cmsHarvester.py

Go to the documentation of this file.

#!/usr/bin/env python

###########################################################################
## File       : cmsHarvest.py
## Authors    : Jeroen Hegeman (jeroen.hegeman@cern.ch)
##              Niklas Pietsch (niklas.pietsch@desy.de)
##              Franseco Costanza (francesco.costanza@desy.de)
## Last change: 20100308


"""Main program to run all kinds of harvesting.

These are the basic kinds of harvesting implemented (contact me if
your favourite is missing):

- RelVal : Run for release validation samples. Makes heavy use of MC
           truth information.

- RelValFS: FastSim RelVal.

- MC : Run for MC samples.

- DQMOffline : Run for real data (could also be run for MC).

For the mappings of these harvesting types to sequence names please
see the setup_harvesting_info() and option_handler_list_types()
methods.

"""
from __future__ import print_function

###########################################################################

__version__ = "3.8.2p1" # (version jump to match release)
__author__ = "Jeroen Hegeman (jeroen.hegeman@cern.ch)," \
             "Niklas Pietsch (niklas.pietsch@desy.de)"

twiki_url = "https://twiki.cern.ch/twiki/bin/view/CMS/CmsHarvester"

###########################################################################

###########################################################################
## TODO list


import os
import sys
import commands
import re
import logging
import optparse
import datetime
import copy
from inspect import getargspec
from random import choice

import six

# These we need to communicate with DBS global DBSAPI
from DBSAPI.dbsApi import DbsApi
import DBSAPI.dbsException
import DBSAPI.dbsApiException
from functools import reduce
# and these we need to parse the DBS output.
global xml
global SAXParseException
import xml.sax
from xml.sax import SAXParseException

import Configuration.PyReleaseValidation
from Configuration.PyReleaseValidation.ConfigBuilder import \
     ConfigBuilder, defaultOptions
# from Configuration.PyReleaseValidation.cmsDriverOptions import options, python_config_filename

#import FWCore.ParameterSet.Config as cms

# Debugging stuff.
import pdb
try:
    import debug_hook
except ImportError:
    pass

###########################################################################
## Helper class: Usage exception.
###########################################################################
class Usage(Exception):
    def __init__(self, msg):
        self.msg = msg
    def __str__(self):
        return repr(self.msg)

    # End of Usage.

###########################################################################
## Helper class: Error exception.
###########################################################################
class Error(Exception):
    def __init__(self, msg):
        self.msg = msg
    def __str__(self):
        return repr(self.msg)

###########################################################################
## Helper class: CMSHarvesterHelpFormatter.
###########################################################################

class CMSHarvesterHelpFormatter(optparse.IndentedHelpFormatter):
    """Helper class to add some customised help output to cmsHarvester.

    We want to add some instructions, as well as a pointer to the CMS
    Twiki.

    """

    def format_usage(self, usage):

        usage_lines = []

        sep_line = "-" * 60
        usage_lines.append(sep_line)
        usage_lines.append("Welcome to the CMS harvester, a (hopefully useful)")
        usage_lines.append("tool to create harvesting configurations.")
        usage_lines.append("For more information please have a look at the CMS Twiki:")
        usage_lines.append("  %s" % twiki_url)
        usage_lines.append(sep_line)
        usage_lines.append("")

        # Since we only add to the output, we now just append the
        # original output from IndentedHelpFormatter.
        usage_lines.append(optparse.IndentedHelpFormatter. \
                           format_usage(self, usage))

        formatted_usage = "\n".join(usage_lines)
        return formatted_usage

    # End of CMSHarvesterHelpFormatter.

###########################################################################
## Helper class: DBSXMLHandler.
###########################################################################

class DBSXMLHandler(xml.sax.handler.ContentHandler):
    """XML handler class to parse DBS results.

    The tricky thing here is that older DBS versions (2.0.5 and
    earlier) return results in a different XML format than newer
    versions. Previously the result values were returned as attributes
    to the `result' element. The new approach returns result values as
    contents of named elements.

    The old approach is handled directly in startElement(), the new
    approach in characters().

    NOTE: All results are returned in the form of string values of
          course!

    """

    # This is the required mapping from the name of the variable we
    # ask for to what we call it ourselves. (Effectively this is the
    # mapping between the old attribute key name and the new element
    # name.)
    mapping = {
        "dataset"        : "PATH",
        "dataset.tag"    : "PROCESSEDDATASET_GLOBALTAG",
        "datatype.type"  : "PRIMARYDSTYPE_TYPE",
        "run"            : "RUNS_RUNNUMBER",
        "run.number"     : "RUNS_RUNNUMBER",
        "file.name"      : "FILES_LOGICALFILENAME",
        "file.numevents" : "FILES_NUMBEROFEVENTS",
        "algo.version"   : "APPVERSION_VERSION",
        "site"           : "STORAGEELEMENT_SENAME",
        }

    def __init__(self, tag_names):
        # This is a list used as stack to keep track of where we are
        # in the element tree.
        self.element_position = []
        self.tag_names = tag_names
        self.results = {}

    def startElement(self, name, attrs):
        self.element_position.append(name)

        self.current_value = []

        #----------

        # This is to catch results from DBS 2.0.5 and earlier.
        if name == "result":
            for name in self.tag_names:
                key = DBSXMLHandler.mapping[name]
                value = str(attrs[key])
                try:
                    self.results[name].append(value)
                except KeyError:
                    self.results[name] = [value]

        #----------

    def endElement(self, name):
        assert self.current_element() == name, \
               "closing unopenend element `%s'" % name

        if self.current_element() in self.tag_names:
            contents = "".join(self.current_value)
            if self.current_element() in self.results:
                self.results[self.current_element()].append(contents)
            else:
                self.results[self.current_element()] = [contents]

        self.element_position.pop()

    def characters(self, content):
        # NOTE: It is possible that the actual contents of the tag
        # gets split into multiple pieces. This method will be called
        # for each of the pieces. This means we have to concatenate
        # everything ourselves.
        if self.current_element() in self.tag_names:
            self.current_value.append(content)

    def current_element(self):
        return self.element_position[-1]

    def check_results_validity(self):
        """Make sure that all results arrays have equal length.

        We should have received complete rows from DBS. I.e. all
        results arrays in the handler should be of equal length.

        """

        results_valid = True

        res_names = self.results.keys()
        if len(res_names) > 1:
            for res_name in res_names[1:]:
                res_tmp = self.results[res_name]
                if len(res_tmp) != len(self.results[res_names[0]]):
                    results_valid = False

        return results_valid

    # End of DBSXMLHandler.

###########################################################################
## CMSHarvester class.
###########################################################################

class CMSHarvester(object):
    """Class to perform CMS harvesting.

    More documentation `obviously' to follow.

    """

    ##########

    def __init__(self, cmd_line_opts=None):
        "Initialize class and process command line options."

        self.version = __version__

        # These are the harvesting types allowed. See the head of this
        # file for more information.
        self.harvesting_types = [
            "RelVal",
            "RelValFS",
            "MC",
            "DQMOffline",
            ]

        # These are the possible harvesting modes:
        #   - Single-step: harvesting takes place on-site in a single
        #   step. For each samples only a single ROOT file containing
        #   the harvesting results is returned.
        #   - Single-step-allow-partial: special hack to allow
        #   harvesting of partial statistics using single-step
        #   harvesting on spread-out samples.
        #   - Two-step: harvesting takes place in two steps. The first
        #   step returns a series of monitoring elenent summaries for
        #   each sample. The second step then merges these summaries
        #   locally and does the real harvesting. This second step
        #   produces the ROOT file containing the harvesting results.
        self.harvesting_modes = [
            "single-step",
            "single-step-allow-partial",
            "two-step"
            ]

        # It is possible to specify a GlobalTag that will override any
        # choices (regarding GlobalTags) made by the cmsHarvester.
        # BUG BUG BUG
        # For the moment, until I figure out a way to obtain the
        # GlobalTag with which a given data (!) dataset was created,
        # it is necessary to specify a GlobalTag for harvesting of
        # data.
        # BUG BUG BUG end
        self.globaltag = None

        # It's also possible to switch off the use of reference
        # histograms altogether.
        self.use_ref_hists = True

        # The database name and account are hard-coded. They are not
        # likely to change before the end-of-life of this tool. But of
        # course there is a way to override this from the command
        # line. One can even override the Frontier connection used for
        # the GlobalTag and for the reference histograms
        # independently. Please only use this for testing purposes.
        self.frontier_connection_name = {}
        self.frontier_connection_name["globaltag"] = "frontier://" \
                                                     "FrontierProd/"
        self.frontier_connection_name["refhists"] = "frontier://" \
                                                    "FrontierProd/"
        self.frontier_connection_overridden = {}
        for key in self.frontier_connection_name.keys():
            self.frontier_connection_overridden[key] = False

        # This contains information specific to each of the harvesting
        # types. Used to create the harvesting configuration. It is
        # filled by setup_harvesting_info().
        self.harvesting_info = None

        ###

        # These are default `unused' values that will be filled in
        # depending on the command line options.

        # The type of harvesting we're doing. See
        # self.harvesting_types for allowed types.
        self.harvesting_type = None

        # The harvesting mode, popularly known as single-step
        # vs. two-step. The thing to remember at this point is that
        # single-step is only possible for samples located completely
        # at a single site (i.e. SE).
        self.harvesting_mode = None
        # BUG BUG BUG
        # Default temporarily set to two-step until we can get staged
        # jobs working with CRAB.
        self.harvesting_mode_default = "single-step"
        # BUG BUG BUG end

        # The input method: are we reading a dataset name (or regexp)
        # directly from the command line or are we reading a file
        # containing a list of dataset specifications. Actually we
        # keep one of each for both datasets and runs.
        self.input_method = {}
        self.input_method["datasets"] = {}
        self.input_method["datasets"]["use"] = None
        self.input_method["datasets"]["ignore"] = None
        self.input_method["runs"] = {}
        self.input_method["runs"]["use"] = None
        self.input_method["runs"]["ignore"] = None
        self.input_method["runs"]["ignore"] = None
        # The name of whatever input we're using.
        self.input_name = {}
        self.input_name["datasets"] = {}
        self.input_name["datasets"]["use"] = None
        self.input_name["datasets"]["ignore"] = None
        self.input_name["runs"] = {}
        self.input_name["runs"]["use"] = None
        self.input_name["runs"]["ignore"] = None

        self.Jsonlumi = False
        self.Jsonfilename = "YourJSON.txt"
        self.Jsonrunfilename = "YourJSON.txt"
        self.todofile = "YourToDofile.txt"

        # If this is true, we're running in `force mode'. In this case
        # the sanity checks are performed but failure will not halt
        # everything.
        self.force_running = None

        # The base path of the output dir in CASTOR.
        self.castor_base_dir = None
        self.castor_base_dir_default = "/castor/cern.ch/" \
                                       "cms/store/temp/" \
                                       "dqm/offline/harvesting_output/"

        # The name of the file to be used for book keeping: which
        # datasets, runs, etc. we have already processed.
        self.book_keeping_file_name = None
        self.book_keeping_file_name_default = "harvesting_accounting.txt"

        # The dataset name to reference histogram name mapping is read
        # from a text file. The name of this file is kept in the
        # following variable.
        self.ref_hist_mappings_file_name = None
        # And this is the default value.
        self.ref_hist_mappings_file_name_default = "harvesting_ref_hist_mappings.txt"

        # Hmmm, hard-coded prefix of the CERN CASTOR area. This is the
        # only supported CASTOR area.
        # NOTE: Make sure this one starts with a `/'.
        self.castor_prefix = "/castor/cern.ch"

        # Normally the central harvesting should be done using the
        # `t1access' grid role. To be able to run without T1 access
        # the --no-t1access flag can be used. This variable keeps
        # track of that special mode.
        self.non_t1access = False
        self.caf_access = False
        self.saveByLumiSection = False
        self.crab_submission = False
        self.nr_max_sites = 1

        self.preferred_site = "no preference"

        # This will become the list of datasets and runs to consider
        self.datasets_to_use = {}
        # and this will become the list of datasets and runs to skip.
        self.datasets_to_ignore = {}
        # This, in turn, will hold all book keeping information.
        self.book_keeping_information = {}
        # And this is where the dataset name to reference histogram
        # name mapping is stored.
        self.ref_hist_mappings = {}

        # We're now also allowing run selection. This means we also
        # have to keep list of runs requested and vetoed by the user.
        self.runs_to_use = {}
        self.runs_to_ignore = {}

        # Cache for CMSSW version availability at different sites.
        self.sites_and_versions_cache = {}

        # Cache for checked GlobalTags.
        self.globaltag_check_cache = []

        # Global flag to see if there were any jobs for which we could
        # not find a matching site.
        self.all_sites_found = True

        # Helper string centrally defined.
        self.no_matching_site_found_str = "no_matching_site_found"

        # Store command line options for later use.
        if cmd_line_opts is None:
            cmd_line_opts = sys.argv[1:]
        self.cmd_line_opts = cmd_line_opts

        # Set up the logger.
        log_handler = logging.StreamHandler()
        # This is the default log formatter, the debug option switches
        # on some more information.
        log_formatter = logging.Formatter("%(message)s")
        log_handler.setFormatter(log_formatter)
        logger = logging.getLogger()
        logger.name = "main"
        logger.addHandler(log_handler)
        self.logger = logger
        # The default output mode is quite verbose.
        self.set_output_level("NORMAL")

        #logger.debug("Initialized successfully")

        # End of __init__.

    ##########

    def cleanup(self):
        "Clean up after ourselves."

        # NOTE: This is the safe replacement of __del__.

        #self.logger.debug("All done -> shutting down")
        logging.shutdown()

        # End of cleanup.

    ##########

    def time_stamp(self):
        "Create a timestamp to use in the created config files."

        time_now = datetime.datetime.utcnow()
        # We don't care about the microseconds.
        time_now = time_now.replace(microsecond = 0)
        time_stamp = "%sUTC" % datetime.datetime.isoformat(time_now)

        # End of time_stamp.
        return time_stamp

    ##########

    def ident_string(self):
        "Spit out an identification string for cmsHarvester.py."

        ident_str = "`cmsHarvester.py " \
                    "version %s': cmsHarvester.py %s" % \
                    (__version__,
                     reduce(lambda x, y: x+' '+y, sys.argv[1:]))

        return ident_str

    ##########

    def format_conditions_string(self, globaltag):
        """Create the conditions string needed for `cmsDriver'.

        Just glueing the FrontierConditions bit in front of it really.

        """

        # Not very robust but okay. The idea is that if the user
        # specified (since this cannot happen with GlobalTags coming
        # from DBS) something containing `conditions', they probably
        # know what they're doing and we should not muck things up. In
        # all other cases we just assume we only received the
        # GlobalTag part and we built the usual conditions string from
        # that .
        if globaltag.lower().find("conditions") > -1:
            conditions_string = globaltag
        else:
            conditions_string = "FrontierConditions_GlobalTag,%s" % \
                                globaltag

        # End of format_conditions_string.
        return conditions_string

    ##########

    def db_account_name_cms_cond_globaltag(self):
        """Return the database account name used to store the GlobalTag.

        The name of the database account depends (albeit weakly) on
        the CMSSW release version.

        """

        # This never changed, unlike the cms_cond_31X_DQM_SUMMARY ->
        # cms_cond_34X_DQM transition.
        account_name = "CMS_COND_31X_GLOBALTAG"

        # End of db_account_name_cms_cond_globaltag.
        return account_name

    ##########

    def db_account_name_cms_cond_dqm_summary(self):
        """See db_account_name_cms_cond_globaltag."""

        account_name = None
        version = self.cmssw_version[6:11]
        if version < "3_4_0":
            account_name = "CMS_COND_31X_DQM_SUMMARY"
        else:
            account_name = "CMS_COND_34X"

        # End of db_account_name_cms_cond_dqm_summary.
        return account_name

    ##########

    def config_file_header(self):
        "Create a nice header to be used to mark the generated files."

        tmp = []

        time_stamp = self.time_stamp()
        ident_str = self.ident_string()
        tmp.append("# %s" % time_stamp)
        tmp.append("# WARNING: This file was created automatically!")
        tmp.append("")
        tmp.append("# Created by %s" % ident_str)

        header = "\n".join(tmp)

        # End of config_file_header.
        return header

    ##########

    def set_output_level(self, output_level):
        """Adjust the level of output generated.

        Choices are:
          - normal  : default level of output
          - quiet   : less output than the default
          - verbose : some additional information
          - debug   : lots more information, may be overwhelming

        NOTE: The debug option is a bit special in the sense that it
              also modifies the output format.

        """

        # NOTE: These levels are hooked up to the ones used in the
        #       logging module.
        output_levels = {
            "NORMAL"  : logging.INFO,
            "QUIET"   : logging.WARNING,
            "VERBOSE" : logging.INFO,
            "DEBUG"   : logging.DEBUG
            }

        output_level = output_level.upper()

        try:
            # Update the logger.
            self.log_level = output_levels[output_level]
            self.logger.setLevel(self.log_level)
        except KeyError:
            # Show a complaint
            self.logger.fatal("Unknown output level `%s'" % ouput_level)
            # and re-raise an exception.
            raise Exception

        # End of set_output_level.

    ##########

    def option_handler_debug(self, option, opt_str, value, parser):
        """Switch to debug mode.

        This both increases the amount of output generated, as well as
        changes the format used (more detailed information is given).

        """

        # Switch to a more informative log formatter for debugging.
        log_formatter_debug = logging.Formatter("[%(levelname)s] " \
                                                # NOTE: funcName was
                                                # only implemented
                                                # starting with python
                                                # 2.5.
                                                #"%(funcName)s() " \
                                                #"@%(filename)s:%(lineno)d " \
                                                "%(message)s")
        # Hmmm, not very nice. This assumes there's only a single
        # handler associated with the current logger.
        log_handler = self.logger.handlers[0]
        log_handler.setFormatter(log_formatter_debug)
        self.set_output_level("DEBUG")

        # End of option_handler_debug.

    ##########

    def option_handler_quiet(self, option, opt_str, value, parser):
        "Switch to quiet mode: less verbose."

        self.set_output_level("QUIET")

        # End of option_handler_quiet.

    ##########

    def option_handler_force(self, option, opt_str, value, parser):
        """Switch on `force mode' in which case we don't brake for nobody.

        In so-called `force mode' all sanity checks are performed but
        we don't halt on failure. Of course this requires some care
        from the user.

        """

        self.logger.debug("Switching on `force mode'.")
        self.force_running = True

        # End of option_handler_force.

    ##########

    def option_handler_harvesting_type(self, option, opt_str, value, parser):
        """Set the harvesting type to be used.

        This checks that no harvesting type is already set, and sets
        the harvesting type to be used to the one specified. If a
        harvesting type is already set an exception is thrown. The
        same happens when an unknown type is specified.

        """

        # Check for (in)valid harvesting types.
        # NOTE: The matching is done in a bit of a complicated
        # way. This allows the specification of the type to be
        # case-insensitive while still ending up with the properly
        # `cased' version afterwards.
        value = value.lower()
        harvesting_types_lowered = [i.lower() for i in self.harvesting_types]
        try:
            type_index = harvesting_types_lowered.index(value)
            # If this works, we now have the index to the `properly
            # cased' version of the harvesting type.
        except ValueError:
            self.logger.fatal("Unknown harvesting type `%s'" % \
                              value)
            self.logger.fatal("  possible types are: %s" %
                              ", ".join(self.harvesting_types))
            raise Usage("Unknown harvesting type `%s'" % \
                        value)

        # Check if multiple (by definition conflicting) harvesting
        # types are being specified.
        if not self.harvesting_type is None:
            msg = "Only one harvesting type should be specified"
            self.logger.fatal(msg)
            raise Usage(msg)
        self.harvesting_type = self.harvesting_types[type_index]

        self.logger.info("Harvesting type to be used: `%s'" % \
                         self.harvesting_type)

        # End of option_handler_harvesting_type.

    ##########

    def option_handler_harvesting_mode(self, option, opt_str, value, parser):
        """Set the harvesting mode to be used.

        Single-step harvesting can be used for samples that are
        located completely at a single site (= SE). Otherwise use
        two-step mode.

        """

        # Check for valid mode.
        harvesting_mode = value.lower()
        if not harvesting_mode in self.harvesting_modes:
            msg = "Unknown harvesting mode `%s'" % harvesting_mode
            self.logger.fatal(msg)
            self.logger.fatal("  possible modes are: %s" % \
                              ", ".join(self.harvesting_modes))
            raise Usage(msg)

        # Check if we've been given only a single mode, otherwise
        # complain.
        if not self.harvesting_mode is None:
            msg = "Only one harvesting mode should be specified"
            self.logger.fatal(msg)
            raise Usage(msg)
        self.harvesting_mode = harvesting_mode

        self.logger.info("Harvesting mode to be used: `%s'" % \
                         self.harvesting_mode)

        # End of option_handler_harvesting_mode.

    ##########

    def option_handler_globaltag(self, option, opt_str, value, parser):
        """Set the GlobalTag to be used, overriding our own choices.

        By default the cmsHarvester will use the GlobalTag with which
        a given dataset was created also for the harvesting. The
        --globaltag option is the way to override this behaviour.

        """

        # Make sure that this flag only occurred once.
        if not self.globaltag is None:
            msg = "Only one GlobalTag should be specified"
            self.logger.fatal(msg)
            raise Usage(msg)
        self.globaltag = value

        self.logger.info("GlobalTag to be used: `%s'" % \
                         self.globaltag)

        # End of option_handler_globaltag.

    ##########

    def option_handler_no_ref_hists(self, option, opt_str, value, parser):
        "Switch use of all reference histograms off."

        self.use_ref_hists = False

        self.logger.warning("Switching off all use of reference histograms")

        # End of option_handler_no_ref_hists.

     ##########

    def option_handler_frontier_connection(self, option, opt_str,
                                           value, parser):
        """Override the default Frontier connection string.

        Please only use this for testing (e.g. when a test payload has
        been inserted into cms_orc_off instead of cms_orc_on).

        This method gets called for three different command line
        options:
        - --frontier-connection,
        - --frontier-connection-for-globaltag,
        - --frontier-connection-for-refhists.
        Appropriate care has to be taken to make sure things are only
        specified once.

        """

        # Figure out with which command line option we've been called.
        frontier_type = opt_str.split("-")[-1]
        if frontier_type == "connection":
            # Main option: change all connection strings.
            frontier_types = self.frontier_connection_name.keys()
        else:
            frontier_types = [frontier_type]

        # Make sure that each Frontier connection is specified only
        # once. (Okay, in a bit of a dodgy way...)
        for connection_name in frontier_types:
            if self.frontier_connection_overridden[connection_name] == True:
                msg = "Please specify either:\n" \
                      "  `--frontier-connection' to change the " \
                      "Frontier connection used for everything, or\n" \
                      "either one or both of\n" \
                      "  `--frontier-connection-for-globaltag' to " \
                      "change the Frontier connection used for the " \
                      "GlobalTag and\n" \
                      "  `--frontier-connection-for-refhists' to change " \
                      "the Frontier connection used for the " \
                      "reference histograms."
                self.logger.fatal(msg)
                raise Usage(msg)

        frontier_prefix = "frontier://"
        if not value.startswith(frontier_prefix):
            msg = "Expecting Frontier connections to start with " \
                  "`%s'. You specified `%s'." % \
                  (frontier_prefix, value)
            self.logger.fatal(msg)
            raise Usage(msg)
        # We also kind of expect this to be either FrontierPrep or
        # FrontierProd (but this is just a warning).
        if value.find("FrontierProd") < 0 and \
               value.find("FrontierProd") < 0:
            msg = "Expecting Frontier connections to contain either " \
                  "`FrontierProd' or `FrontierPrep'. You specified " \
                  "`%s'. Are you sure?" % \
                  value
            self.logger.warning(msg)

        if not value.endswith("/"):
            value += "/"

        for connection_name in frontier_types:
            self.frontier_connection_name[connection_name] = value
            self.frontier_connection_overridden[connection_name] = True

            frontier_type_str = "unknown"
            if connection_name == "globaltag":
                frontier_type_str = "the GlobalTag"
            elif connection_name == "refhists":
                frontier_type_str = "the reference histograms"

            self.logger.warning("Overriding default Frontier " \
                                "connection for %s " \
                                "with `%s'" % \
                                (frontier_type_str,
                                 self.frontier_connection_name[connection_name]))

        # End of option_handler_frontier_connection

    ##########

    def option_handler_input_todofile(self, option, opt_str, value, parser):

        self.todofile = value
        # End of option_handler_input_todofile.

    ##########

    def option_handler_input_Jsonfile(self, option, opt_str, value, parser):

        self.Jsonfilename = value
        # End of option_handler_input_Jsonfile.

    ##########

    def option_handler_input_Jsonrunfile(self, option, opt_str, value, parser):

        self.Jsonrunfilename = value
        # End of option_handler_input_Jsonrunfile.

    ##########

    def option_handler_input_spec(self, option, opt_str, value, parser):
        """TODO TODO TODO
        Document this...

        """

        # Figure out if we were called for the `use these' or the
        # `ignore these' case.
        if opt_str.lower().find("ignore") > -1:
            spec_type = "ignore"
        else:
            spec_type = "use"

        # Similar: are we being called for datasets or for runs?
        if opt_str.lower().find("dataset") > -1:
            select_type = "datasets"
        else:
            select_type = "runs"

        if not self.input_method[select_type][spec_type] is None:
            msg = "Please only specify one input method " \
                  "(for the `%s' case)" % opt_str
            self.logger.fatal(msg)
            raise Usage(msg)

        input_method = opt_str.replace("-", "").replace("ignore", "")
        self.input_method[select_type][spec_type] = input_method
        self.input_name[select_type][spec_type] = value

        self.logger.debug("Input method for the `%s' case: %s" % \
                          (spec_type, input_method))

        # End of option_handler_input_spec

    ##########

    def option_handler_book_keeping_file(self, option, opt_str, value, parser):
        """Store the name of the file to be used for book keeping.

        The only check done here is that only a single book keeping
        file is specified.

        """

        file_name = value

        if not self.book_keeping_file_name is None:
            msg = "Only one book keeping file should be specified"
            self.logger.fatal(msg)
            raise Usage(msg)
        self.book_keeping_file_name = file_name

        self.logger.info("Book keeping file to be used: `%s'" % \
                         self.book_keeping_file_name)

        # End of option_handler_book_keeping_file.

    ##########

    def option_handler_ref_hist_mapping_file(self, option, opt_str, value, parser):
        """Store the name of the file for the ref. histogram mapping.

        """

        file_name = value

        if not self.ref_hist_mappings_file_name is None:
            msg = "Only one reference histogram mapping file " \
                  "should be specified"
            self.logger.fatal(msg)
            raise Usage(msg)
        self.ref_hist_mappings_file_name = file_name

        self.logger.info("Reference histogram mapping file " \
                         "to be used: `%s'" % \
                         self.ref_hist_mappings_file_name)

        # End of option_handler_ref_hist_mapping_file.

    ##########

    # OBSOLETE OBSOLETE OBSOLETE

##    def option_handler_dataset_name(self, option, opt_str, value, parser):
##        """Specify the name(s) of the dataset(s) to be processed.

##        It is checked to make sure that no dataset name or listfile
##        names are given yet. If all is well (i.e. we still have a
##        clean slate) the dataset name is stored for later use,
##        otherwise a Usage exception is raised.

##        """

##        if not self.input_method is None:
##            if self.input_method == "dataset":
##                raise Usage("Please only feed me one dataset specification")
##            elif self.input_method == "listfile":
##                raise Usage("Cannot specify both dataset and input list file")
##            else:
##                assert False, "Unknown input method `%s'" % self.input_method
##        self.input_method = "dataset"
##        self.input_name = value
##        self.logger.info("Input method used: %s" % self.input_method)

##        # End of option_handler_dataset_name.

##    ##########

##    def option_handler_listfile_name(self, option, opt_str, value, parser):
##        """Specify the input list file containing datasets to be processed.

##        It is checked to make sure that no dataset name or listfile
##        names are given yet. If all is well (i.e. we still have a
##        clean slate) the listfile name is stored for later use,
##        otherwise a Usage exception is raised.

##        """

##        if not self.input_method is None:
##            if self.input_method == "listfile":
##                raise Usage("Please only feed me one list file")
##            elif self.input_method == "dataset":
##                raise Usage("Cannot specify both dataset and input list file")
##            else:
##                assert False, "Unknown input method `%s'" % self.input_method
##        self.input_method = "listfile"
##        self.input_name = value
##        self.logger.info("Input method used: %s" % self.input_method)

##        # End of option_handler_listfile_name.

    # OBSOLETE OBSOLETE OBSOLETE end

    ##########

    def option_handler_castor_dir(self, option, opt_str, value, parser):
        """Specify where on CASTOR the output should go.

        At the moment only output to CERN CASTOR is
        supported. Eventually the harvested results should go into the
        central place for DQM on CASTOR anyway.

        """

        # Check format of specified CASTOR area.
        castor_dir = value
        #castor_dir = castor_dir.lstrip(os.path.sep)
        castor_prefix = self.castor_prefix

        # Add a leading slash if necessary and clean up the path.
        castor_dir = os.path.join(os.path.sep, castor_dir)
        self.castor_base_dir = os.path.normpath(castor_dir)

        self.logger.info("CASTOR (base) area to be used: `%s'" % \
                         self.castor_base_dir)

        # End of option_handler_castor_dir.

    ##########

    def option_handler_no_t1access(self, option, opt_str, value, parser):
        """Set the self.no_t1access flag to try and create jobs that
        run without special `t1access' role.

        """

        self.non_t1access = True

        self.logger.warning("Running in `non-t1access' mode. " \
                            "Will try to create jobs that run " \
                            "without special rights but no " \
                            "further promises...")

        # End of option_handler_no_t1access.

    ##########

    def option_handler_caf_access(self, option, opt_str, value, parser):
        """Set the self.caf_access flag to try and create jobs that
        run on the CAF.

        """
        self.caf_access = True

        self.logger.warning("Running in `caf_access' mode. " \
                            "Will try to create jobs that run " \
                            "on CAF but no" \
                            "further promises...")

        # End of option_handler_caf_access.

   ##########

    def option_handler_saveByLumiSection(self, option, opt_str, value, parser):
        """Set process.dqmSaver.saveByLumiSectiont=1 in cfg harvesting file
        """
        self.saveByLumiSection = True

        self.logger.warning("waning concerning saveByLumiSection option")

        # End of option_handler_saveByLumiSection.


    ##########

    def option_handler_crab_submission(self, option, opt_str, value, parser):
        """Crab jobs are not created and
        "submitted automatically",
        """
        self.crab_submission = True

        # End of option_handler_crab_submission.

    ##########

    def option_handler_sites(self, option, opt_str, value, parser):

        self.nr_max_sites = value

    ##########

    def option_handler_preferred_site(self, option, opt_str, value, parser):

        self.preferred_site = value

    ##########

    def option_handler_list_types(self, option, opt_str, value, parser):
        """List all harvesting types and their mappings.

        This lists all implemented harvesting types with their
        corresponding mappings to sequence names. This had to be
        separated out from the help since it depends on the CMSSW
        version and was making things a bit of a mess.

        NOTE: There is no way (at least not that I could come up with)
        to code this in a neat generic way that can be read both by
        this method and by setup_harvesting_info(). Please try hard to
        keep these two methods in sync!

        """

        sep_line = "-" * 50
        sep_line_short = "-" * 20

        print(sep_line)
        print("The following harvesting types are available:")
        print(sep_line)

        print("`RelVal' maps to:")
        print("  pre-3_3_0           : HARVESTING:validationHarvesting")
        print("  3_4_0_pre2 and later: HARVESTING:validationHarvesting+dqmHarvesting")
        print("  Exceptions:")
        print("    3_3_0_pre1-4        : HARVESTING:validationHarvesting")
        print("    3_3_0_pre6          : HARVESTING:validationHarvesting")
        print("    3_4_0_pre1          : HARVESTING:validationHarvesting")

        print(sep_line_short)

        print("`RelValFS' maps to:")
        print("  always              : HARVESTING:validationHarvestingFS")

        print(sep_line_short)

        print("`MC' maps to:")
        print("    always          : HARVESTING:validationprodHarvesting")

        print(sep_line_short)

        print("`DQMOffline' maps to:")
        print("  always              : HARVESTING:dqmHarvesting")

        print(sep_line)

        # We're done, let's quit. (This is the same thing optparse
        # does after printing the help.)
        raise SystemExit

        # End of option_handler_list_types.

    ##########

    def setup_harvesting_info(self):
        """Fill our dictionary with all info needed to understand
        harvesting.

        This depends on the CMSSW version since at some point the
        names and sequences were modified.

        NOTE: There is no way (at least not that I could come up with)
        to code this in a neat generic way that can be read both by
        this method and by option_handler_list_types(). Please try
        hard to keep these two methods in sync!

        """

        assert not self.cmssw_version is None, \
               "ERROR setup_harvesting() requires " \
               "self.cmssw_version to be set!!!"

        harvesting_info = {}

        # This is the version-independent part.
        harvesting_info["DQMOffline"] = {}
        harvesting_info["DQMOffline"]["beamspot"] = None
        harvesting_info["DQMOffline"]["eventcontent"] = None
        harvesting_info["DQMOffline"]["harvesting"] = "AtRunEnd"

        harvesting_info["RelVal"] = {}
        harvesting_info["RelVal"]["beamspot"] = None
        harvesting_info["RelVal"]["eventcontent"] = None
        harvesting_info["RelVal"]["harvesting"] = "AtRunEnd"

        harvesting_info["RelValFS"] = {}
        harvesting_info["RelValFS"]["beamspot"] = None
        harvesting_info["RelValFS"]["eventcontent"] = None
        harvesting_info["RelValFS"]["harvesting"] = "AtRunEnd"

        harvesting_info["MC"] = {}
        harvesting_info["MC"]["beamspot"] = None
        harvesting_info["MC"]["eventcontent"] = None
        harvesting_info["MC"]["harvesting"] = "AtRunEnd"

        # This is the version-dependent part. And I know, strictly
        # speaking it's not necessary to fill in all three types since
        # in a single run we'll only use one type anyway. This does
        # look more readable, however, and required less thought from
        # my side when I put this together.

        # DEBUG DEBUG DEBUG
        # Check that we understand our own version naming.
        assert self.cmssw_version.startswith("CMSSW_")
        # DEBUG DEBUG DEBUG end

        version = self.cmssw_version[6:]

        #----------

        # RelVal
        step_string = None
        if version < "3_3_0":
            step_string = "validationHarvesting"
        elif version in ["3_3_0_pre1", "3_3_0_pre2",
                         "3_3_0_pre3", "3_3_0_pre4",
                         "3_3_0_pre6", "3_4_0_pre1"]:
            step_string = "validationHarvesting"
        else:
            step_string = "validationHarvesting+dqmHarvesting"

        harvesting_info["RelVal"]["step_string"] = step_string

        # DEBUG DEBUG DEBUG
        # Let's make sure we found something.
        assert not step_string is None, \
               "ERROR Could not decide a RelVal harvesting sequence " \
               "for CMSSW version %s" % self.cmssw_version
        # DEBUG DEBUG DEBUG end

        #----------

        # RelVal
        step_string = "validationHarvestingFS"

        harvesting_info["RelValFS"]["step_string"] = step_string

        #----------

        # MC
        step_string = "validationprodHarvesting"

        harvesting_info["MC"]["step_string"] = step_string

        # DEBUG DEBUG DEBUG
        # Let's make sure we found something.
        assert not step_string is None, \
               "ERROR Could not decide a MC harvesting " \
               "sequence for CMSSW version %s" % self.cmssw_version
        # DEBUG DEBUG DEBUG end

        #----------

        # DQMOffline
        step_string = "dqmHarvesting"

        harvesting_info["DQMOffline"]["step_string"] = step_string

        #----------

        self.harvesting_info = harvesting_info

        self.logger.info("Based on the CMSSW version (%s) " \
                         "I decided to use the `HARVESTING:%s' " \
                         "sequence for %s harvesting" % \
                         (self.cmssw_version,
                          self.harvesting_info[self.harvesting_type]["step_string"],
                          self.harvesting_type))

        # End of setup_harvesting_info.

    ##########

    def create_castor_path_name_common(self, dataset_name):
        """Build the common part of the output path to be used on
        CASTOR.

        This consists of the CASTOR area base path specified by the
        user and a piece depending on the data type (data vs. MC), the
        harvesting type and the dataset name followed by a piece
        containing the run number and event count. (See comments in
        create_castor_path_name_special for details.) This method
        creates the common part, without run number and event count.

        """

        castor_path = self.castor_base_dir

        ###

        # The data type: data vs. mc.
        datatype = self.datasets_information[dataset_name]["datatype"]
        datatype = datatype.lower()
        castor_path = os.path.join(castor_path, datatype)

        # The harvesting type.
        harvesting_type = self.harvesting_type
        harvesting_type = harvesting_type.lower()
        castor_path = os.path.join(castor_path, harvesting_type)

        # The CMSSW release version (only the `digits'). Note that the
        # CMSSW version used here is the version used for harvesting,
        # not the one from the dataset. This does make the results
        # slightly harder to find. On the other hand it solves
        # problems in case one re-harvests a given dataset with a
        # different CMSSW version, which would lead to ambiguous path
        # names. (Of course for many cases the harvesting is done with
        # the same CMSSW version the dataset was created with.)
        release_version = self.cmssw_version
        release_version = release_version.lower(). \
                          replace("cmssw", ""). \
                          strip("_")
        castor_path = os.path.join(castor_path, release_version)

        # The dataset name.
        dataset_name_escaped = self.escape_dataset_name(dataset_name)
        castor_path = os.path.join(castor_path, dataset_name_escaped)

        ###

        castor_path = os.path.normpath(castor_path)

        # End of create_castor_path_name_common.
        return castor_path

    ##########

    def create_castor_path_name_special(self,
                                        dataset_name, run_number,
                                        castor_path_common):
        """Create the specialised part of the CASTOR output dir name.

        NOTE: To avoid clashes with `incremental harvesting'
        (re-harvesting when a dataset grows) we have to include the
        event count in the path name. The underlying `problem' is that
        CRAB does not overwrite existing output files so if the output
        file already exists CRAB will fail to copy back the output.

        NOTE: It's not possible to create different kinds of
        harvesting jobs in a single call to this tool. However, in
        principle it could be possible to create both data and MC jobs
        in a single go.

        NOTE: The number of events used in the path name is the
        _total_ number of events in the dataset/run at the time of
        harvesting. If we're doing partial harvesting the final
        results will reflect lower statistics. This is a) the easiest
        to code and b) the least likely to lead to confusion if
        someone ever decides to swap/copy around file blocks between
        sites.

        """

        castor_path = castor_path_common

        ###

        # The run number part.
        castor_path = os.path.join(castor_path, "run_%d" % run_number)

        ###

        # The event count (i.e. the number of events we currently see
        # for this dataset).
        #nevents = self.datasets_information[dataset_name] \
        #          ["num_events"][run_number]
        castor_path = os.path.join(castor_path, "nevents")

        ###

        castor_path = os.path.normpath(castor_path)

        # End of create_castor_path_name_special.
        return castor_path

    ##########

    def create_and_check_castor_dirs(self):
        """Make sure all required CASTOR output dirs exist.

        This checks the CASTOR base dir specified by the user as well
        as all the subdirs required by the current set of jobs.

        """

        self.logger.info("Checking (and if necessary creating) CASTOR " \
                         "output area(s)...")

        # Call the real checker method for the base dir.
        self.create_and_check_castor_dir(self.castor_base_dir)

        # Now call the checker for all (unique) subdirs.
        castor_dirs = []
        for (dataset_name, runs) in six.iteritems(self.datasets_to_use):

            for run in runs:
                castor_dirs.append(self.datasets_information[dataset_name] \
                                   ["castor_path"][run])
        castor_dirs_unique = sorted(set(castor_dirs))
        # This can take some time. E.g. CRAFT08 has > 300 runs, each
        # of which will get a new directory. So we show some (rough)
        # info in between.
        ndirs = len(castor_dirs_unique)
        step = max(ndirs / 10, 1)
        for (i, castor_dir) in enumerate(castor_dirs_unique):
            if (i + 1) % step == 0 or \
                   (i + 1) == ndirs:
                self.logger.info("  %d/%d" % \
                                 (i + 1, ndirs))
            self.create_and_check_castor_dir(castor_dir)

            # Now check if the directory is empty. If (an old version
            # of) the output file already exists CRAB will run new
            # jobs but never copy the results back. We assume the user
            # knows what they are doing and only issue a warning in
            # case the directory is not empty.
            self.logger.debug("Checking if path `%s' is empty" % \
                              castor_dir)
            cmd = "rfdir %s" % castor_dir
            (status, output) = commands.getstatusoutput(cmd)
            if status != 0:
                msg = "Could not access directory `%s'" \
                      " !!! This is bad since I should have just" \
                      " created it !!!" % castor_dir
                self.logger.fatal(msg)
                raise Error(msg)
            if len(output) > 0:
                self.logger.warning("Output directory `%s' is not empty:" \
                                    " new jobs will fail to" \
                                    " copy back output" % \
                                    castor_dir)

        # End of create_and_check_castor_dirs.

    ##########

    def create_and_check_castor_dir(self, castor_dir):
        """Check existence of the give CASTOR dir, if necessary create
        it.

        Some special care has to be taken with several things like
        setting the correct permissions such that CRAB can store the
        output results. Of course this means that things like
        /castor/cern.ch/ and user/j/ have to be recognised and treated
        properly.

        NOTE: Only CERN CASTOR area (/castor/cern.ch/) supported for
        the moment.

        NOTE: This method uses some slightly tricky caching to make
        sure we don't keep over and over checking the same base paths.

        """

        ###

        # Local helper function to fully split a path into pieces.
        def split_completely(path):
            (parent_path, name) = os.path.split(path)
            if name == "":
                return (parent_path, )
            else:
                return split_completely(parent_path) + (name, )

        ###

        # Local helper function to check rfio (i.e. CASTOR)
        # directories.
        def extract_permissions(rfstat_output):
            """Parse the output from rfstat and return the
            5-digit permissions string."""

            permissions_line = [i for i in output.split("\n") \
                                if i.lower().find("protection") > -1]
            regexp = re.compile(".*\(([0123456789]{5})\).*")
            match = regexp.search(rfstat_output)
            if not match or len(match.groups()) != 1:
                msg = "Could not extract permissions " \
                      "from output: %s" % rfstat_output
                self.logger.fatal(msg)
                raise Error(msg)
            permissions = match.group(1)

            # End of extract_permissions.
            return permissions

        ###

        # These are the pieces of CASTOR directories that we do not
        # want to touch when modifying permissions.

        # NOTE: This is all a bit involved, basically driven by the
        # fact that one wants to treat the `j' directory of
        # `/castor/cern.ch/user/j/jhegeman/' specially.
        # BUG BUG BUG
        # This should be simplified, for example by comparing to the
        # CASTOR prefix or something like that.
        # BUG BUG BUG end
        castor_paths_dont_touch = {
            0: ["/", "castor", "cern.ch", "cms", "store", "temp",
                "dqm", "offline", "user"],
            -1: ["user", "store"]
            }

        self.logger.debug("Checking CASTOR path `%s'" % castor_dir)

        ###

        # First we take the full CASTOR path apart.
        castor_path_pieces = split_completely(castor_dir)

        # Now slowly rebuild the CASTOR path and see if a) all
        # permissions are set correctly and b) the final destination
        # exists.
        path = ""
        check_sizes = sorted(castor_paths_dont_touch.keys())
        len_castor_path_pieces = len(castor_path_pieces)
        for piece_index in xrange (len_castor_path_pieces):
            skip_this_path_piece = False
            piece = castor_path_pieces[piece_index]
##            self.logger.debug("Checking CASTOR path piece `%s'" % \
##                              piece)
            for check_size in check_sizes:
                # Do we need to do anything with this?
                if (piece_index + check_size) > -1:
##                    self.logger.debug("Checking `%s' against `%s'" % \
##                                      (castor_path_pieces[piece_index + check_size],
##                                       castor_paths_dont_touch[check_size]))
                    if castor_path_pieces[piece_index + check_size] in castor_paths_dont_touch[check_size]:
##                        self.logger.debug("  skipping")
                        skip_this_path_piece = True
##                    else:
##                        # Piece not in the list, fine.
##                        self.logger.debug("  accepting")
            # Add piece to the path we're building.
##            self.logger.debug("!!! Skip path piece `%s'? %s" % \
##                              (piece, str(skip_this_path_piece)))
##            self.logger.debug("Adding piece to path...")
            path = os.path.join(path, piece)
##            self.logger.debug("Path is now `%s'" % \
##                              path)

            # Hmmmm, only at this point can we do some caching. Not
            # ideal, but okay.
            try:
                if path in self.castor_path_checks_cache:
                    continue
            except AttributeError:
                # This only happens the first time around.
                self.castor_path_checks_cache = []
            self.castor_path_checks_cache.append(path)

            # Now, unless we're supposed to skip this piece of the
            # path, let's make sure it exists and set the permissions
            # correctly for use by CRAB. This means that:
            # - the final output directory should (at least) have
            #   permissions 775
            # - all directories above that should (at least) have
            #   permissions 755.

            # BUT: Even though the above permissions are the usual
            # ones to used when setting up CASTOR areas for grid job
            # output, there is one caveat in case multiple people are
            # working in the same CASTOR area. If user X creates
            # /a/b/c/ and user Y wants to create /a/b/d/ he/she does
            # not have sufficient rights. So: we set all dir
            # permissions to 775 to avoid this.

            if not skip_this_path_piece:

                # Ok, first thing: let's make sure this directory
                # exists.
                # NOTE: The nice complication is of course that the
                # usual os.path.isdir() etc. methods don't work for an
                # rfio filesystem. So we call rfstat and interpret an
                # error as meaning that the path does not exist.
                self.logger.debug("Checking if path `%s' exists" % \
                                  path)
                cmd = "rfstat %s" % path
                (status, output) = commands.getstatusoutput(cmd)
                if status != 0:
                    # Path does not exist, let's try and create it.
                    self.logger.debug("Creating path `%s'" % path)
                    cmd = "nsmkdir -m 775 %s" % path
                    (status, output) = commands.getstatusoutput(cmd)
                    if status != 0:
                        msg = "Could not create directory `%s'" % path
                        self.logger.fatal(msg)
                        raise Error(msg)
                    cmd = "rfstat %s" % path
                    (status, output) = commands.getstatusoutput(cmd)
                # Now check that it looks like a directory. If I'm not
                # mistaken one can deduce this from the fact that the
                # (octal) permissions string starts with `40' (instead
                # of `100').
                permissions = extract_permissions(output)
                if not permissions.startswith("40"):
                    msg = "Path `%s' is not a directory(?)" % path
                    self.logger.fatal(msg)
                    raise Error(msg)

                # Figure out the current permissions for this
                # (partial) path.
                self.logger.debug("Checking permissions for path `%s'" % path)
                cmd = "rfstat %s" % path
                (status, output) = commands.getstatusoutput(cmd)
                if status != 0:
                    msg = "Could not obtain permissions for directory `%s'" % \
                          path
                    self.logger.fatal(msg)
                    raise Error(msg)
                # Take the last three digits of the permissions.
                permissions = extract_permissions(output)[-3:]

                # Now if necessary fix permissions.
                # NOTE: Be careful never to `downgrade' permissions.
                if piece_index == (len_castor_path_pieces - 1):
                    # This means we're looking at the final
                    # destination directory.
                    permissions_target = "775"
                else:
                    # `Only' an intermediate directory.
                    permissions_target = "775"

                # Compare permissions.
                permissions_new = []
                for (i, j) in zip(permissions, permissions_target):
                    permissions_new.append(str(max(int(i), int(j))))
                permissions_new = "".join(permissions_new)
                self.logger.debug("  current permissions: %s" % \
                                  permissions)
                self.logger.debug("  target permissions : %s" % \
                                  permissions_target)
                if permissions_new != permissions:
                    # We have to modify the permissions.
                    self.logger.debug("Changing permissions of `%s' " \
                                      "to %s (were %s)" % \
                                      (path, permissions_new, permissions))
                    cmd = "rfchmod %s %s" % (permissions_new, path)
                    (status, output) = commands.getstatusoutput(cmd)
                    if status != 0:
                        msg = "Could not change permissions for path `%s' " \
                              "to %s" % (path, permissions_new)
                        self.logger.fatal(msg)
                        raise Error(msg)

                self.logger.debug("  Permissions ok (%s)" % permissions_new)

        # End of create_and_check_castor_dir.

    ##########

    def pick_a_site(self, sites, cmssw_version):

        # Create list of forbidden sites
        sites_forbidden = []

        if (self.preferred_site == "CAF") or (self.preferred_site == "caf.cern.ch"):
            self.caf_access = True

        if self.caf_access == False:
            sites_forbidden.append("caf.cern.ch")

        # These are the T1 sites. These are only forbidden if we're
        # running in non-T1 mode.
        # Source:
        # https://cmsweb.cern.ch/sitedb/sitelist/?naming_scheme=ce
        # Hard-coded, yes. Not nice, no.

        all_t1 = [
                "srm-cms.cern.ch",
                "ccsrm.in2p3.fr",
                "cmssrm-fzk.gridka.de",
                "cmssrm.fnal.gov",
                "gridka-dCache.fzk.de",
                "srm-cms.gridpp.rl.ac.uk",
                "srm.grid.sinica.edu.tw",
                "srm2.grid.sinica.edu.tw",
                "srmcms.pic.es",
                "storm-fe-cms.cr.cnaf.infn.it"
                ]

        country_codes = {
              "CAF" : "caf.cern.ch",
              "CH" : "srm-cms.cern.ch",
              "FR" : "ccsrm.in2p3.fr",
              "DE" : "cmssrm-fzk.gridka.de",
              "GOV" : "cmssrm.fnal.gov",
              "DE2" : "gridka-dCache.fzk.de",
              "UK" : "srm-cms.gridpp.rl.ac.uk",
              "TW" : "srm.grid.sinica.edu.tw",
              "TW2" : "srm2.grid.sinica.edu.tw",
              "ES" : "srmcms.pic.es",
              "IT" : "storm-fe-cms.cr.cnaf.infn.it"
                }

        if self.non_t1access: 
            sites_forbidden.extend(all_t1)

        for site in sites_forbidden:
            if site in sites:
                sites.remove(site)

        if self.preferred_site in country_codes:
            self.preferred_site = country_codes[self.preferred_site]

        if self.preferred_site != "no preference":
            if self.preferred_site in sites:
                sites = [self.preferred_site]
            else:
                sites= []

        #print sites

        # Looks like we have to do some caching here, otherwise things
        # become waaaay toooo sloooooow. So that's what the
        # sites_and_versions_cache does.

        # NOTE: Keep this set to None!
        site_name = None
        cmd = None
        while len(sites) > 0 and \
              site_name is None:

            # Create list of t1_sites
            t1_sites = []
            for site in sites:
                if site in all_t1:
                    t1_sites.append(site)
                if site == "caf.cern.ch":
                    t1_sites.append(site)

            # If avilable pick preferred site
            #if self.preferred_site in sites:
            #  se_name = self.preferred_site
            # Else, if available pick t1 site

            if len(t1_sites) > 0:
                se_name = choice(t1_sites)
            # Else pick any site
            else:
                se_name = choice(sites)

            # But check that it hosts the CMSSW version we want.

            if se_name in self.sites_and_versions_cache and \
                   cmssw_version in self.sites_and_versions_cache[se_name]:
                if self.sites_and_versions_cache[se_name][cmssw_version]:
                    site_name = se_name
                    break
                else:
                    self.logger.debug("  --> rejecting site `%s'" % se_name)
                    sites.remove(se_name)

            else:
                self.logger.info("Checking if site `%s' " \
                                 "has CMSSW version `%s'" % \
                                 (se_name, cmssw_version))
                self.sites_and_versions_cache[se_name] = {}

                # TODO TODO TODO
                # Test for SCRAM architecture removed as per request
                # from Andreas.
                # scram_arch = os.getenv("SCRAM_ARCH")
                # cmd = "lcg-info --list-ce " \
                #       "--query '" \
                #       "Tag=VO-cms-%s," \
                #       "Tag=VO-cms-%s," \
                #       "CEStatus=Production," \
                #       "CloseSE=%s'" % \
                #       (cmssw_version, scram_arch, se_name)
                # TODO TODO TODO end

                cmd = "lcg-info --list-ce " \
                      "--query '" \
                      "Tag=VO-cms-%s," \
                      "CEStatus=Production," \
                      "CloseSE=%s'" % \
                      (cmssw_version, se_name)
                (status, output) = commands.getstatusoutput(cmd)
                if status != 0:
                    self.logger.error("Could not check site information " \
                                      "for site `%s'" % se_name)
                else:
                    if (len(output) > 0) or (se_name == "caf.cern.ch"):
                        self.sites_and_versions_cache[se_name][cmssw_version] = True
                        site_name = se_name
                        break
                    else:
                        self.sites_and_versions_cache[se_name][cmssw_version] = False
                        self.logger.debug("  --> rejecting site `%s'" % se_name)
                        sites.remove(se_name)

        if site_name is self.no_matching_site_found_str:
            self.logger.error("  --> no matching site found")
            self.logger.error("    --> Your release or SCRAM " \
                              "architecture may not be available" \
                              "anywhere on the (LCG) grid.")
            if not cmd is None:
                self.logger.debug("      (command used: `%s')" % cmd)
        else:
            self.logger.debug("  --> selected site `%s'" % site_name)

        # Return something more descriptive (than `None') in case we
        # found nothing.
        if site_name is None:
            site_name = self.no_matching_site_found_str
            # Keep track of our global flag signifying that this
            # happened.
            self.all_sites_found = False

        # End of pick_a_site.
        return site_name

    ##########

    def parse_cmd_line_options(self):

        # Set up the command line parser. Note that we fix up the help
        # formatter so that we can add some text pointing people to
        # the Twiki etc.
        parser = optparse.OptionParser(version="%s %s" % \
                                       ("%prog", self.version),
                                       formatter=CMSHarvesterHelpFormatter())

        self.option_parser = parser

        # The debug switch.
        parser.add_option("-d", "--debug",
                          help="Switch on debug mode",
                          action="callback",
                          callback=self.option_handler_debug)

        # The quiet switch.
        parser.add_option("-q", "--quiet",
                          help="Be less verbose",
                          action="callback",
                          callback=self.option_handler_quiet)

        # The force switch. If this switch is used sanity checks are
        # performed but failures do not lead to aborts. Use with care.
        parser.add_option("", "--force",
                          help="Force mode. Do not abort on sanity check "
                          "failures",
                          action="callback",
                          callback=self.option_handler_force)

        # Choose between the different kinds of harvesting.
        parser.add_option("", "--harvesting_type",
                          help="Harvesting type: %s" % \
                          ", ".join(self.harvesting_types),
                          action="callback",
                          callback=self.option_handler_harvesting_type,
                          type="string",
                          metavar="HARVESTING_TYPE")

        # Choose between single-step and two-step mode.
        parser.add_option("", "--harvesting_mode",
                          help="Harvesting mode: %s (default = %s)" % \
                          (", ".join(self.harvesting_modes),
                          self.harvesting_mode_default),
                          action="callback",
                          callback=self.option_handler_harvesting_mode,
                          type="string",
                          metavar="HARVESTING_MODE")

        # Override the GlobalTag chosen by the cmsHarvester.
        parser.add_option("", "--globaltag",
                          help="GlobalTag to use. Default is the ones " \
                          "the dataset was created with for MC, for data" \
                          "a GlobalTag has to be specified.",
                          action="callback",
                          callback=self.option_handler_globaltag,
                          type="string",
                          metavar="GLOBALTAG")

        # Allow switching off of reference histograms.
        parser.add_option("", "--no-ref-hists",
                          help="Don't use any reference histograms",
                          action="callback",
                          callback=self.option_handler_no_ref_hists)

        # Allow the default (i.e. the one that should be used)
        # Frontier connection to be overridden.
        parser.add_option("", "--frontier-connection",
                          help="Use this Frontier connection to find " \
                          "GlobalTags and LocalTags (for reference " \
                          "histograms).\nPlease only use this for " \
                          "testing.",
                          action="callback",
                          callback=self.option_handler_frontier_connection,
                          type="string",
                          metavar="FRONTIER")

        # Similar to the above but specific to the Frontier connection
        # to be used for the GlobalTag.
        parser.add_option("", "--frontier-connection-for-globaltag",
                          help="Use this Frontier connection to find " \
                          "GlobalTags.\nPlease only use this for " \
                          "testing.",
                          action="callback",
                          callback=self.option_handler_frontier_connection,
                          type="string",
                          metavar="FRONTIER")

        # Similar to the above but specific to the Frontier connection
        # to be used for the reference histograms.
        parser.add_option("", "--frontier-connection-for-refhists",
                          help="Use this Frontier connection to find " \
                          "LocalTags (for reference " \
                          "histograms).\nPlease only use this for " \
                          "testing.",
                          action="callback",
                          callback=self.option_handler_frontier_connection,
                          type="string",
                          metavar="FRONTIER")

        # Option to specify the name (or a regexp) of the dataset(s)
        # to be used.
        parser.add_option("", "--dataset",
                          help="Name (or regexp) of dataset(s) to process",
                          action="callback",
                          #callback=self.option_handler_dataset_name,
                          callback=self.option_handler_input_spec,
                          type="string",
                          #dest="self.input_name",
                          metavar="DATASET")

        # Option to specify the name (or a regexp) of the dataset(s)
        # to be ignored.
        parser.add_option("", "--dataset-ignore",
                          help="Name (or regexp) of dataset(s) to ignore",
                          action="callback",
                          callback=self.option_handler_input_spec,
                          type="string",
                          metavar="DATASET-IGNORE")

        # Option to specify the name (or a regexp) of the run(s)
        # to be used.
        parser.add_option("", "--runs",
                          help="Run number(s) to process",
                          action="callback",
                          callback=self.option_handler_input_spec,
                          type="string",
                          metavar="RUNS")                   

        # Option to specify the name (or a regexp) of the run(s)
        # to be ignored.
        parser.add_option("", "--runs-ignore",
                          help="Run number(s) to ignore",
                          action="callback",
                          callback=self.option_handler_input_spec,
                          type="string",
                          metavar="RUNS-IGNORE")

        # Option to specify a file containing a list of dataset names
        # (or regexps) to be used.
        parser.add_option("", "--datasetfile",
                          help="File containing list of dataset names " \
                          "(or regexps) to process",
                          action="callback",
                          #callback=self.option_handler_listfile_name,
                          callback=self.option_handler_input_spec,
                          type="string",
                          #dest="self.input_name",
                          metavar="DATASETFILE")

        # Option to specify a file containing a list of dataset names
        # (or regexps) to be ignored.
        parser.add_option("", "--datasetfile-ignore",
                          help="File containing list of dataset names " \
                          "(or regexps) to ignore",
                          action="callback",
                          callback=self.option_handler_input_spec,
                          type="string",
                          metavar="DATASETFILE-IGNORE")

        # Option to specify a file containing a list of runs to be
        # used.
        parser.add_option("", "--runslistfile",
                          help="File containing list of run numbers " \
                          "to process",
                          action="callback",
                          callback=self.option_handler_input_spec,
                          type="string",
                          metavar="RUNSLISTFILE")

        # Option to specify a file containing a list of runs
        # to be ignored.
        parser.add_option("", "--runslistfile-ignore",
                          help="File containing list of run numbers " \
                          "to ignore",
                          action="callback",
                          callback=self.option_handler_input_spec,
                          type="string",
                          metavar="RUNSLISTFILE-IGNORE")

        # Option to specify a Jsonfile contaning a list of runs
        # to be used.
        parser.add_option("", "--Jsonrunfile",
                          help="Jsonfile containing dictionary of run/lumisections pairs. " \
                          "All lumisections of runs contained in dictionary are processed.",
                          action="callback",
                          callback=self.option_handler_input_Jsonrunfile,
                          type="string",
                          metavar="JSONRUNFILE")

        # Option to specify a Jsonfile contaning a dictionary of run/lumisections pairs
        # to be used.
        parser.add_option("", "--Jsonfile",
                          help="Jsonfile containing dictionary of run/lumisections pairs. " \
                          "Only specified lumisections of runs contained in dictionary are processed.",
                          action="callback",
                          callback=self.option_handler_input_Jsonfile,
                          type="string",
                          metavar="JSONFILE")

        # Option to specify a ToDo file contaning a list of runs
        # to be used.
        parser.add_option("", "--todo-file",
                          help="Todo file containing a list of runs to process.",
                          action="callback",
                          callback=self.option_handler_input_todofile,
                          type="string",
                          metavar="TODO-FILE")

        # Option to specify which file to use for the dataset name to
        # reference histogram name mappings.
        parser.add_option("", "--refhistmappingfile",
                          help="File to be use for the reference " \
                          "histogram mappings. Default: `%s'." % \
                          self.ref_hist_mappings_file_name_default,
                          action="callback",
                          callback=self.option_handler_ref_hist_mapping_file,
                          type="string",
                          metavar="REFHISTMAPPING-FILE")

        # Specify the place in CASTOR where the output should go.
        # NOTE: Only output to CASTOR is supported for the moment,
        # since the central DQM results place is on CASTOR anyway.
        parser.add_option("", "--castordir",
                          help="Place on CASTOR to store results. " \
                          "Default: `%s'." % \
                          self.castor_base_dir_default,
                          action="callback",
                          callback=self.option_handler_castor_dir,
                          type="string",
                          metavar="CASTORDIR")

        # Use this to try and create jobs that will run without
        # special `t1access' role.
        parser.add_option("", "--no-t1access",
                          help="Try to create jobs that will run " \
                          "without special `t1access' role",
                          action="callback",
                          callback=self.option_handler_no_t1access)

        # Use this to create jobs that may run on CAF
        parser.add_option("", "--caf-access",
                          help="Crab jobs may run " \
                          "on CAF",
                          action="callback",
                          callback=self.option_handler_caf_access)

        # set process.dqmSaver.saveByLumiSection=1 in harvesting cfg file
        parser.add_option("", "--saveByLumiSection",
                          help="set saveByLumiSection=1 in harvesting cfg file",
                          action="callback",
                          callback=self.option_handler_saveByLumiSection)

        # Use this to enable automatic creation and submission of crab jobs
        parser.add_option("", "--automatic-crab-submission",
                          help="Crab jobs are created and " \
                          "submitted automatically",
                          action="callback",
                          callback=self.option_handler_crab_submission)

        # Option to set the max number of sites, each
        #job is submitted to 
        parser.add_option("", "--max-sites",
                          help="Max. number of sites each job is submitted to",
                          action="callback",
                          callback=self.option_handler_sites,
                          type="int") 

        # Option to set the preferred site
        parser.add_option("", "--site",
                          help="Crab jobs are submitted to specified site. T1 sites may be shortened by the following (country) codes: \
                          srm-cms.cern.ch : CH \
                          ccsrm.in2p3.fr : FR \
                          cmssrm-fzk.gridka.de : DE \
                          cmssrm.fnal.gov : GOV \
                          gridka-dCache.fzk.de : DE2 \
                          rm-cms.gridpp.rl.ac.uk : UK \
                          srm.grid.sinica.edu.tw : TW \
                          srm2.grid.sinica.edu.tw : TW2 \
                          srmcms.pic.es : ES \
                          storm-fe-cms.cr.cnaf.infn.it : IT",
                          action="callback",
                          callback=self.option_handler_preferred_site,
                          type="string") 

        # This is the command line flag to list all harvesting
        # type-to-sequence mappings.
        parser.add_option("-l", "--list",
                          help="List all harvesting types and their" \
                          "corresponding sequence names",
                          action="callback",
                          callback=self.option_handler_list_types)

        # If nothing was specified: tell the user how to do things the
        # next time and exit.
        # NOTE: We just use the OptParse standard way of doing this by
        #       acting as if a '--help' was specified.
        if len(self.cmd_line_opts) < 1:
            self.cmd_line_opts = ["--help"]

        # Some trickery with the options. Why? Well, since these
        # options change the output level immediately from the option
        # handlers, the results differ depending on where they are on
        # the command line. Let's just make sure they are at the
        # front.
        # NOTE: Not very efficient or sophisticated, but it works and
        # it only has to be done once anyway.
        for i in ["-d", "--debug",
                  "-q", "--quiet"]:
            if i in self.cmd_line_opts:
                self.cmd_line_opts.remove(i)
                self.cmd_line_opts.insert(0, i)

        # Everything is set up, now parse what we were given.
        parser.set_defaults()
        (self.options, self.args) = parser.parse_args(self.cmd_line_opts)

        # End of parse_cmd_line_options.

    ##########

    def check_input_status(self):
        """Check completeness and correctness of input information.

        Check that all required information has been specified and
        that, at least as far as can be easily checked, it makes
        sense.

        NOTE: This is also where any default values are applied.

        """

        self.logger.info("Checking completeness/correctness of input...")

        # The cmsHarvester does not take (i.e. understand) any
        # arguments so there should not be any.
        if len(self.args) > 0:
            msg = "Sorry but I don't understand `%s'" % \
                  (" ".join(self.args))
            self.logger.fatal(msg)
            raise Usage(msg)

        # BUG BUG BUG
        # While we wait for some bugs left and right to get fixed, we
        # disable two-step.
        if self.harvesting_mode == "two-step":
            msg = "--------------------\n" \
                  "  Sorry, but for the moment (well, till it works)" \
                  "  the two-step mode has been disabled.\n" \
                  "--------------------\n"
            self.logger.fatal(msg)
            raise Error(msg)
        # BUG BUG BUG end

        # We need a harvesting method to be specified
        if self.harvesting_type is None:
            msg = "Please specify a harvesting type"
            self.logger.fatal(msg)
            raise Usage(msg)
        # as well as a harvesting mode.
        if self.harvesting_mode is None:
            self.harvesting_mode = self.harvesting_mode_default
            msg = "No harvesting mode specified --> using default `%s'" % \
                  self.harvesting_mode
            self.logger.warning(msg)
            #raise Usage(msg)

        ###

        # We need an input method so we can find the dataset name(s).
        if self.input_method["datasets"]["use"] is None:
            msg = "Please specify an input dataset name " \
                  "or a list file name"
            self.logger.fatal(msg)
            raise Usage(msg)

        # DEBUG DEBUG DEBUG
        # If we get here, we should also have an input name.
        assert not self.input_name["datasets"]["use"] is None
        # DEBUG DEBUG DEBUG end

        ###

        # The same holds for the reference histogram mapping file (if
        # we're using references).
        if self.use_ref_hists:
            if self.ref_hist_mappings_file_name is None:
                self.ref_hist_mappings_file_name = self.ref_hist_mappings_file_name_default
                msg = "No reference histogram mapping file specified --> " \
                      "using default `%s'" % \
                      self.ref_hist_mappings_file_name
                self.logger.warning(msg)

        ###

        # We need to know where to put the stuff (okay, the results)
        # on CASTOR.
        if self.castor_base_dir is None:
            self.castor_base_dir = self.castor_base_dir_default
            msg = "No CASTOR area specified -> using default `%s'" % \
                  self.castor_base_dir
            self.logger.warning(msg)
            #raise Usage(msg)

        # Only the CERN CASTOR area is supported.
        if not self.castor_base_dir.startswith(self.castor_prefix):
            msg = "CASTOR area does not start with `%s'" % \
                  self.castor_prefix
            self.logger.fatal(msg)
            if self.castor_base_dir.find("castor") > -1 and \
               not self.castor_base_dir.find("cern.ch") > -1:
                self.logger.fatal("Only CERN CASTOR is supported")
            raise Usage(msg)

        ###

        # TODO TODO TODO
        # This should be removed in the future, once I find out how to
        # get the config file used to create a given dataset from DBS.

        # For data we need to have a GlobalTag. (For MC we can figure
        # it out by ourselves.)
        if self.globaltag is None:
            self.logger.warning("No GlobalTag specified. This means I cannot")
            self.logger.warning("run on data, only on MC.")
            self.logger.warning("I will skip all data datasets.")

        # TODO TODO TODO end

        # Make sure the GlobalTag ends with `::All'.
        if not self.globaltag is None:
            if not self.globaltag.endswith("::All"):
                self.logger.warning("Specified GlobalTag `%s' does " \
                                    "not end in `::All' --> " \
                                    "appending this missing piece" % \
                                    self.globaltag)
                self.globaltag = "%s::All" % self.globaltag

        ###

        # Dump some info about the Frontier connections used.
        for (key, value) in six.iteritems(self.frontier_connection_name):
            frontier_type_str = "unknown"
            if key == "globaltag":
                frontier_type_str = "the GlobalTag"
            elif key == "refhists":
                frontier_type_str = "the reference histograms"
            non_str = None
            if self.frontier_connection_overridden[key] == True:
                non_str = "non-"
            else:
                non_str = ""
            self.logger.info("Using %sdefault Frontier " \
                             "connection for %s: `%s'" % \
                             (non_str, frontier_type_str, value))

        ###

        # End of check_input_status.

    ##########

    def check_cmssw(self):
        """Check if CMSSW is setup.

        """

        # Try to access the CMSSW_VERSION environment variable. If
        # it's something useful we consider CMSSW to be set up
        # properly. Otherwise we raise an error.
        cmssw_version = os.getenv("CMSSW_VERSION")
        if cmssw_version is None:
            self.logger.fatal("It seems CMSSW is not setup...")
            self.logger.fatal("($CMSSW_VERSION is empty)")
            raise Error("ERROR: CMSSW needs to be setup first!")

        self.cmssw_version = cmssw_version
        self.logger.info("Found CMSSW version %s properly set up" % \
                          self.cmssw_version)

        # End of check_cmsssw.
        return True

    ##########

    def check_dbs(self):
        """Check if DBS is setup.

        """

        # Try to access the DBSCMD_HOME environment variable. If this
        # looks useful we consider DBS to be set up
        # properly. Otherwise we raise an error.
        dbs_home = os.getenv("DBSCMD_HOME")
        if dbs_home is None:
            self.logger.fatal("It seems DBS is not setup...")
            self.logger.fatal("  $DBSCMD_HOME is empty")
            raise Error("ERROR: DBS needs to be setup first!")

##        # Now we try to do a very simple DBS search. If that works
##        # instead of giving us the `Unsupported API call' crap, we
##        # should be good to go.
##        # NOTE: Not ideal, I know, but it reduces the amount of
##        #       complaints I get...
##        cmd = "dbs search --query=\"find dataset where dataset = impossible\""
##        (status, output) = commands.getstatusoutput(cmd)
##        pdb.set_trace()
##        if status != 0 or \
##           output.lower().find("unsupported api call") > -1:
##            self.logger.fatal("It seems DBS is not setup...")
##            self.logger.fatal("  %s returns crap:" % cmd)
##            for line in output.split("\n"):
##                self.logger.fatal("  %s" % line)
##            raise Error("ERROR: DBS needs to be setup first!")

        self.logger.debug("Found DBS properly set up")

        # End of check_dbs.
        return True

    ##########

    def setup_dbs(self):
        """Setup the Python side of DBS.

        For more information see the DBS Python API documentation:
        https://twiki.cern.ch/twiki/bin/view/CMS/DBSApiDocumentation

        """

        try:
            args={}
            args["url"]= "http://cmsdbsprod.cern.ch/cms_dbs_prod_global/" \
                         "servlet/DBSServlet"
            api = DbsApi(args)
            self.dbs_api = api

        except DBSAPI.dbsApiException.DbsApiException as ex:
            self.logger.fatal("Caught DBS API exception %s: %s "  % \
                              (ex.getClassName(), ex.getErrorMessage()))
            if ex.getErrorCode() not in (None, ""):
                logger.debug("DBS exception error code: ", ex.getErrorCode())
            raise

        # End of setup_dbs.

    ##########

    def dbs_resolve_dataset_name(self, dataset_name):
        """Use DBS to resolve a wildcarded dataset name.

        """

        # DEBUG DEBUG DEBUG
        # If we get here DBS should have been set up already.
        assert not self.dbs_api is None
        # DEBUG DEBUG DEBUG end

        # Some minor checking to make sure that whatever we've been
        # given as dataset name actually sounds like a dataset name.
        if not (dataset_name.startswith("/") and \
                dataset_name.endswith("RECO")):
            self.logger.warning("Dataset name `%s' does not sound " \
                                 "like a valid dataset name!" % \
                                dataset_name)

        #----------

        api = self.dbs_api
        dbs_query = "find dataset where dataset like %s " \
                    "and dataset.status = VALID" % \
                    dataset_name
        try:
            api_result = api.executeQuery(dbs_query)
        except DBSAPI.dbsApiException.DbsApiException:
            msg = "ERROR: Could not execute DBS query"
            self.logger.fatal(msg)
            raise Error(msg)

        # Setup parsing.
        handler = DBSXMLHandler(["dataset"])
        parser = xml.sax.make_parser()
        parser.setContentHandler(handler)

        # Parse.
        try:
            xml.sax.parseString(api_result, handler)
        except SAXParseException:
            msg = "ERROR: Could not parse DBS server output"
            self.logger.fatal(msg)
            raise Error(msg)

        # DEBUG DEBUG DEBUG
        assert(handler.check_results_validity()), "ERROR The DBSXMLHandler screwed something up!"
        # DEBUG DEBUG DEBUG end

        # Extract the results.
        datasets = handler.results.values()[0]

        # End of dbs_resolve_dataset_name.
        return datasets

    ##########

    def dbs_resolve_cmssw_version(self, dataset_name):
        """Ask DBS for the CMSSW version used to create this dataset.

        """

        # DEBUG DEBUG DEBUG
        # If we get here DBS should have been set up already.
        assert not self.dbs_api is None
        # DEBUG DEBUG DEBUG end

        api = self.dbs_api
        dbs_query = "find algo.version where dataset = %s " \
                    "and dataset.status = VALID" % \
                    dataset_name
        try:
            api_result = api.executeQuery(dbs_query)
        except DBSAPI.dbsApiException.DbsApiException:
            msg = "ERROR: Could not execute DBS query"
            self.logger.fatal(msg)
            raise Error(msg)

        handler = DBSXMLHandler(["algo.version"])
        parser = xml.sax.make_parser()
        parser.setContentHandler(handler)

        try:
            xml.sax.parseString(api_result, handler)
        except SAXParseException:
            msg = "ERROR: Could not parse DBS server output"
            self.logger.fatal(msg)
            raise Error(msg)

        # DEBUG DEBUG DEBUG
        assert(handler.check_results_validity()), "ERROR The DBSXMLHandler screwed something up!"
        # DEBUG DEBUG DEBUG end

        cmssw_version = handler.results.values()[0]

        # DEBUG DEBUG DEBUG
        assert len(cmssw_version) == 1
        # DEBUG DEBUG DEBUG end

        cmssw_version = cmssw_version[0]

        # End of dbs_resolve_cmssw_version.
        return cmssw_version

    ##########

##    def dbs_resolve_dataset_number_of_events(self, dataset_name):
##        """Ask DBS across how many events this dataset has been spread
##        out.

##        This is especially useful to check that we do not submit a job
##        supposed to run on a complete sample that is not contained at
##        a single site.

##        """

##        # DEBUG DEBUG DEBUG
##        # If we get here DBS should have been set up already.
##        assert not self.dbs_api is None
##        # DEBUG DEBUG DEBUG end

##        api = self.dbs_api
##        dbs_query = "find count(site) where dataset = %s " \
##                    "and dataset.status = VALID" % \
##                    dataset_name
##        try:
##            api_result = api.executeQuery(dbs_query)
##        except DbsApiException:
##            raise Error("ERROR: Could not execute DBS query")

##        try:
##            num_events = []
##            class Handler(xml.sax.handler.ContentHandler):
##                def startElement(self, name, attrs):
##                    if name == "result":
##                        num_events.append(str(attrs["COUNT_STORAGEELEMENT"]))
##            xml.sax.parseString(api_result, Handler())
##        except SAXParseException:
##            raise Error("ERROR: Could not parse DBS server output")

##        # DEBUG DEBUG DEBUG
##        assert len(num_events) == 1
##        # DEBUG DEBUG DEBUG end

##        num_events = int(num_events[0])

##        # End of dbs_resolve_dataset_number_of_events.
##        return num_events

    ##########

    def dbs_resolve_runs(self, dataset_name):
        """Ask DBS for the list of runs in a given dataset.

        # NOTE: This does not (yet?) skip/remove empty runs. There is
        # a bug in the DBS entry run.numevents (i.e. it always returns
        # zero) which should be fixed in the `next DBS release'.
        # See also:
        #   https://savannah.cern.ch/bugs/?53452
        #   https://savannah.cern.ch/bugs/?53711

        """

        # TODO TODO TODO
        # We should remove empty runs as soon as the above mentioned
        # bug is fixed.
        # TODO TODO TODO end

        # DEBUG DEBUG DEBUG
        # If we get here DBS should have been set up already.
        assert not self.dbs_api is None
        # DEBUG DEBUG DEBUG end

        api = self.dbs_api
        dbs_query = "find run where dataset = %s " \
                    "and dataset.status = VALID" % \
                    dataset_name
        try:
            api_result = api.executeQuery(dbs_query)
        except DBSAPI.dbsApiException.DbsApiException:
            msg = "ERROR: Could not execute DBS query"
            self.logger.fatal(msg)
            raise Error(msg)

        handler = DBSXMLHandler(["run"])
        parser = xml.sax.make_parser()
        parser.setContentHandler(handler)

        try:
            xml.sax.parseString(api_result, handler)
        except SAXParseException:
            msg = "ERROR: Could not parse DBS server output"
            self.logger.fatal(msg)
            raise Error(msg)

        # DEBUG DEBUG DEBUG
        assert(handler.check_results_validity()), "ERROR The DBSXMLHandler screwed something up!"
        # DEBUG DEBUG DEBUG end

        runs = handler.results.values()[0]
        # Turn strings into integers.
        runs = sorted([int(i) for i in runs])

        # End of dbs_resolve_runs.
        return runs

    ##########

    def dbs_resolve_globaltag(self, dataset_name):
        """Ask DBS for the globaltag corresponding to a given dataset.

        # BUG BUG BUG
        # This does not seem to work for data datasets? E.g. for
        # /Cosmics/Commissioning08_CRAFT0831X_V1_311_ReReco_FromSuperPointing_v1/RAW-RECO
        # Probaly due to the fact that the GlobalTag changed during
        # datataking...
        BUG BUG BUG end

        """

        # DEBUG DEBUG DEBUG
        # If we get here DBS should have been set up already.
        assert not self.dbs_api is None
        # DEBUG DEBUG DEBUG end

        api = self.dbs_api
        dbs_query = "find dataset.tag where dataset = %s " \
                    "and dataset.status = VALID" % \
                    dataset_name
        try:
            api_result = api.executeQuery(dbs_query)
        except DBSAPI.dbsApiException.DbsApiException:
            msg = "ERROR: Could not execute DBS query"
            self.logger.fatal(msg)
            raise Error(msg)

        handler = DBSXMLHandler(["dataset.tag"])
        parser = xml.sax.make_parser()
        parser.setContentHandler(parser)

        try:
            xml.sax.parseString(api_result, handler)
        except SAXParseException:
            msg = "ERROR: Could not parse DBS server output"
            self.logger.fatal(msg)
            raise Error(msg)

        # DEBUG DEBUG DEBUG
        assert(handler.check_results_validity()), "ERROR The DBSXMLHandler screwed something up!"
        # DEBUG DEBUG DEBUG end

        globaltag = handler.results.values()[0]

        # DEBUG DEBUG DEBUG
        assert len(globaltag) == 1
        # DEBUG DEBUG DEBUG end

        globaltag = globaltag[0]

        # End of dbs_resolve_globaltag.
        return globaltag

    ##########

    def dbs_resolve_datatype(self, dataset_name):
        """Ask DBS for the the data type (data or mc) of a given
        dataset.

        """

        # DEBUG DEBUG DEBUG
        # If we get here DBS should have been set up already.
        assert not self.dbs_api is None
        # DEBUG DEBUG DEBUG end

        api = self.dbs_api
        dbs_query = "find datatype.type where dataset = %s " \
                    "and dataset.status = VALID" % \
                    dataset_name
        try:
            api_result = api.executeQuery(dbs_query)
        except DBSAPI.dbsApiException.DbsApiException:
            msg = "ERROR: Could not execute DBS query"
            self.logger.fatal(msg)
            raise Error(msg)

        handler = DBSXMLHandler(["datatype.type"])
        parser = xml.sax.make_parser()
        parser.setContentHandler(handler)

        try:
            xml.sax.parseString(api_result, handler)
        except SAXParseException:
            msg = "ERROR: Could not parse DBS server output"
            self.logger.fatal(msg)
            raise Error(msg)

        # DEBUG DEBUG DEBUG
        assert(handler.check_results_validity()), "ERROR The DBSXMLHandler screwed something up!"
        # DEBUG DEBUG DEBUG end

        datatype = handler.results.values()[0]

        # DEBUG DEBUG DEBUG
        assert len(datatype) == 1
        # DEBUG DEBUG DEBUG end

        datatype = datatype[0]

        # End of dbs_resolve_datatype.
        return datatype

    ##########

    # OBSOLETE OBSOLETE OBSOLETE
    # This method is no longer used.

    def dbs_resolve_number_of_events(self, dataset_name, run_number=None):
        """Determine the number of events in a given dataset (and run).

        Ask DBS for the number of events in a dataset. If a run number
        is specified the number of events returned is that in that run
        of that dataset. If problems occur we throw an exception.

        # BUG BUG BUG
        # Since DBS does not return the number of events correctly,
        # neither for runs nor for whole datasets, we have to work
        # around that a bit...
        # BUG BUG BUG end

        """

        # DEBUG DEBUG DEBUG
        # If we get here DBS should have been set up already.
        assert not self.dbs_api is None
        # DEBUG DEBUG DEBUG end

        api = self.dbs_api
        dbs_query = "find file.name, file.numevents where dataset = %s " \
                    "and dataset.status = VALID" % \
                    dataset_name
        if not run_number is None:
            dbs_query = dbq_query + (" and run = %d" % run_number)
        try:
            api_result = api.executeQuery(dbs_query)
        except DBSAPI.dbsApiException.DbsApiException:
            msg = "ERROR: Could not execute DBS query"
            self.logger.fatal(msg)
            raise Error(msg)

        handler = DBSXMLHandler(["file.name", "file.numevents"])
        parser = xml.sax.make_parser()
        parser.setContentHandler(handler)

        try:
            xml.sax.parseString(api_result, handler)
        except SAXParseException:
            msg = "ERROR: Could not parse DBS server output"
            self.logger.fatal(msg)
            raise Error(msg)

        # DEBUG DEBUG DEBUG
        assert(handler.check_results_validity()), "ERROR The DBSXMLHandler screwed something up!"
        # DEBUG DEBUG DEBUG end

        num_events = sum(handler.results["file.numevents"])

        # End of dbs_resolve_number_of_events.
        return num_events

    # OBSOLETE OBSOLETE OBSOLETE end

    ##########

##    def dbs_resolve_dataset_number_of_sites(self, dataset_name):
##        """Ask DBS across how many sites this dataset has been spread
##        out.

##        This is especially useful to check that we do not submit a job
##        supposed to run on a complete sample that is not contained at
##        a single site.

##        """

##        # DEBUG DEBUG DEBUG
##        # If we get here DBS should have been set up already.
##        assert not self.dbs_api is None
##        # DEBUG DEBUG DEBUG end

##        api = self.dbs_api
##        dbs_query = "find count(site) where dataset = %s " \
##                    "and dataset.status = VALID" % \
##                    dataset_name
##        try:
##            api_result = api.executeQuery(dbs_query)
##        except DbsApiException:
##            raise Error("ERROR: Could not execute DBS query")

##        try:
##            num_sites = []
##            class Handler(xml.sax.handler.ContentHandler):
##                def startElement(self, name, attrs):
##                    if name == "result":
##                        num_sites.append(str(attrs["COUNT_STORAGEELEMENT"]))
##            xml.sax.parseString(api_result, Handler())
##        except SAXParseException:
##            raise Error("ERROR: Could not parse DBS server output")

##        # DEBUG DEBUG DEBUG
##        assert len(num_sites) == 1
##        # DEBUG DEBUG DEBUG end

##        num_sites = int(num_sites[0])

##        # End of dbs_resolve_dataset_number_of_sites.
##        return num_sites

    ##########

##    def dbs_check_dataset_spread(self, dataset_name):
##        """Figure out across how many sites this dataset is spread.

##        NOTE: This is something we need to figure out per run, since
##        we want to submit harvesting jobs per run.

##        Basically three things can happen with a given dataset:
##        - the whole dataset is available on a single site,
##        - the whole dataset is available (mirrored) at multiple sites,
##        - the dataset is spread across multiple sites and there is no
##          single site containing the full dataset in one place.

##        NOTE: If all goes well, it should not be possible that
##        anything but a _full_ dataset is mirrored. So we ignore the
##        possibility in which for example one site contains the full
##        dataset and two others mirror half of it.
##        ANOTHER NOTE: According to some people this last case _could_
##        actually happen. I will not design for it, but make sure it
##        ends up as a false negative, in which case we just loose some
##        efficiency and treat the dataset (unnecessarily) as
##        spread-out.

##        We don't really care about the first two possibilities, but in
##        the third case we need to make sure to run the harvesting in
##        two-step mode.

##        This method checks with DBS which of the above cases is true
##        for the dataset name given, and returns a 1 for the first two
##        cases, and the number of sites across which the dataset is
##        spread for the third case.

##        The way in which this is done is by asking how many files each
##        site has for the dataset. In the first case there is only one
##        site, in the second case all sites should have the same number
##        of files (i.e. the total number of files in the dataset) and
##        in the third case the file counts from all sites should add up
##        to the total file count for the dataset.

##        """

##        # DEBUG DEBUG DEBUG
##        # If we get here DBS should have been set up already.
##        assert not self.dbs_api is None
##        # DEBUG DEBUG DEBUG end

##        api = self.dbs_api
##        dbs_query = "find run, run.numevents, site, file.count " \
##                    "where dataset = %s " \
##                    "and dataset.status = VALID" % \
##                    dataset_name
##        try:
##            api_result = api.executeQuery(dbs_query)
##        except DbsApiException:
##            msg = "ERROR: Could not execute DBS query"
##            self.logger.fatal(msg)
##            raise Error(msg)

##        # Index things by run number. No cross-check is done to make
##        # sure we get results for each and every run in the
##        # dataset. I'm not sure this would make sense since we'd be
##        # cross-checking DBS info with DBS info anyway. Note that we
##        # use the file count per site to see if we're dealing with an
##        # incomplete vs. a mirrored dataset.
##        sample_info = {}
##        try:
##            class Handler(xml.sax.handler.ContentHandler):
##                def startElement(self, name, attrs):
##                    if name == "result":
##                        run_number = int(attrs["RUNS_RUNNUMBER"])
##                        site_name = str(attrs["STORAGEELEMENT_SENAME"])
##                        file_count = int(attrs["COUNT_FILES"])
##                        # BUG BUG BUG
##                        # Doh! For some reason DBS never returns any other
##                        # event count than zero.
##                        event_count = int(attrs["RUNS_NUMBEROFEVENTS"])
##                        # BUG BUG BUG end
##                        info = (site_name, file_count, event_count)
##                        try:
##                            sample_info[run_number].append(info)
##                        except KeyError:
##                            sample_info[run_number] = [info]
##            xml.sax.parseString(api_result, Handler())
##        except SAXParseException:
##            msg = "ERROR: Could not parse DBS server output"
##            self.logger.fatal(msg)
##            raise Error(msg)

##        # Now translate this into a slightly more usable mapping.
##        sites = {}
##        for (run_number, site_info) in six.iteritems(sample_info):
##            # Quick-n-dirty trick to see if all file counts are the
##            # same.
##            unique_file_counts = set([i[1] for i in site_info])
##            if len(unique_file_counts) == 1:
##                # Okay, so this must be a mirrored dataset.
##                # We have to pick one but we have to be careful. We
##                # cannot submit to things like a T0, a T1, or CAF.
##                site_names = [self.pick_a_site([i[0] for i in site_info])]
##                nevents = [site_info[0][2]]
##            else:
##                # Looks like this is a spread-out sample.
##                site_names = [i[0] for i in site_info]
##                nevents = [i[2] for i in site_info]
##            sites[run_number] = zip(site_names, nevents)

##        self.logger.debug("Sample `%s' spread is:" % dataset_name)
##        run_numbers = sites.keys()
##        run_numbers.sort()
##        for run_number in run_numbers:
##            self.logger.debug("  run # %6d: %d sites (%s)" % \
##                              (run_number,
##                               len(sites[run_number]),
##                               ", ".join([i[0] for i in sites[run_number]])))

##        # End of dbs_check_dataset_spread.
##        return sites

##    # DEBUG DEBUG DEBUG
##    # Just kept for debugging now.
##    def dbs_check_dataset_spread_old(self, dataset_name):
##        """Figure out across how many sites this dataset is spread.

##        NOTE: This is something we need to figure out per run, since
##        we want to submit harvesting jobs per run.

##        Basically three things can happen with a given dataset:
##        - the whole dataset is available on a single site,
##        - the whole dataset is available (mirrored) at multiple sites,
##        - the dataset is spread across multiple sites and there is no
##          single site containing the full dataset in one place.

##        NOTE: If all goes well, it should not be possible that
##        anything but a _full_ dataset is mirrored. So we ignore the
##        possibility in which for example one site contains the full
##        dataset and two others mirror half of it.
##        ANOTHER NOTE: According to some people this last case _could_
##        actually happen. I will not design for it, but make sure it
##        ends up as a false negative, in which case we just loose some
##        efficiency and treat the dataset (unnecessarily) as
##        spread-out.

##        We don't really care about the first two possibilities, but in
##        the third case we need to make sure to run the harvesting in
##        two-step mode.

##        This method checks with DBS which of the above cases is true
##        for the dataset name given, and returns a 1 for the first two
##        cases, and the number of sites across which the dataset is
##        spread for the third case.

##        The way in which this is done is by asking how many files each
##        site has for the dataset. In the first case there is only one
##        site, in the second case all sites should have the same number
##        of files (i.e. the total number of files in the dataset) and
##        in the third case the file counts from all sites should add up
##        to the total file count for the dataset.

##        """

##        # DEBUG DEBUG DEBUG
##        # If we get here DBS should have been set up already.
##        assert not self.dbs_api is None
##        # DEBUG DEBUG DEBUG end

##        api = self.dbs_api
##        dbs_query = "find run, run.numevents, site, file.count " \
##                    "where dataset = %s " \
##                    "and dataset.status = VALID" % \
##                    dataset_name
##        try:
##            api_result = api.executeQuery(dbs_query)
##        except DbsApiException:
##            msg = "ERROR: Could not execute DBS query"
##            self.logger.fatal(msg)
##            raise Error(msg)

##        # Index things by run number. No cross-check is done to make
##        # sure we get results for each and every run in the
##        # dataset. I'm not sure this would make sense since we'd be
##        # cross-checking DBS info with DBS info anyway. Note that we
##        # use the file count per site to see if we're dealing with an
##        # incomplete vs. a mirrored dataset.
##        sample_info = {}
##        try:
##            class Handler(xml.sax.handler.ContentHandler):
##                def startElement(self, name, attrs):
##                    if name == "result":
##                        run_number = int(attrs["RUNS_RUNNUMBER"])
##                        site_name = str(attrs["STORAGEELEMENT_SENAME"])
##                        file_count = int(attrs["COUNT_FILES"])
##                        # BUG BUG BUG
##                        # Doh! For some reason DBS never returns any other
##                        # event count than zero.
##                        event_count = int(attrs["RUNS_NUMBEROFEVENTS"])
##                        # BUG BUG BUG end
##                        info = (site_name, file_count, event_count)
##                        try:
##                            sample_info[run_number].append(info)
##                        except KeyError:
##                            sample_info[run_number] = [info]
##            xml.sax.parseString(api_result, Handler())
##        except SAXParseException:
##            msg = "ERROR: Could not parse DBS server output"
##            self.logger.fatal(msg)
##            raise Error(msg)

##        # Now translate this into a slightly more usable mapping.
##        sites = {}
##        for (run_number, site_info) in six.iteritems(sample_info):
##            # Quick-n-dirty trick to see if all file counts are the
##            # same.
##            unique_file_counts = set([i[1] for i in site_info])
##            if len(unique_file_counts) == 1:
##                # Okay, so this must be a mirrored dataset.
##                # We have to pick one but we have to be careful. We
##                # cannot submit to things like a T0, a T1, or CAF.
##                site_names = [self.pick_a_site([i[0] for i in site_info])]
##                nevents = [site_info[0][2]]
##            else:
##                # Looks like this is a spread-out sample.
##                site_names = [i[0] for i in site_info]
##                nevents = [i[2] for i in site_info]
##            sites[run_number] = zip(site_names, nevents)

##        self.logger.debug("Sample `%s' spread is:" % dataset_name)
##        run_numbers = sites.keys()
##        run_numbers.sort()
##        for run_number in run_numbers:
##            self.logger.debug("  run # %6d: %d site(s) (%s)" % \
##                              (run_number,
##                               len(sites[run_number]),
##                               ", ".join([i[0] for i in sites[run_number]])))

##        # End of dbs_check_dataset_spread_old.
##        return sites
##    # DEBUG DEBUG DEBUG end

    ##########

    def dbs_check_dataset_spread(self, dataset_name):
        """Figure out the number of events in each run of this dataset.

        This is a more efficient way of doing this than calling
        dbs_resolve_number_of_events for each run.

        """

        self.logger.debug("Checking spread of dataset `%s'" % dataset_name)

        # DEBUG DEBUG DEBUG
        # If we get here DBS should have been set up already.
        assert not self.dbs_api is None
        # DEBUG DEBUG DEBUG end

        api = self.dbs_api
        dbs_query = "find run.number, site, file.name, file.numevents " \
                    "where dataset = %s " \
                    "and dataset.status = VALID" % \
                    dataset_name
        try:
            api_result = api.executeQuery(dbs_query)
        except DBSAPI.dbsApiException.DbsApiException:
            msg = "ERROR: Could not execute DBS query"
            self.logger.fatal(msg)
            raise Error(msg)

        handler = DBSXMLHandler(["run.number", "site", "file.name", "file.numevents"])
        parser = xml.sax.make_parser()
        parser.setContentHandler(handler)

        try:
            # OBSOLETE OBSOLETE OBSOLETE
##            class Handler(xml.sax.handler.ContentHandler):
##                def startElement(self, name, attrs):
##                    if name == "result":
##                        site_name = str(attrs["STORAGEELEMENT_SENAME"])
##                        # TODO TODO TODO
##                        # Ugly hack to get around cases like this:
##                        #   $ dbs search --query="find dataset, site, file.count where dataset=/RelValQCD_Pt_3000_3500/CMSSW_3_3_0_pre1-STARTUP31X_V4-v1/GEN-SIM-RECO"
##                        #   Using DBS instance at: http://cmsdbsprod.cern.ch/cms_dbs_prod_global/servlet/DBSServlet
##                        #   Processing ... \
##                        #   PATH    STORAGEELEMENT_SENAME   COUNT_FILES
##                        #   _________________________________________________________________________________
##                        #   /RelValQCD_Pt_3000_3500/CMSSW_3_3_0_pre1-STARTUP31X_V4-v1/GEN-SIM-RECO          1
##                        #   /RelValQCD_Pt_3000_3500/CMSSW_3_3_0_pre1-STARTUP31X_V4-v1/GEN-SIM-RECO  cmssrm.fnal.gov 12
##                        #   /RelValQCD_Pt_3000_3500/CMSSW_3_3_0_pre1-STARTUP31X_V4-v1/GEN-SIM-RECO  srm-cms.cern.ch 12
##                        if len(site_name) < 1:
##                            return
##                        # TODO TODO TODO end
##                        run_number = int(attrs["RUNS_RUNNUMBER"])
##                        file_name = str(attrs["FILES_LOGICALFILENAME"])
##                        nevents = int(attrs["FILES_NUMBEROFEVENTS"])
##                        # I know, this is a bit of a kludge.
##                        if not files_info.has_key(run_number):
##                            # New run.
##                            files_info[run_number] = {}
##                            files_info[run_number][file_name] = (nevents,
##                                                                 [site_name])
##                        elif not files_info[run_number].has_key(file_name):
##                            # New file for a known run.
##                            files_info[run_number][file_name] = (nevents,
##                                                                 [site_name])
##                        else:
##                            # New entry for a known file for a known run.
##                            # DEBUG DEBUG DEBUG
##                            # Each file should have the same number of
##                            # events independent of the site it's at.
##                            assert nevents == files_info[run_number][file_name][0]
##                            # DEBUG DEBUG DEBUG end
##                            files_info[run_number][file_name][1].append(site_name)
            # OBSOLETE OBSOLETE OBSOLETE end
            xml.sax.parseString(api_result, handler)
        except SAXParseException:
            msg = "ERROR: Could not parse DBS server output"
            self.logger.fatal(msg)
            raise Error(msg)

        # DEBUG DEBUG DEBUG
        assert(handler.check_results_validity()), "ERROR The DBSXMLHandler screwed something up!"
        # DEBUG DEBUG DEBUG end

        # Now reshuffle all results a bit so we can more easily use
        # them later on. (Remember that all arrays in the results
        # should have equal length.)
        files_info = {}
        for (index, site_name) in enumerate(handler.results["site"]):
            # Ugly hack to get around cases like this:
            #   $ dbs search --query="find dataset, site, file.count where dataset=/RelValQCD_Pt_3000_3500/CMSSW_3_3_0_pre1-STARTUP31X_V4-v1/GEN-SIM-RECO"
            #   Using DBS instance at: http://cmsdbsprod.cern.ch/cms_dbs_prod_global/servlet/DBSServlet
            #   Processing ... \
            #   PATH    STORAGEELEMENT_SENAME   COUNT_FILES
            #   _________________________________________________________________________________
            #   /RelValQCD_Pt_3000_3500/CMSSW_3_3_0_pre1-STARTUP31X_V4-v1/GEN-SIM-RECO          1
            #   /RelValQCD_Pt_3000_3500/CMSSW_3_3_0_pre1-STARTUP31X_V4-v1/GEN-SIM-RECO  cmssrm.fnal.gov 12
            #   /RelValQCD_Pt_3000_3500/CMSSW_3_3_0_pre1-STARTUP31X_V4-v1/GEN-SIM-RECO  srm-cms.cern.ch 12
            if len(site_name) < 1:
                continue
            run_number = int(handler.results["run.number"][index])
            file_name = handler.results["file.name"][index]
            nevents = int(handler.results["file.numevents"][index])

            # I know, this is a bit of a kludge.
            if run_number not in files_info:
                # New run.
                files_info[run_number] = {}
                files_info[run_number][file_name] = (nevents,
                                                     [site_name])
            elif file_name not in files_info[run_number]:
                # New file for a known run.
                files_info[run_number][file_name] = (nevents,
                                                     [site_name])
            else:
                # New entry for a known file for a known run.
                # DEBUG DEBUG DEBUG
                # Each file should have the same number of
                # events independent of the site it's at.
                assert nevents == files_info[run_number][file_name][0]
                # DEBUG DEBUG DEBUG end
                files_info[run_number][file_name][1].append(site_name)

        # Remove any information for files that are not available
        # anywhere. NOTE: After introducing the ugly hack above, this
        # is a bit redundant, but let's keep it for the moment.
        for run_number in files_info.keys():
            files_without_sites = [i for (i, j) in \
                                   files_info[run_number].items() \
                                   if len(j[1]) < 1]
            if len(files_without_sites) > 0:
                self.logger.warning("Removing %d file(s)" \
                                    " with empty site names" % \
                                    len(files_without_sites))
                for file_name in files_without_sites:
                    del files_info[run_number][file_name]
                    # files_info[run_number][file_name] = (files_info \
                    #                                     [run_number] \
                    #                                      [file_name][0], [])

        # And another bit of a kludge.
        num_events_catalog = {}
        for run_number in files_info.keys():
            site_names = list(set([j for i in files_info[run_number].values() for j in i[1]]))

            # NOTE: The term `mirrored' does not have the usual
            # meaning here. It basically means that we can apply
            # single-step harvesting.
            mirrored = None
            if len(site_names) > 1:
                # Now we somehow need to figure out if we're dealing
                # with a mirrored or a spread-out dataset. The rule we
                # use here is that we're dealing with a spread-out
                # dataset unless we can find at least one site
                # containing exactly the full list of files for this
                # dataset that DBS knows about. In that case we just
                # use only that site.
                all_file_names = files_info[run_number].keys()
                all_file_names = set(all_file_names)
                sites_with_complete_copies = []
                for site_name in site_names:
                    files_at_site = [i for (i, (j, k)) \
                                     in files_info[run_number].items() \
                                     if site_name in k]
                    files_at_site = set(files_at_site)
                    if files_at_site == all_file_names:
                        sites_with_complete_copies.append(site_name)
                if len(sites_with_complete_copies) < 1:
                    # This dataset/run is available at more than one
                    # site, but no one has a complete copy. So this is
                    # a spread-out sample.
                    mirrored = False
                else:
                    if len(sites_with_complete_copies) > 1:
                        # This sample is available (and complete) at
                        # more than one site. Definitely mirrored.
                        mirrored = True
                    else:
                        # This dataset/run is available at more than
                        # one site and at least one of them has a
                        # complete copy. Even if this is only a single
                        # site, let's call this `mirrored' and run the
                        # single-step harvesting.
                        mirrored = True

##                site_names_ref = set(files_info[run_number].values()[0][1])
##                for site_names_tmp in files_info[run_number].values()[1:]:
##                    if set(site_names_tmp[1]) != site_names_ref:
##                        mirrored = False
##                        break

                if mirrored:
                    self.logger.debug("    -> run appears to be `mirrored'")
                else:
                    self.logger.debug("    -> run appears to be spread-out")

                if mirrored and \
                       len(sites_with_complete_copies) != len(site_names):
                    # Remove any references to incomplete sites if we
                    # have at least one complete site (and if there
                    # are incomplete sites).
                    for (file_name, (i, sites)) in files_info[run_number].items():
                        complete_sites = [site for site in sites \
                                          if site in sites_with_complete_copies]
                        files_info[run_number][file_name] = (i, complete_sites)
                    site_names = sites_with_complete_copies

            self.logger.debug("  for run #%d:" % run_number)
            num_events_catalog[run_number] = {}
            num_events_catalog[run_number]["all_sites"] = sum([i[0] for i in files_info[run_number].values()])
            if len(site_names) < 1:
                self.logger.debug("    run is not available at any site")
                self.logger.debug("      (but should contain %d events" % \
                                  num_events_catalog[run_number]["all_sites"])
            else:
                self.logger.debug("    at all sites combined there are %d events" % \
                                  num_events_catalog[run_number]["all_sites"])
                for site_name in site_names:
                    num_events_catalog[run_number][site_name] = sum([i[0] for i in files_info[run_number].values() if site_name in i[1]])
                    self.logger.debug("    at site `%s' there are %d events" % \
                                      (site_name, num_events_catalog[run_number][site_name]))
            num_events_catalog[run_number]["mirrored"] = mirrored

        # End of dbs_check_dataset_spread.
        return num_events_catalog

    # Beginning of old version.
##    def dbs_check_dataset_num_events(self, dataset_name):
##        """Figure out the number of events in each run of this dataset.

##        This is a more efficient way of doing this than calling
##        dbs_resolve_number_of_events for each run.

##        # BUG BUG BUG
##        # This might very well not work at all for spread-out samples. (?)
##        # BUG BUG BUG end

##        """

##        # DEBUG DEBUG DEBUG
##        # If we get here DBS should have been set up already.
##        assert not self.dbs_api is None
##        # DEBUG DEBUG DEBUG end

##        api = self.dbs_api
##        dbs_query = "find run.number, file.name, file.numevents where dataset = %s " \
##                    "and dataset.status = VALID" % \
##                    dataset_name
##        try:
##            api_result = api.executeQuery(dbs_query)
##        except DbsApiException:
##            msg = "ERROR: Could not execute DBS query"
##            self.logger.fatal(msg)
##            raise Error(msg)

##        try:
##            files_info = {}
##            class Handler(xml.sax.handler.ContentHandler):
##                def startElement(self, name, attrs):
##                    if name == "result":
##                        run_number = int(attrs["RUNS_RUNNUMBER"])
##                        file_name = str(attrs["FILES_LOGICALFILENAME"])
##                        nevents = int(attrs["FILES_NUMBEROFEVENTS"])
##                        try:
##                            files_info[run_number][file_name] = nevents
##                        except KeyError:
##                            files_info[run_number] = {file_name: nevents}
##            xml.sax.parseString(api_result, Handler())
##        except SAXParseException:
##            msg = "ERROR: Could not parse DBS server output"
##            self.logger.fatal(msg)
##            raise Error(msg)

##        num_events_catalog = {}
##        for run_number in files_info.keys():
##            num_events_catalog[run_number] = sum(files_info[run_number].values())

##        # End of dbs_check_dataset_num_events.
##        return num_events_catalog
    # End of old version.

    ##########

    def build_dataset_list(self, input_method, input_name):
        """Build a list of all datasets to be processed.

        """

        dataset_names = []

        # It may be, but only for the list of datasets to ignore, that
        # the input method and name are None because nothing was
        # specified. In that case just an empty list is returned.
        if input_method is None:
            pass
        elif input_method == "dataset":
            # Input comes from a dataset name directly on the command
            # line. But, this can also contain wildcards so we need
            # DBS to translate it conclusively into a list of explicit
            # dataset names.
            self.logger.info("Asking DBS for dataset names")
            dataset_names = self.dbs_resolve_dataset_name(input_name)
        elif input_method == "datasetfile":
            # In this case a file containing a list of dataset names
            # is specified. Still, each line may contain wildcards so
            # this step also needs help from DBS.
            # NOTE: Lines starting with a `#' are ignored.
            self.logger.info("Reading input from list file `%s'" % \
                             input_name)
            try:
                listfile = open("/afs/cern.ch/cms/CAF/CMSCOMM/COMM_DQM/harvesting/bin/%s" %input_name, "r")
                print("open listfile")
                for dataset in listfile:
                    # Skip empty lines.
                    dataset_stripped = dataset.strip()
                    if len(dataset_stripped) < 1:
                        continue
                    # Skip lines starting with a `#'.
                    if dataset_stripped[0] != "#":
                        dataset_names.extend(self. \
                                             dbs_resolve_dataset_name(dataset_stripped))
                listfile.close()
            except IOError:
                msg = "ERROR: Could not open input list file `%s'" % \
                      input_name
                self.logger.fatal(msg)
                raise Error(msg)
        else:
            # DEBUG DEBUG DEBUG
            # We should never get here.
            assert False, "Unknown input method `%s'" % input_method
            # DEBUG DEBUG DEBUG end

        # Remove duplicates from the dataset list.
        # NOTE: There should not be any duplicates in any list coming
        # from DBS, but maybe the user provided a list file with less
        # care.
        # Store for later use.
        dataset_names = sorted(set(dataset_names))


        # End of build_dataset_list.
        return dataset_names

    ##########

    def build_dataset_use_list(self):
        """Build a list of datasets to process.

        """

        self.logger.info("Building list of datasets to consider...")

        input_method = self.input_method["datasets"]["use"]
        input_name = self.input_name["datasets"]["use"]
        dataset_names = self.build_dataset_list(input_method,
                                                input_name)
        self.datasets_to_use = dict(list(zip(dataset_names,
                                        [None] * len(dataset_names))))

        self.logger.info("  found %d dataset(s) to process:" % \
                         len(dataset_names))
        for dataset in dataset_names:
            self.logger.info("  `%s'" % dataset)

        # End of build_dataset_use_list.

    ##########

    def build_dataset_ignore_list(self):
        """Build a list of datasets to ignore.

        NOTE: We should always have a list of datasets to process, but
        it may be that we don't have a list of datasets to ignore.

        """

        self.logger.info("Building list of datasets to ignore...")

        input_method = self.input_method["datasets"]["ignore"]
        input_name = self.input_name["datasets"]["ignore"]
        dataset_names = self.build_dataset_list(input_method,
                                                input_name)
        self.datasets_to_ignore = dict(list(zip(dataset_names,
                                           [None] * len(dataset_names))))

        self.logger.info("  found %d dataset(s) to ignore:" % \
                         len(dataset_names))
        for dataset in dataset_names:
            self.logger.info("  `%s'" % dataset)

        # End of build_dataset_ignore_list.

    ##########

    def build_runs_list(self, input_method, input_name):

        runs = []

        # A list of runs (either to use or to ignore) is not
        # required. This protects against `empty cases.'
        if input_method is None:
            pass
        elif input_method == "runs":
            # A list of runs was specified directly from the command
            # line.
            self.logger.info("Reading list of runs from the " \
                             "command line")
            runs.extend([int(i.strip()) \
                         for i in input_name.split(",") \
                         if len(i.strip()) > 0])
        elif input_method == "runslistfile":
            # We were passed a file containing a list of runs.
            self.logger.info("Reading list of runs from file `%s'" % \
                             input_name)
            try:
                listfile = open(input_name, "r")
                for run in listfile:
                    # Skip empty lines.
                    run_stripped = run.strip()
                    if len(run_stripped) < 1:
                        continue
                    # Skip lines starting with a `#'.
                    if run_stripped[0] != "#":
                        runs.append(int(run_stripped))
                listfile.close()
            except IOError:
                msg = "ERROR: Could not open input list file `%s'" % \
                      input_name
                self.logger.fatal(msg)
                raise Error(msg)

        else:
            # DEBUG DEBUG DEBUG
            # We should never get here.
            assert False, "Unknown input method `%s'" % input_method
            # DEBUG DEBUG DEBUG end

        # Remove duplicates, sort and done.
        runs = list(set(runs))

        # End of build_runs_list().
        return runs

    ##########

    def build_runs_use_list(self):
        """Build a list of runs to process.

        """

        self.logger.info("Building list of runs to consider...")

        input_method = self.input_method["runs"]["use"]
        input_name = self.input_name["runs"]["use"]
        runs = self.build_runs_list(input_method, input_name)
        self.runs_to_use = dict(list(zip(runs, [None] * len(runs))))

        self.logger.info("  found %d run(s) to process:" % \
                         len(runs))
        if len(runs) > 0:
            self.logger.info("  %s" % ", ".join([str(i) for i in runs]))

        # End of build_runs_list().

    ##########

    def build_runs_ignore_list(self):
        """Build a list of runs to ignore.

        NOTE: We should always have a list of runs to process, but
        it may be that we don't have a list of runs to ignore.

        """

        self.logger.info("Building list of runs to ignore...")

        input_method = self.input_method["runs"]["ignore"]
        input_name = self.input_name["runs"]["ignore"]
        runs = self.build_runs_list(input_method, input_name)
        self.runs_to_ignore = dict(list(zip(runs, [None] * len(runs))))

        self.logger.info("  found %d run(s) to ignore:" % \
                         len(runs))
        if len(runs) > 0:
            self.logger.info("  %s" % ", ".join([str(i) for i in runs]))

        # End of build_runs_ignore_list().

    ##########

    def process_dataset_ignore_list(self):
        """Update the list of datasets taking into account the ones to
        ignore.

        Both lists have been generated before from DBS and both are
        assumed to be unique.

        NOTE: The advantage of creating the ignore list from DBS (in
        case a regexp is given) and matching that instead of directly
        matching the ignore criterion against the list of datasets (to
        consider) built from DBS is that in the former case we're sure
        that all regexps are treated exactly as DBS would have done
        without the cmsHarvester.

        NOTE: This only removes complete samples. Exclusion of single
        runs is done by the book keeping. So the assumption is that a
        user never wants to harvest just part (i.e. n out of N runs)
        of a sample.

        """

        self.logger.info("Processing list of datasets to ignore...")

        self.logger.debug("Before processing ignore list there are %d " \
                          "datasets in the list to be processed" % \
                          len(self.datasets_to_use))

        # Simple approach: just loop and search.
        dataset_names_filtered = copy.deepcopy(self.datasets_to_use)
        for dataset_name in self.datasets_to_use.keys():
            if dataset_name in self.datasets_to_ignore.keys():
                del dataset_names_filtered[dataset_name]

        self.logger.info("  --> Removed %d dataset(s)" % \
                         (len(self.datasets_to_use) -
                          len(dataset_names_filtered)))

        self.datasets_to_use = dataset_names_filtered

        self.logger.debug("After processing ignore list there are %d " \
                          "datasets in the list to be processed" % \
                          len(self.datasets_to_use))

    # End of process_dataset_ignore_list.

    ##########

    def process_runs_use_and_ignore_lists(self):

        self.logger.info("Processing list of runs to use and ignore...")

        # This basically adds all runs in a dataset to be processed,
        # except for any runs that are not specified in the `to use'
        # list and any runs that are specified in the `to ignore'
        # list.

        # NOTE: It is assumed that those lists make sense. The input
        # should be checked against e.g. overlapping `use' and
        # `ignore' lists.

        runs_to_use = self.runs_to_use
        runs_to_ignore = self.runs_to_ignore

        for dataset_name in self.datasets_to_use:
            runs_in_dataset = self.datasets_information[dataset_name]["runs"]

            # First some sanity checks.
            runs_to_use_tmp = []
            for run in runs_to_use:
                if not run in runs_in_dataset:
                    self.logger.warning("Dataset `%s' does not contain " \
                                        "requested run %d " \
                                        "--> ignoring `use' of this run" % \
                                        (dataset_name, run))
                else:
                    runs_to_use_tmp.append(run)

            if len(runs_to_use) > 0:
                runs = runs_to_use_tmp
                self.logger.info("Using %d out of %d runs " \
                                 "of dataset `%s'" % \
                                 (len(runs), len(runs_in_dataset),
                                  dataset_name))
            else:
                runs = runs_in_dataset

            if len(runs_to_ignore) > 0:
                runs_tmp = []
                for run in runs:
                    if not run in runs_to_ignore:
                        runs_tmp.append(run)
                self.logger.info("Ignoring %d out of %d runs " \
                                 "of dataset `%s'" % \
                                 (len(runs)- len(runs_tmp),
                                  len(runs_in_dataset),
                                  dataset_name))
                runs = runs_tmp

            if self.todofile != "YourToDofile.txt":
                runs_todo = []
                print("Reading runs from file /afs/cern.ch/cms/CAF/CMSCOMM/COMM_DQM/harvesting/%s" %self.todofile)
                cmd="grep %s /afs/cern.ch/cms/CAF/CMSCOMM/COMM_DQM/harvesting/%s | cut -f5 -d' '" %(dataset_name,self.todofile)
                (status, output)=commands.getstatusoutput(cmd)
                for run in runs:
                    run_str="%s" %run
                    if run_str in output:
                        runs_todo.append(run)
                self.logger.info("Using %d runs " \
                                 "of dataset `%s'" % \
                                 (len(runs_todo),
                                  dataset_name))
                runs=runs_todo

            Json_runs = []
            if self.Jsonfilename != "YourJSON.txt":
                good_runs = []
                self.Jsonlumi = True
                # We were passed a Jsonfile containing a dictionary of
                # run/lunisection-pairs
                self.logger.info("Reading runs and lumisections from file `%s'" % \
                                self.Jsonfilename)
                try:
                    Jsonfile = open(self.Jsonfilename, "r")
                    for names in Jsonfile:
                        dictNames= eval(str(names))
                        for key in dictNames:
                            intkey=int(key)
                            Json_runs.append(intkey)
                    Jsonfile.close()
                except IOError:
                    msg = "ERROR: Could not open Jsonfile `%s'" % \
                          input_name
                    self.logger.fatal(msg)
                    raise Error(msg)
                for run in runs:
                    if run in Json_runs:
                        good_runs.append(run)
                self.logger.info("Using %d runs " \
                                 "of dataset `%s'" % \
                                 (len(good_runs),
                                  dataset_name))
                runs=good_runs
            if (self.Jsonrunfilename != "YourJSON.txt") and (self.Jsonfilename == "YourJSON.txt"):
                good_runs = []
                # We were passed a Jsonfile containing a dictionary of
                # run/lunisection-pairs
                self.logger.info("Reading runs from file `%s'" % \
                                self.Jsonrunfilename)
                try:
                    Jsonfile = open(self.Jsonrunfilename, "r")
                    for names in Jsonfile:
                        dictNames= eval(str(names))
                        for key in dictNames:
                            intkey=int(key)
                            Json_runs.append(intkey)
                    Jsonfile.close()
                except IOError:
                    msg = "ERROR: Could not open Jsonfile `%s'" % \
                          input_name
                    self.logger.fatal(msg)
                    raise Error(msg)
                for run in runs:
                    if run in Json_runs: 
                        good_runs.append(run)
                self.logger.info("Using %d runs " \
                                 "of dataset `%s'" % \
                                 (len(good_runs),
                                  dataset_name))
                runs=good_runs

            self.datasets_to_use[dataset_name] = runs

        # End of process_runs_use_and_ignore_lists().

    ##########

    def singlify_datasets(self):
        """Remove all but the largest part of all datasets.

        This allows us to harvest at least part of these datasets
        using single-step harvesting until the two-step approach
        works.

        """

        # DEBUG DEBUG DEBUG
        assert self.harvesting_mode == "single-step-allow-partial"
        # DEBUG DEBUG DEBUG end

        for dataset_name in self.datasets_to_use:
            for run_number in self.datasets_information[dataset_name]["runs"]:
                max_events = max(self.datasets_information[dataset_name]["sites"][run_number].values())
                sites_with_max_events = [i[0] for i in self.datasets_information[dataset_name]["sites"][run_number].items() if i[1] == max_events]
                self.logger.warning("Singlifying dataset `%s', " \
                                    "run %d" % \
                                    (dataset_name, run_number))
                cmssw_version = self.datasets_information[dataset_name] \
                                ["cmssw_version"]
                selected_site = self.pick_a_site(sites_with_max_events,
                                                 cmssw_version)

                # Let's tell the user that we're manhandling this dataset.
                nevents_old = self.datasets_information[dataset_name]["num_events"][run_number]
                self.logger.warning("  --> " \
                                    "only harvesting partial statistics: " \
                                    "%d out of %d events (5.1%f%%) " \
                                    "at site `%s'" % \
                                    (max_events,
                                     nevents_old,
                                     100. * max_events / nevents_old,
                                     selected_site))
                self.logger.warning("!!! Please note that the number of " \
                                    "events in the output path name will " \
                                    "NOT reflect the actual statistics in " \
                                    "the harvested results !!!")

                # We found the site with the highest statistics and
                # the corresponding number of events. (CRAB gets upset
                # if we ask for more events than there are at a given
                # site.) Now update this information in our main
                # datasets_information variable.
                self.datasets_information[dataset_name]["sites"][run_number] = {selected_site: max_events}
                self.datasets_information[dataset_name]["num_events"][run_number] = max_events
                #self.datasets_information[dataset_name]["sites"][run_number] = [selected_site]

        # End of singlify_datasets.

    ##########

    def check_dataset_list(self):
        """Check list of dataset names for impossible ones.

        Two kinds of checks are done:
        - Checks for things that do not make sense. These lead to
          errors and skipped datasets.
        - Sanity checks. For these warnings are issued but the user is
          considered to be the authoritative expert.

        Checks performed:
        - The CMSSW version encoded in the dataset name should match
          self.cmssw_version. This is critical.
        - There should be some events in the dataset/run. This is
          critical in the sense that CRAB refuses to create jobs for
          zero events. And yes, this does happen in practice. E.g. the
          reprocessed CRAFT08 datasets contain runs with zero events.
        - A cursory check is performed to see if the harvesting type
          makes sense for the data type. This should prevent the user
          from inadvertently running RelVal for data.
        - It is not possible to run single-step harvesting jobs on
          samples that are not fully contained at a single site.
        - Each dataset/run has to be available at at least one site.

        """

        self.logger.info("Performing sanity checks on dataset list...")

        dataset_names_after_checks = copy.deepcopy(self.datasets_to_use)

        for dataset_name in self.datasets_to_use.keys():

            # Check CMSSW version.
            version_from_dataset = self.datasets_information[dataset_name] \
                                   ["cmssw_version"]
            if version_from_dataset != self.cmssw_version:
                msg = "  CMSSW version mismatch for dataset `%s' " \
                      "(%s vs. %s)" % \
                      (dataset_name,
                       self.cmssw_version, version_from_dataset)
                if self.force_running:
                    # Expert mode: just warn, then continue.
                    self.logger.warning("%s " \
                                        "--> `force mode' active: " \
                                        "run anyway" % msg)
                else:
                    del dataset_names_after_checks[dataset_name]
                    self.logger.warning("%s " \
                                        "--> skipping" % msg)
                    continue

            ###

            # Check that the harvesting type makes sense for the
            # sample. E.g. normally one would not run the DQMOffline
            # harvesting on Monte Carlo.
            # TODO TODO TODO
            # This should be further refined.
            suspicious = False
            datatype = self.datasets_information[dataset_name]["datatype"]
            if datatype == "data":
                # Normally only DQM harvesting is run on data.
                if self.harvesting_type != "DQMOffline":
                    suspicious = True
            elif datatype == "mc":
                if self.harvesting_type == "DQMOffline":
                    suspicious = True
            else:
                # Doh!
                assert False, "ERROR Impossible data type `%s' " \
                       "for dataset `%s'" % \
                       (datatype, dataset_name)
            if suspicious:
                msg = "  Normally one does not run `%s' harvesting " \
                      "on %s samples, are you sure?" % \
                      (self.harvesting_type, datatype)
                if self.force_running:
                    self.logger.warning("%s " \
                                        "--> `force mode' active: " \
                                        "run anyway" % msg)
                else:
                    del dataset_names_after_checks[dataset_name]
                    self.logger.warning("%s " \
                                        "--> skipping" % msg)
                    continue

            # TODO TODO TODO end

            ###

            # BUG BUG BUG
            # For the moment, due to a problem with DBS, I cannot
            # figure out the GlobalTag for data by myself. (For MC
            # it's no problem.) This means that unless a GlobalTag was
            # specified from the command line, we will have to skip
            # any data datasets.

            if datatype == "data":
                if self.globaltag is None:
                    msg = "For data datasets (like `%s') " \
                          "we need a GlobalTag" % \
                          dataset_name
                    del dataset_names_after_checks[dataset_name]
                    self.logger.warning("%s " \
                                        "--> skipping" % msg)
                    continue

            # BUG BUG BUG end

            ###

            # Check if the GlobalTag exists and (if we're using
            # reference histograms) if it's ready to be used with
            # reference histograms.
            globaltag = self.datasets_information[dataset_name]["globaltag"]
            if not globaltag in self.globaltag_check_cache:
                if self.check_globaltag(globaltag):
                    self.globaltag_check_cache.append(globaltag)
                else:
                    msg = "Something is wrong with GlobalTag `%s' " \
                          "used by dataset `%s'!" % \
                          (globaltag, dataset_name)
                    if self.use_ref_hists:
                        msg += "\n(Either it does not exist or it " \
                               "does not contain the required key to " \
                               "be used with reference histograms.)"
                    else:
                        msg += "\n(It probably just does not exist.)"
                    self.logger.fatal(msg)
                    raise Usage(msg)

            ###

            # Require that each run is available at least somewhere.
            runs_without_sites = [i for (i, j) in \
                                  self.datasets_information[dataset_name] \
                                  ["sites"].items() \
                                  if len(j) < 1 and \
                                  i in self.datasets_to_use[dataset_name]]
            if len(runs_without_sites) > 0:
                for run_without_sites in runs_without_sites:
                    try:
                        dataset_names_after_checks[dataset_name].remove(run_without_sites)
                    except KeyError:
                        pass
                self.logger.warning("  removed %d unavailable run(s) " \
                                    "from dataset `%s'" % \
                                    (len(runs_without_sites), dataset_name))
                self.logger.debug("    (%s)" % \
                                  ", ".join([str(i) for i in \
                                             runs_without_sites]))

            ###

            # Unless we're running two-step harvesting: only allow
            # samples located on a single site.
            if not self.harvesting_mode == "two-step":
                for run_number in self.datasets_to_use[dataset_name]:
                    # DEBUG DEBUG DEBUG
##                    if self.datasets_information[dataset_name]["num_events"][run_number] != 0:
##                        pdb.set_trace()
                    # DEBUG DEBUG DEBUG end
                    num_sites = len(self.datasets_information[dataset_name] \
                                ["sites"][run_number])
                    if num_sites > 1 and \
                           not self.datasets_information[dataset_name] \
                           ["mirrored"][run_number]:
                        # Cannot do this with a single-step job, not
                        # even in force mode. It just does not make
                        # sense.
                        msg = "  Dataset `%s', run %d is spread across more " \
                              "than one site.\n" \
                              "  Cannot run single-step harvesting on " \
                              "samples spread across multiple sites" % \
                              (dataset_name, run_number)
                        try:
                            dataset_names_after_checks[dataset_name].remove(run_number)
                        except KeyError:
                            pass
                        self.logger.warning("%s " \
                                            "--> skipping" % msg)

            ###

            # Require that the dataset/run is non-empty.
            # NOTE: To avoid reconsidering empty runs/datasets next
            # time around, we do include them in the book keeping.
            # BUG BUG BUG
            # This should sum only over the runs that we use!
            tmp = [j for (i, j) in self.datasets_information \
                   [dataset_name]["num_events"].items() \
                   if i in self.datasets_to_use[dataset_name]]
            num_events_dataset = sum(tmp)
            # BUG BUG BUG end
            if num_events_dataset < 1:
                msg = "  dataset `%s' is empty" % dataset_name
                del dataset_names_after_checks[dataset_name]
                self.logger.warning("%s " \
                                    "--> skipping" % msg)
                # Update the book keeping with all the runs in the dataset.
                # DEBUG DEBUG DEBUG
                #assert set([j for (i, j) in self.datasets_information \
                #            [dataset_name]["num_events"].items() \
                #            if i in self.datasets_to_use[dataset_name]]) == \
                #           set([0])
                # DEBUG DEBUG DEBUG end
                #self.book_keeping_information[dataset_name] = self.datasets_information \
                #                                              [dataset_name]["num_events"]
                continue

            tmp = [i for i in \
                   self.datasets_information[dataset_name] \
                   ["num_events"].items() if i[1] < 1]
            tmp = [i for i in tmp if i[0] in self.datasets_to_use[dataset_name]]
            empty_runs = dict(tmp)
            if len(empty_runs) > 0:
                for empty_run in empty_runs:
                    try:
                        dataset_names_after_checks[dataset_name].remove(empty_run)
                    except KeyError:
                        pass
                self.logger.info("  removed %d empty run(s) from dataset `%s'" % \
                                 (len(empty_runs), dataset_name))
                self.logger.debug("    (%s)" % \
                                  ", ".join([str(i) for i in empty_runs]))

        ###

        # If we emptied out a complete dataset, remove the whole
        # thing.
        dataset_names_after_checks_tmp = copy.deepcopy(dataset_names_after_checks)
        for (dataset_name, runs) in six.iteritems(dataset_names_after_checks):
            if len(runs) < 1:
                self.logger.warning("  Removing dataset without any runs " \
                                    "(left) `%s'" % \
                                    dataset_name)
                del dataset_names_after_checks_tmp[dataset_name]
        dataset_names_after_checks = dataset_names_after_checks_tmp

        ###

        self.logger.warning("  --> Removed %d dataset(s)" % \
                            (len(self.datasets_to_use) -
                             len(dataset_names_after_checks)))

        # Now store the modified version of the dataset list.
        self.datasets_to_use = dataset_names_after_checks

        # End of check_dataset_list.

    ##########

    def escape_dataset_name(self, dataset_name):
        """Escape a DBS dataset name.

        Escape a DBS dataset name such that it does not cause trouble
        with the file system. This means turning each `/' into `__',
        except for the first one which is just removed.

        """

        escaped_dataset_name = dataset_name
        escaped_dataset_name = escaped_dataset_name.strip("/")
        escaped_dataset_name = escaped_dataset_name.replace("/", "__")

        return escaped_dataset_name

    ##########

    # BUG BUG BUG
    # This is a bit of a redundant method, isn't it?
    def create_config_file_name(self, dataset_name, run_number):
        """Generate the name of the configuration file to be run by
        CRAB.

        Depending on the harvesting mode (single-step or two-step)
        this is the name of the real harvesting configuration or the
        name of the first-step ME summary extraction configuration.

        """

        if self.harvesting_mode == "single-step":
            config_file_name = self.create_harvesting_config_file_name(dataset_name)
        elif self.harvesting_mode == "single-step-allow-partial":
            config_file_name = self.create_harvesting_config_file_name(dataset_name)
##            # Only add the alarming piece to the file name if this is
##            # a spread-out dataset.
##            pdb.set_trace()
##            if self.datasets_information[dataset_name] \
##                   ["mirrored"][run_number] == False:
##                config_file_name = config_file_name.replace(".py", "_partial.py")
        elif self.harvesting_mode == "two-step":
            config_file_name = self.create_me_summary_config_file_name(dataset_name)
        else:
            assert False, "ERROR Unknown harvesting mode `%s'" % \
                   self.harvesting_mode

        # End of create_config_file_name.
        return config_file_name
    # BUG BUG BUG end

    ##########

    def create_harvesting_config_file_name(self, dataset_name):
        "Generate the name to be used for the harvesting config file."

        file_name_base = "harvesting.py"
        dataset_name_escaped = self.escape_dataset_name(dataset_name)
        config_file_name = file_name_base.replace(".py",
                                                  "_%s.py" % \
                                                  dataset_name_escaped)

        # End of create_harvesting_config_file_name.
        return config_file_name

    ##########

    def create_me_summary_config_file_name(self, dataset_name):
        "Generate the name of the ME summary extraction config file."

        file_name_base = "me_extraction.py"
        dataset_name_escaped = self.escape_dataset_name(dataset_name)
        config_file_name = file_name_base.replace(".py",
                                                  "_%s.py" % \
                                                  dataset_name_escaped)

        # End of create_me_summary_config_file_name.
        return config_file_name

    ##########

    def create_output_file_name(self, dataset_name, run_number=None):
        """Create the name of the output file name to be used.

        This is the name of the output file of the `first step'. In
        the case of single-step harvesting this is already the final
        harvesting output ROOT file. In the case of two-step
        harvesting it is the name of the intermediary ME summary
        file.

        """

        # BUG BUG BUG
        # This method has become a bit of a mess. Originally it was
        # nice to have one entry point for both single- and two-step
        # output file names. However, now the former needs the run
        # number, while the latter does not even know about run
        # numbers. This should be fixed up a bit.
        # BUG BUG BUG end

        if self.harvesting_mode == "single-step":
            # DEBUG DEBUG DEBUG
            assert not run_number is None
            # DEBUG DEBUG DEBUG end
            output_file_name = self.create_harvesting_output_file_name(dataset_name, run_number)
        elif self.harvesting_mode == "single-step-allow-partial":
            # DEBUG DEBUG DEBUG
            assert not run_number is None
            # DEBUG DEBUG DEBUG end
            output_file_name = self.create_harvesting_output_file_name(dataset_name, run_number)
        elif self.harvesting_mode == "two-step":
            # DEBUG DEBUG DEBUG
            assert run_number is None
            # DEBUG DEBUG DEBUG end
            output_file_name = self.create_me_summary_output_file_name(dataset_name)
        else:
            # This should not be possible, but hey...
            assert False, "ERROR Unknown harvesting mode `%s'" % \
                   self.harvesting_mode

        # End of create_harvesting_output_file_name.
        return output_file_name

    ##########

    def create_harvesting_output_file_name(self, dataset_name, run_number):
        """Generate the name to be used for the harvesting output file.

        This harvesting output file is the _final_ ROOT output file
        containing the harvesting results. In case of two-step
        harvesting there is an intermediate ME output file as well.

        """

        dataset_name_escaped = self.escape_dataset_name(dataset_name)

        # Hmmm, looking at the code for the DQMFileSaver this might
        # actually be the place where the first part of this file
        # naming scheme comes from.
        # NOTE: It looks like the `V0001' comes from the DQM
        # version. This is something that cannot be looked up from
        # here, so let's hope it does not change too often.
        output_file_name = "DQM_V0001_R%09d__%s.root" % \
                           (run_number, dataset_name_escaped)
        if self.harvesting_mode.find("partial") > -1:
            # Only add the alarming piece to the file name if this is
            # a spread-out dataset.
            if self.datasets_information[dataset_name] \
                   ["mirrored"][run_number] == False:
                output_file_name = output_file_name.replace(".root", \
                                                            "_partial.root")

        # End of create_harvesting_output_file_name.
        return output_file_name

    ##########

    def create_me_summary_output_file_name(self, dataset_name):
        """Generate the name of the intermediate ME file name to be
        used in two-step harvesting.

        """

        dataset_name_escaped = self.escape_dataset_name(dataset_name)
        output_file_name = "me_summary_%s.root" % \
                           dataset_name_escaped

        # End of create_me_summary_output_file_name.
        return output_file_name

    ##########

    def create_multicrab_block_name(self, dataset_name, run_number, index):
        """Create the block name to use for this dataset/run number.

        This is what appears in the brackets `[]' in multicrab.cfg. It
        is used as the name of the job and to create output
        directories.

        """

        dataset_name_escaped = self.escape_dataset_name(dataset_name)
        block_name = "%s_%09d_%s" % (dataset_name_escaped, run_number, index)

        # End of create_multicrab_block_name.
        return block_name

    ##########

    def create_crab_config(self):
        """Create a CRAB configuration for a given job.

        NOTE: This is _not_ a complete (as in: submittable) CRAB
        configuration. It is used to store the common settings for the
        multicrab configuration.

        NOTE: Only CERN CASTOR area (/castor/cern.ch/) is supported.

        NOTE: According to CRAB, you `Must define exactly two of
        total_number_of_events, events_per_job, or
        number_of_jobs.'. For single-step harvesting we force one job,
        for the rest we don't really care.

        # BUG BUG BUG
        # With the current version of CRAB (2.6.1), in which Daniele
        # fixed the behaviour of no_block_boundary for me, one _has to
        # specify_ the total_number_of_events and one single site in
        # the se_white_list.
        # BUG BUG BUG end

        """

        tmp = []

        # This is the stuff we will need to fill in.
        castor_prefix = self.castor_prefix

        tmp.append(self.config_file_header())
        tmp.append("")

        ## CRAB
        ##------
        tmp.append("[CRAB]")
        tmp.append("jobtype = cmssw")
        tmp.append("")

        ## GRID
        ##------
        tmp.append("[GRID]")
        tmp.append("virtual_organization=cms")
        tmp.append("")

        ## USER
        ##------
        tmp.append("[USER]")
        tmp.append("copy_data = 1")
        tmp.append("")

        ## CMSSW
        ##-------
        tmp.append("[CMSSW]")
        tmp.append("# This reveals data hosted on T1 sites,")
        tmp.append("# which is normally hidden by CRAB.")
        tmp.append("show_prod = 1")
        tmp.append("number_of_jobs = 1")
        if self.Jsonlumi == True:
            tmp.append("lumi_mask = %s" % self.Jsonfilename)
            tmp.append("total_number_of_lumis = -1")
        else:
            if self.harvesting_type == "DQMOffline":
                tmp.append("total_number_of_lumis = -1")
            else:
                tmp.append("total_number_of_events = -1")
        if self.harvesting_mode.find("single-step") > -1:
            tmp.append("# Force everything to run in one job.")
            tmp.append("no_block_boundary = 1")
        tmp.append("")

        ## CAF
        ##-----
        tmp.append("[CAF]")

        crab_config = "\n".join(tmp)

        # End of create_crab_config.
        return crab_config

    ##########

    def create_multicrab_config(self):
        """Create a multicrab.cfg file for all samples.

        This creates the contents for a multicrab.cfg file that uses
        the crab.cfg file (generated elsewhere) for the basic settings
        and contains blocks for each run of each dataset.

        # BUG BUG BUG
        # The fact that it's necessary to specify the se_white_list
        # and the total_number_of_events is due to our use of CRAB
        # version 2.6.1. This should no longer be necessary in the
        # future.
        # BUG BUG BUG end

        """

        cmd="who i am | cut -f1 -d' '"
        (status, output)=commands.getstatusoutput(cmd)
        UserName = output

        if self.caf_access == True:
            print("Extracting %s as user name" %UserName) 

        number_max_sites = self.nr_max_sites + 1

        multicrab_config_lines = []
        multicrab_config_lines.append(self.config_file_header())
        multicrab_config_lines.append("")
        multicrab_config_lines.append("[MULTICRAB]")
        multicrab_config_lines.append("cfg = crab.cfg")
        multicrab_config_lines.append("")

        dataset_names = sorted(self.datasets_to_use.keys())

        for dataset_name in dataset_names:
            runs = self.datasets_to_use[dataset_name]
            dataset_name_escaped = self.escape_dataset_name(dataset_name)
            castor_prefix = self.castor_prefix

            for run in runs:

                # CASTOR output dir.
                castor_dir = self.datasets_information[dataset_name] \
                                 ["castor_path"][run]

                cmd = "rfdir %s" % castor_dir
                (status, output) = commands.getstatusoutput(cmd)

                if len(output) <= 0:

                    # DEBUG DEBUG DEBUG
                    # We should only get here if we're treating a
                    # dataset/run that is fully contained at a single
                    # site.
                    assert (len(self.datasets_information[dataset_name] \
                            ["sites"][run]) == 1) or \
                            self.datasets_information[dataset_name]["mirrored"]
                    # DEBUG DEBUG DEBUG end

                    site_names = self.datasets_information[dataset_name] \
                             ["sites"][run].keys()

                    for i in range(1, number_max_sites, 1):
                        if len(site_names) > 0: 
                            index = "site_%02d" % (i)

                            config_file_name = self. \
                                       create_config_file_name(dataset_name, run)
                            output_file_name = self. \
                                       create_output_file_name(dataset_name, run)


                            # If we're looking at a mirrored dataset we just pick
                            # one of the sites. Otherwise there is nothing to
                            # choose.

                            # Loop variable
                            loop = 0

                            if len(site_names) > 1:
                                cmssw_version = self.datasets_information[dataset_name] \
                                            ["cmssw_version"]
                                self.logger.info("Picking site for mirrored dataset " \
                                             "`%s', run %d" % \
                                             (dataset_name, run))
                                site_name = self.pick_a_site(site_names, cmssw_version)
                                if site_name in site_names:
                                    site_names.remove(site_name)

                            else:
                                site_name = site_names[0]
                                site_names.remove(site_name)

                            if site_name is self.no_matching_site_found_str:
                                if loop < 1:
                                    break

                            nevents = self.datasets_information[dataset_name]["num_events"][run]

                            # The block name.
                            multicrab_block_name = self.create_multicrab_block_name( \
                                dataset_name, run, index)
                            multicrab_config_lines.append("[%s]" % \
                                                      multicrab_block_name)

                            ## CRAB
                            ##------
                            if site_name == "caf.cern.ch":
                                multicrab_config_lines.append("CRAB.use_server=0")
                                multicrab_config_lines.append("CRAB.scheduler=caf")
                            else:
                                multicrab_config_lines.append("scheduler = glite")

                            ## GRID
                            ##------
                            if site_name == "caf.cern.ch":
                                pass
                            else:
                                multicrab_config_lines.append("GRID.se_white_list = %s" % \
                                                      site_name)
                                multicrab_config_lines.append("# This removes the default blacklisting of T1 sites.")
                                multicrab_config_lines.append("GRID.remove_default_blacklist = 1")
                                multicrab_config_lines.append("GRID.rb = CERN")
                                if not self.non_t1access:
                                    multicrab_config_lines.append("GRID.role = t1access")

                            ## USER
                            ##------

                            castor_dir = castor_dir.replace(castor_prefix, "")
                            multicrab_config_lines.append("USER.storage_element=srm-cms.cern.ch")
                            multicrab_config_lines.append("USER.user_remote_dir = %s" % \
                                                      castor_dir)
                            multicrab_config_lines.append("USER.check_user_remote_dir=0")

                            if site_name == "caf.cern.ch":
                                multicrab_config_lines.append("USER.storage_path=%s" % castor_prefix)
                                #multicrab_config_lines.append("USER.storage_element=T2_CH_CAF")
                                #castor_dir = castor_dir.replace("/cms/store/caf/user/%s" %UserName, "")
                                #multicrab_config_lines.append("USER.user_remote_dir = %s" % \
                                #             castor_dir)
                            else:
                                multicrab_config_lines.append("USER.storage_path=/srm/managerv2?SFN=%s" % castor_prefix)
                                #multicrab_config_lines.append("USER.user_remote_dir = %s" % \
                                #             castor_dir)
                                #multicrab_config_lines.append("USER.storage_element=srm-cms.cern.ch")

                            ## CMSSW
                            ##-------
                            multicrab_config_lines.append("CMSSW.pset = %s" % \
                                                       config_file_name)
                            multicrab_config_lines.append("CMSSW.datasetpath = %s" % \
                                                      dataset_name)
                            multicrab_config_lines.append("CMSSW.runselection = %d" % \
                                                  run)

                            if self.Jsonlumi == True:
                                pass
                            else:
                                if self.harvesting_type == "DQMOffline":
                                    pass
                                else:
                                    multicrab_config_lines.append("CMSSW.total_number_of_events = %d" % \
                                                                  nevents)
                            # The output file name.
                            multicrab_config_lines.append("CMSSW.output_file = %s" % \
                                                      output_file_name)

                            ## CAF
                            ##-----
                            if site_name == "caf.cern.ch":
                                multicrab_config_lines.append("CAF.queue=cmscaf1nd")


                            # End of block.
                            multicrab_config_lines.append("")

                            loop = loop + 1

                            self.all_sites_found = True

        multicrab_config = "\n".join(multicrab_config_lines)

        # End of create_multicrab_config.
        return multicrab_config

    ##########

    def check_globaltag(self, globaltag=None):
        """Check if globaltag exists.

        Check if globaltag exists as GlobalTag in the database given
        by self.frontier_connection_name['globaltag']. If globaltag is
        None, self.globaltag is used instead.

        If we're going to use reference histograms this method also
        checks for the existence of the required key in the GlobalTag.

        """

        if globaltag is None:
            globaltag = self.globaltag

        # All GlobalTags should end in `::All', right?
        if globaltag.endswith("::All"):
            globaltag = globaltag[:-5]

        connect_name = self.frontier_connection_name["globaltag"]
        # BUG BUG BUG
        # There is a bug in cmscond_tagtree_list: some magic is
        # missing from the implementation requiring one to specify
        # explicitly the name of the squid to connect to. Since the
        # cmsHarvester can only be run from the CERN network anyway,
        # cmsfrontier:8000 is hard-coded in here. Not nice but it
        # works.
        connect_name = connect_name.replace("frontier://",
                                            "frontier://cmsfrontier:8000/")
        # BUG BUG BUG end
        connect_name += self.db_account_name_cms_cond_globaltag()

        tag_exists = self.check_globaltag_exists(globaltag, connect_name)

        #----------

        tag_contains_ref_hist_key = False
        if self.use_ref_hists and tag_exists:
            # Check for the key required to use reference histograms.
            tag_contains_ref_hist_key = self.check_globaltag_contains_ref_hist_key(globaltag, connect_name)

        #----------

        if self.use_ref_hists:
            ret_val = tag_exists and tag_contains_ref_hist_key
        else:
            ret_val = tag_exists

        #----------

        # End of check_globaltag.
        return ret_val

    ##########

    def check_globaltag_exists(self, globaltag, connect_name):
        """Check if globaltag exists.

        """

        self.logger.info("Checking existence of GlobalTag `%s'" % \
                         globaltag)
        self.logger.debug("  (Using database connection `%s')" % \
                          connect_name)

        cmd = "cmscond_tagtree_list -c %s -T %s" % \
              (connect_name, globaltag)
        (status, output) = commands.getstatusoutput(cmd)
        if status != 0 or \
               output.find("error") > -1:
            msg = "Could not check existence of GlobalTag `%s' in `%s'" % \
                  (globaltag, connect_name)
            if output.find(".ALL_TABLES not found") > -1:
                msg = "%s\n" \
                      "Missing database account `%s'" % \
                      (msg, output.split(".ALL_TABLES")[0].split()[-1])
            self.logger.fatal(msg)
            self.logger.debug("Command used:")
            self.logger.debug("  %s" % cmd)
            self.logger.debug("Output received:")
            self.logger.debug(output)
            raise Error(msg)
        if output.find("does not exist") > -1:
            self.logger.debug("GlobalTag `%s' does not exist in `%s':" % \
                              (globaltag, connect_name))
            self.logger.debug("Output received:")
            self.logger.debug(output)
            tag_exists = False
        else:
            tag_exists = True
        self.logger.info("  GlobalTag exists? -> %s" % tag_exists)

        # End of check_globaltag_exists.
        return tag_exists

    ##########

    def check_globaltag_contains_ref_hist_key(self, globaltag, connect_name):
        """Check if globaltag contains the required RefHistos key.

        """

        # Check for the key required to use reference histograms.
        tag_contains_key = None
        ref_hist_key = "RefHistos"
        self.logger.info("Checking existence of reference " \
                         "histogram key `%s' in GlobalTag `%s'" % \
                         (ref_hist_key, globaltag))
        self.logger.debug("  (Using database connection `%s')" % \
                              connect_name)
        cmd = "cmscond_tagtree_list -c %s -T %s -n %s" % \
              (connect_name, globaltag, ref_hist_key)
        (status, output) = commands.getstatusoutput(cmd)
        if status != 0 or \
               output.find("error") > -1:
            msg = "Could not check existence of key `%s'" % \
                  (ref_hist_key, connect_name)
            self.logger.fatal(msg)
            self.logger.debug("Command used:")
            self.logger.debug("  %s" % cmd)
            self.logger.debug("Output received:")
            self.logger.debug("  %s" % output)
            raise Error(msg)
        if len(output) < 1:
            self.logger.debug("Required key for use of reference " \
                              "histograms `%s' does not exist " \
                              "in GlobalTag `%s':" % \
                              (ref_hist_key, globaltag))
            self.logger.debug("Output received:")
            self.logger.debug(output)
            tag_contains_key = False
        else:
            tag_contains_key = True

        self.logger.info("  GlobalTag contains `%s' key? -> %s" % \
                         (ref_hist_key, tag_contains_key))

        # End of check_globaltag_contains_ref_hist_key.
        return tag_contains_key

    ##########

    def check_ref_hist_tag(self, tag_name):
        """Check the existence of tag_name in database connect_name.

        Check if tag_name exists as a reference histogram tag in the
        database given by self.frontier_connection_name['refhists'].

        """

        connect_name = self.frontier_connection_name["refhists"]
        connect_name += self.db_account_name_cms_cond_dqm_summary()

        self.logger.debug("Checking existence of reference " \
                          "histogram tag `%s'" % \
                          tag_name)
        self.logger.debug("  (Using database connection `%s')" % \
                          connect_name)

        cmd = "cmscond_list_iov -c %s" % \
              connect_name
        (status, output) = commands.getstatusoutput(cmd)
        if status != 0:
            msg = "Could not check existence of tag `%s' in `%s'" % \
                  (tag_name, connect_name)
            self.logger.fatal(msg)
            self.logger.debug("Command used:")
            self.logger.debug("  %s" % cmd)
            self.logger.debug("Output received:")
            self.logger.debug(output)
            raise Error(msg)
        if not tag_name in output.split():
            self.logger.debug("Reference histogram tag `%s' " \
                              "does not exist in `%s'" % \
                              (tag_name, connect_name))
            self.logger.debug(" Existing tags: `%s'" % \
                              "', `".join(output.split()))
            tag_exists = False
        else:
            tag_exists = True
        self.logger.debug("  Reference histogram tag exists? " \
                          "-> %s" % tag_exists)

        # End of check_ref_hist_tag.
        return tag_exists

    ##########

    def create_es_prefer_snippet(self, dataset_name):
        """Build the es_prefer snippet for the reference histograms.

        The building of the snippet is wrapped in some care-taking
        code that figures out the name of the reference histogram set
        and makes sure the corresponding tag exists.

        """

        # Figure out the name of the reference histograms tag.
        # NOTE: The existence of these tags has already been checked.
        ref_hist_tag_name = self.ref_hist_mappings[dataset_name]

        connect_name = self.frontier_connection_name["refhists"]
        connect_name += self.db_account_name_cms_cond_dqm_summary()
        record_name = "DQMReferenceHistogramRootFileRcd"

        # Build up the code snippet.
        code_lines = []
        code_lines.append("from CondCore.DBCommon.CondDBSetup_cfi import *")
        code_lines.append("process.ref_hist_source = cms.ESSource(\"PoolDBESSource\", CondDBSetup,")
        code_lines.append("                                       connect = cms.string(\"%s\")," % connect_name)
        code_lines.append("                                       toGet = cms.VPSet(cms.PSet(record = cms.string(\"%s\")," % record_name)
        code_lines.append("                                                                  tag = cms.string(\"%s\"))," % ref_hist_tag_name)
        code_lines.append("                                                         )")
        code_lines.append("                                       )")
        code_lines.append("process.es_prefer_ref_hist_source = cms.ESPrefer(\"PoolDBESSource\", \"ref_hist_source\")")

        snippet = "\n".join(code_lines)

        # End of create_es_prefer_snippet.
        return snippet

    ##########

    def create_harvesting_config(self, dataset_name):
        """Create the Python harvesting configuration for harvesting.

        The basic configuration is created by
        Configuration.PyReleaseValidation.ConfigBuilder. (This mimics
        what cmsDriver.py does.) After that we add some specials
        ourselves.

        NOTE: On one hand it may not be nice to circumvent
        cmsDriver.py, on the other hand cmsDriver.py does not really
        do anything itself. All the real work is done by the
        ConfigBuilder so there is not much risk that we miss out on
        essential developments of cmsDriver in the future.

        """

        # Setup some options needed by the ConfigBuilder.
        config_options = defaultOptions

        # These are fixed for all kinds of harvesting jobs. Some of
        # them are not needed for the harvesting config, but to keep
        # the ConfigBuilder happy.
        config_options.name = "harvesting"
        config_options.scenario = "pp"
        config_options.number = 1
        config_options.arguments = self.ident_string()
        config_options.evt_type = config_options.name
        config_options.customisation_file = None
        config_options.filein = "dummy_value"
        config_options.filetype = "EDM"
        # This seems to be new in CMSSW 3.3.X, no clue what it does.
        config_options.gflash = "dummy_value"
        # This seems to be new in CMSSW 3.3.0.pre6, no clue what it
        # does.
        #config_options.himix = "dummy_value"
        config_options.dbsquery = ""

        ###

        # These options depend on the type of harvesting we're doing
        # and are stored in self.harvesting_info.

        config_options.step = "HARVESTING:%s" % \
                              self.harvesting_info[self.harvesting_type] \
                              ["step_string"]
        config_options.beamspot = self.harvesting_info[self.harvesting_type] \
                                  ["beamspot"]
        config_options.eventcontent = self.harvesting_info \
                                      [self.harvesting_type] \
                                      ["eventcontent"]
        config_options.harvesting = self.harvesting_info \
                                    [self.harvesting_type] \
                                    ["harvesting"]

        ###

        # This one is required (see also above) for each dataset.

        datatype = self.datasets_information[dataset_name]["datatype"]
        config_options.isMC = (datatype.lower() == "mc")
        config_options.isData = (datatype.lower() == "data")
        globaltag = self.datasets_information[dataset_name]["globaltag"]

        config_options.conditions = self.format_conditions_string(globaltag)

        ###

        if "with_input" in getargspec(ConfigBuilder.__init__)[0]:
            # This is the case for 3.3.X.
            config_builder = ConfigBuilder(config_options, with_input=True)
        else:
            # This is the case in older CMSSW versions.
            config_builder = ConfigBuilder(config_options)
        config_builder.prepare(True)
        config_contents = config_builder.pythonCfgCode

        ###

        # Add our signature to the top of the configuration.  and add
        # some markers to the head and the tail of the Python code
        # generated by the ConfigBuilder.
        marker_lines = []
        sep = "#" * 30
        marker_lines.append(sep)
        marker_lines.append("# Code between these markers was generated by")
        marker_lines.append("# Configuration.PyReleaseValidation." \
                            "ConfigBuilder")

        marker_lines.append(sep)
        marker = "\n".join(marker_lines)

        tmp = [self.config_file_header()]
        tmp.append("")
        tmp.append(marker)
        tmp.append("")
        tmp.append(config_contents)
        tmp.append("")
        tmp.append(marker)
        tmp.append("")
        config_contents = "\n".join(tmp)

        ###

        # Now we add some stuff of our own.
        customisations = [""]

        customisations.append("# Now follow some customisations")
        customisations.append("")
        connect_name = self.frontier_connection_name["globaltag"]
        connect_name += self.db_account_name_cms_cond_globaltag()
        customisations.append("process.GlobalTag.connect = \"%s\"" % \
                              connect_name)


        if self.saveByLumiSection == True:
            customisations.append("process.dqmSaver.saveByLumiSection = 1")
        

        customisations.append("")

        # About the reference histograms... For data there is only one
        # set of references and those are picked up automatically
        # based on the GlobalTag. For MC we have to do some more work
        # since the reference histograms to be used depend on the MC
        # sample at hand. In this case we glue in an es_prefer snippet
        # to pick up the references. We do this only for RelVals since
        # for MC there are no meaningful references so far.

        # NOTE: Due to the lack of meaningful references for
        # MC samples reference histograms are explicitly
        # switched off in this case.

        use_es_prefer = (self.harvesting_type == "RelVal")
        use_refs = use_es_prefer or \
                   (not self.harvesting_type == "MC")
        # Allow global override.
        use_refs = use_refs and self.use_ref_hists

        if not use_refs:
            # Disable reference histograms explicitly. The histograms
            # are loaded by the dqmRefHistoRootFileGetter
            # EDAnalyzer. This analyzer can be run from several
            # sequences. Here we remove it from each sequence that
            # exists.
            customisations.append("print \"Not using reference histograms\"")
            customisations.append("if hasattr(process, \"dqmRefHistoRootFileGetter\"):")
            customisations.append("    for (sequence_name, sequence) in six.iteritems(process.sequences):")
            customisations.append("        if sequence.remove(process.dqmRefHistoRootFileGetter):")
            customisations.append("            print \"Removed process.dqmRefHistoRootFileGetter from sequence `%s'\" % \\")
            customisations.append("                  sequence_name")
            customisations.append("process.dqmSaver.referenceHandling = \"skip\"")
        else:
            # This makes sure all reference histograms are saved to
            # the output ROOT file.
            customisations.append("process.dqmSaver.referenceHandling = \"all\"")
            if use_es_prefer:
                es_prefer_snippet = self.create_es_prefer_snippet(dataset_name)
                customisations.append(es_prefer_snippet)

        # Make sure we get the `workflow' correct. As far as I can see
        # this is only important for the output file name.
        workflow_name = dataset_name
        if self.harvesting_mode == "single-step-allow-partial":
            workflow_name += "_partial"
        customisations.append("process.dqmSaver.workflow = \"%s\"" % \
                              workflow_name)

        # BUG BUG BUG
        # This still does not work. The current two-step harvesting
        # efforts are on hold waiting for the solution to come from
        # elsewhere. (In this case the elsewhere is Daniele Spiga.)

##        # In case this file is the second step (the real harvesting
##        # step) of the two-step harvesting we have to tell it to use
##        # our local files.
##        if self.harvesting_mode == "two-step":
##            castor_dir = self.datasets_information[dataset_name] \
##                         ["castor_path"][run]
##            customisations.append("")
##            customisations.append("# This is the second step (the real")
##            customisations.append("# harvesting step) of a two-step")
##            customisations.append("# harvesting procedure.")
##            # BUG BUG BUG
##            # To be removed in production version.
##            customisations.append("import pdb")
##            # BUG BUG BUG end
##            customisations.append("import commands")
##            customisations.append("import os")
##            customisations.append("castor_dir = \"%s\"" % castor_dir)
##            customisations.append("cmd = \"rfdir %s\" % castor_dir")
##            customisations.append("(status, output) = commands.getstatusoutput(cmd)")
##            customisations.append("if status != 0:")
##            customisations.append("    print \"ERROR\"")
##            customisations.append("    raise Exception, \"ERROR\"")
##            customisations.append("file_names = [os.path.join(\"rfio:%s\" % path, i) for i in output.split() if i.startswith(\"EDM_summary\") and i.endswith(\".root\")]")
##            #customisations.append("pdb.set_trace()")
##            customisations.append("process.source.fileNames = cms.untracked.vstring(*file_names)")
##            customisations.append("")

        # BUG BUG BUG end

        config_contents = config_contents + "\n".join(customisations)

        ###

        # End of create_harvesting_config.
        return config_contents

##    ##########

##    def create_harvesting_config_two_step(self, dataset_name):
##        """Create the Python harvesting configuration for two-step
##        harvesting.

##        """

##        # BUG BUG BUG
##        config_contents = self.create_harvesting_config_single_step(dataset_name)
##        # BUG BUG BUG end

##        # End of create_harvesting_config_two_step.
##        return config_contents

    ##########

    def create_me_extraction_config(self, dataset_name):
        """

        """

        # Big chunk of hard-coded Python. Not such a big deal since
        # this does not do much and is not likely to break.
        tmp = []
        tmp.append(self.config_file_header())
        tmp.append("")
        tmp.append("import FWCore.ParameterSet.Config as cms")
        tmp.append("")
        tmp.append("process = cms.Process(\"ME2EDM\")")
        tmp.append("")
        tmp.append("# Import of standard configurations")
        tmp.append("process.load(\"Configuration/EventContent/EventContent_cff\")")
        tmp.append("")
        tmp.append("# We don't really process any events, just keep this set to one to")
        tmp.append("# make sure things work.")
        tmp.append("process.maxEvents = cms.untracked.PSet(")
        tmp.append("    input = cms.untracked.int32(1)")
        tmp.append("    )")
        tmp.append("")
        tmp.append("process.options = cms.untracked.PSet(")
        tmp.append("    Rethrow = cms.untracked.vstring(\"ProductNotFound\")")
        tmp.append("    )")
        tmp.append("")
        tmp.append("process.source = cms.Source(\"PoolSource\",")
        tmp.append("                            processingMode = \\")
        tmp.append("                            cms.untracked.string(\"RunsAndLumis\"),")
        tmp.append("                            fileNames = \\")
        tmp.append("                            cms.untracked.vstring(\"no_file_specified\")")
        tmp.append("                            )")
        tmp.append("")
        tmp.append("# Output definition: drop everything except for the monitoring.")
        tmp.append("process.output = cms.OutputModule(")
        tmp.append("    \"PoolOutputModule\",")
        tmp.append("    outputCommands = \\")
        tmp.append("    cms.untracked.vstring(\"drop *\", \\")
        tmp.append("                          \"keep *_MEtoEDMConverter_*_*\"),")
        output_file_name = self. \
                           create_output_file_name(dataset_name)
        tmp.append("    fileName = \\")
        tmp.append("    cms.untracked.string(\"%s\")," % output_file_name)
        tmp.append("    dataset = cms.untracked.PSet(")
        tmp.append("    dataTier = cms.untracked.string(\"RECO\"),")
        tmp.append("    filterName = cms.untracked.string(\"\")")
        tmp.append("    )")
        tmp.append("    )")
        tmp.append("")
        tmp.append("# Additional output definition")
        tmp.append("process.out_step = cms.EndPath(process.output)")
        tmp.append("")
        tmp.append("# Schedule definition")
        tmp.append("process.schedule = cms.Schedule(process.out_step)")
        tmp.append("")

        config_contents = "\n".join(tmp)

        # End of create_me_extraction_config.
        return config_contents

    ##########

##    def create_harvesting_config(self, dataset_name):
##        """Create the Python harvesting configuration for a given job.

##        NOTE: The reason to have a single harvesting configuration per
##        sample is to be able to specify the GlobalTag corresponding to
##        each sample. Since it has been decided that (apart from the
##        prompt reco) datasets cannot contain runs with different
##        GlobalTags, we don't need a harvesting config per run.

##        NOTE: This is the place where we distinguish between
##        single-step and two-step harvesting modes (at least for the
##        Python job configuration).

##        """

##        ###

##        if self.harvesting_mode == "single-step":
##            config_contents = self.create_harvesting_config_single_step(dataset_name)
##        elif self.harvesting_mode == "two-step":
##            config_contents = self.create_harvesting_config_two_step(dataset_name)
##        else:
##            # Impossible harvesting mode, we should never get here.
##            assert False, "ERROR: unknown harvesting mode `%s'" % \
##                   self.harvesting_mode

##        ###

##        # End of create_harvesting_config.
##        return config_contents

    ##########

    def write_crab_config(self):
        """Write a CRAB job configuration Python file.

        """

        self.logger.info("Writing CRAB configuration...")

        file_name_base = "crab.cfg"

        # Create CRAB configuration.
        crab_contents = self.create_crab_config()

        # Write configuration to file.
        crab_file_name = file_name_base
        try:
            crab_file = file(crab_file_name, "w")
            crab_file.write(crab_contents)
            crab_file.close()
        except IOError:
            self.logger.fatal("Could not write " \
                              "CRAB configuration to file `%s'" % \
                              crab_file_name)
            raise Error("ERROR: Could not write to file `%s'!" % \
                        crab_file_name)

        # End of write_crab_config.

    ##########

    def write_multicrab_config(self):
        """Write a multi-CRAB job configuration Python file.

        """

        self.logger.info("Writing multi-CRAB configuration...")

        file_name_base = "multicrab.cfg"

        # Create multi-CRAB configuration.
        multicrab_contents = self.create_multicrab_config()

        # Write configuration to file.
        multicrab_file_name = file_name_base
        try:
            multicrab_file = file(multicrab_file_name, "w")
            multicrab_file.write(multicrab_contents)
            multicrab_file.close()
        except IOError:
            self.logger.fatal("Could not write " \
                              "multi-CRAB configuration to file `%s'" % \
                              multicrab_file_name)
            raise Error("ERROR: Could not write to file `%s'!" % \
                        multicrab_file_name)

        # End of write_multicrab_config.

    ##########

    def write_harvesting_config(self, dataset_name):
        """Write a harvesting job configuration Python file.

        NOTE: This knows nothing about single-step or two-step
        harvesting. That's all taken care of by
        create_harvesting_config.

        """

        self.logger.debug("Writing harvesting configuration for `%s'..." % \
                          dataset_name)

        # Create Python configuration.
        config_contents = self.create_harvesting_config(dataset_name)

        # Write configuration to file.
        config_file_name = self. \
                           create_harvesting_config_file_name(dataset_name)
        try:
            config_file = file(config_file_name, "w")
            config_file.write(config_contents)
            config_file.close()
        except IOError:
            self.logger.fatal("Could not write " \
                              "harvesting configuration to file `%s'" % \
                              config_file_name)
            raise Error("ERROR: Could not write to file `%s'!" % \
                        config_file_name)

        # End of write_harvesting_config.

    ##########

    def write_me_extraction_config(self, dataset_name):
        """Write an ME-extraction configuration Python file.

        This `ME-extraction' (ME = Monitoring Element) is the first
        step of the two-step harvesting.

        """

        self.logger.debug("Writing ME-extraction configuration for `%s'..." % \
                          dataset_name)

        # Create Python configuration.
        config_contents = self.create_me_extraction_config(dataset_name)

        # Write configuration to file.
        config_file_name = self. \
                           create_me_summary_config_file_name(dataset_name)
        try:
            config_file = file(config_file_name, "w")
            config_file.write(config_contents)
            config_file.close()
        except IOError:
            self.logger.fatal("Could not write " \
                              "ME-extraction configuration to file `%s'" % \
                              config_file_name)
            raise Error("ERROR: Could not write to file `%s'!" % \
                        config_file_name)

        # End of write_me_extraction_config.

    ##########


    def ref_hist_mappings_needed(self, dataset_name=None):
        """Check if we need to load and check the reference mappings.

        For data the reference histograms should be taken
        automatically from the GlobalTag, so we don't need any
        mappings. For RelVals we need to know a mapping to be used in
        the es_prefer code snippet (different references for each of
        the datasets.)

        WARNING: This implementation is a bit convoluted.

        """

        # If no dataset name given, do everything, otherwise check
        # only this one dataset.
        if not dataset_name is None:
            data_type = self.datasets_information[dataset_name] \
                        ["datatype"]
            mappings_needed = (data_type == "mc")
            # DEBUG DEBUG DEBUG
            if not mappings_needed:
                assert data_type == "data"
            # DEBUG DEBUG DEBUG end
        else:
            tmp = [self.ref_hist_mappings_needed(dataset_name) \
                   for dataset_name in \
                   self.datasets_information.keys()]
            mappings_needed = (True in tmp)

        # End of ref_hist_mappings_needed.
        return mappings_needed

    ##########

    def load_ref_hist_mappings(self):
        """Load the reference histogram mappings from file.

        The dataset name to reference histogram name mappings are read
        from a text file specified in self.ref_hist_mappings_file_name.

        """

        # DEBUG DEBUG DEBUG
        assert len(self.ref_hist_mappings) < 1, \
               "ERROR Should not be RE-loading " \
               "reference histogram mappings!"
        # DEBUG DEBUG DEBUG end

        self.logger.info("Loading reference histogram mappings " \
                         "from file `%s'" % \
                         self.ref_hist_mappings_file_name)

        mappings_lines = None
        try:
            mappings_file = file(self.ref_hist_mappings_file_name, "r")
            mappings_lines = mappings_file.readlines()
            mappings_file.close()
        except IOError:
            msg = "ERROR: Could not open reference histogram mapping "\
                  "file `%s'" % self.ref_hist_mappings_file_name
            self.logger.fatal(msg)
            raise Error(msg)

        ##########

        # The format we expect is: two white-space separated pieces
        # per line. The first the dataset name for which the reference
        # should be used, the second one the name of the reference
        # histogram in the database.

        for mapping in mappings_lines:
            # Skip comment lines.
            if not mapping.startswith("#"):
                mapping = mapping.strip()
                if len(mapping) > 0:
                    mapping_pieces = mapping.split()
                    if len(mapping_pieces) != 2:
                        msg = "ERROR: The reference histogram mapping " \
                              "file contains a line I don't " \
                              "understand:\n  %s" % mapping
                        self.logger.fatal(msg)
                        raise Error(msg)
                    dataset_name = mapping_pieces[0].strip()
                    ref_hist_name = mapping_pieces[1].strip()
                    # We don't want people to accidentally specify
                    # multiple mappings for the same dataset. Just
                    # don't accept those cases.
                    if dataset_name in self.ref_hist_mappings:
                        msg = "ERROR: The reference histogram mapping " \
                              "file contains multiple mappings for " \
                              "dataset `%s'."
                        self.logger.fatal(msg)
                        raise Error(msg)

                    # All is well that ends well.
                    self.ref_hist_mappings[dataset_name] = ref_hist_name

        ##########

        self.logger.info("  Successfully loaded %d mapping(s)" % \
                         len(self.ref_hist_mappings))
        max_len = max([len(i) for i in self.ref_hist_mappings.keys()])
        for (map_from, map_to) in six.iteritems(self.ref_hist_mappings):
            self.logger.info("    %-*s -> %s" % \
                              (max_len, map_from, map_to))

        # End of load_ref_hist_mappings.

    ##########

    def check_ref_hist_mappings(self):
        """Make sure all necessary reference histograms exist.

        Check that for each of the datasets to be processed a
        reference histogram is specified and that that histogram
        exists in the database.

        NOTE: There's a little complication here. Since this whole
        thing was designed to allow (in principle) harvesting of both
        data and MC datasets in one go, we need to be careful to check
        the availability fof reference mappings only for those
        datasets that need it.

        """

        self.logger.info("Checking reference histogram mappings")

        for dataset_name in self.datasets_to_use:
            try:
                ref_hist_name = self.ref_hist_mappings[dataset_name]
            except KeyError:
                msg = "ERROR: No reference histogram mapping found " \
                      "for dataset `%s'" % \
                      dataset_name
                self.logger.fatal(msg)
                raise Error(msg)

            if not self.check_ref_hist_tag(ref_hist_name):
                msg = "Reference histogram tag `%s' " \
                      "(used for dataset `%s') does not exist!" % \
                      (ref_hist_name, dataset_name)
                self.logger.fatal(msg)
                raise Usage(msg)

        self.logger.info("  Done checking reference histogram mappings.")

        # End of check_ref_hist_mappings.

    ##########

    def build_datasets_information(self):
        """Obtain all information on the datasets that we need to run.

        Use DBS to figure out all required information on our
        datasets, like the run numbers and the GlobalTag. All
        information is stored in the datasets_information member
        variable.

        """

        # Get a list of runs in the dataset.
        # NOTE: The harvesting has to be done run-by-run, so we
        # split up datasets based on the run numbers. Strictly
        # speaking this is not (yet?) necessary for Monte Carlo
        # since all those samples use run number 1. Still, this
        # general approach should work for all samples.

        # Now loop over all datasets in the list and process them.
        # NOTE: This processing has been split into several loops
        # to be easier to follow, sacrificing a bit of efficiency.
        self.datasets_information = {}
        self.logger.info("Collecting information for all datasets to process")
        dataset_names = sorted(self.datasets_to_use.keys())
        for dataset_name in dataset_names:

            # Tell the user which dataset: nice with many datasets.
            sep_line = "-" * 30
            self.logger.info(sep_line)
            self.logger.info("  `%s'" % dataset_name)
            self.logger.info(sep_line)

            runs = self.dbs_resolve_runs(dataset_name)
            self.logger.info("    found %d run(s)" % len(runs))
            if len(runs) > 0:
                self.logger.debug("      run number(s): %s" % \
                                  ", ".join([str(i) for i in runs]))
            else:
                # DEBUG DEBUG DEBUG
                # This should never happen after the DBS checks.
                self.logger.warning("  --> skipping dataset "
                                    "without any runs")
                assert False, "Panic: found a dataset without runs " \
                       "after DBS checks!"
                # DEBUG DEBUG DEBUG end

            cmssw_version = self.dbs_resolve_cmssw_version(dataset_name)
            self.logger.info("    found CMSSW version `%s'" % cmssw_version)

            # Figure out if this is data or MC.
            datatype = self.dbs_resolve_datatype(dataset_name)
            self.logger.info("    sample is data or MC? --> %s" % \
                             datatype)

            ###

            # Try and figure out the GlobalTag to be used.
            if self.globaltag is None:
                globaltag = self.dbs_resolve_globaltag(dataset_name)
            else:
                globaltag = self.globaltag

            self.logger.info("    found GlobalTag `%s'" % globaltag)

            # DEBUG DEBUG DEBUG
            if globaltag == "":
                # Actually we should not even reach this point, after
                # our dataset sanity checks.
                assert datatype == "data", \
                       "ERROR Empty GlobalTag for MC dataset!!!"
            # DEBUG DEBUG DEBUG end

            ###

            # DEBUG DEBUG DEBUG
            #tmp = self.dbs_check_dataset_spread_old(dataset_name)
            # DEBUG DEBUG DEBUG end
            sites_catalog = self.dbs_check_dataset_spread(dataset_name)

            # Extract the total event counts.
            num_events = {}
            for run_number in sites_catalog.keys():
                num_events[run_number] = sites_catalog \
                                         [run_number]["all_sites"]
                del sites_catalog[run_number]["all_sites"]

            # Extract the information about whether or not datasets
            # are mirrored.
            mirror_catalog = {}
            for run_number in sites_catalog.keys():
                mirror_catalog[run_number] = sites_catalog \
                                             [run_number]["mirrored"]
                del sites_catalog[run_number]["mirrored"]

            # BUG BUG BUG
            # I think I could now get rid of that and just fill the
            # "sites" entry with the `inverse' of this
            # num_events_catalog(?).
            #num_sites = self.dbs_resolve_dataset_number_of_sites(dataset_name)
            #sites_catalog = self.dbs_check_dataset_spread(dataset_name)
            #sites_catalog = dict(zip(num_events_catalog.keys(),
            #                         [[j for i in num_events_catalog.values() for j in i.keys()]]))
            # BUG BUG BUG end

##            # DEBUG DEBUG DEBUG
##            # This is probably only useful to make sure we don't muck
##            # things up, right?
##            # Figure out across how many sites this sample has been spread.
##            if num_sites == 1:
##                self.logger.info("    sample is contained at a single site")
##            else:
##                self.logger.info("    sample is spread across %d sites" % \
##                                 num_sites)
##            if num_sites < 1:
##                # NOTE: This _should not_ happen with any valid dataset.
##                self.logger.warning("  --> skipping dataset which is not " \
##                                    "hosted anywhere")
##            # DEBUG DEBUG DEBUG end

            # Now put everything in a place where we can find it again
            # if we need it.
            self.datasets_information[dataset_name] = {}
            self.datasets_information[dataset_name]["runs"] = runs
            self.datasets_information[dataset_name]["cmssw_version"] = \
                                                                     cmssw_version
            self.datasets_information[dataset_name]["globaltag"] = globaltag
            self.datasets_information[dataset_name]["datatype"] = datatype
            self.datasets_information[dataset_name]["num_events"] = num_events
            self.datasets_information[dataset_name]["mirrored"] = mirror_catalog
            self.datasets_information[dataset_name]["sites"] = sites_catalog

            # Each run of each dataset has a different CASTOR output
            # path.
            castor_path_common = self.create_castor_path_name_common(dataset_name)
            self.logger.info("    output will go into `%s'" % \
                             castor_path_common)

            castor_paths = dict(list(zip(runs,
                                    [self.create_castor_path_name_special(dataset_name, i, castor_path_common) \
                                     for i in runs])))
            for path_name in castor_paths.values():
                self.logger.debug("      %s" % path_name)
            self.datasets_information[dataset_name]["castor_path"] = \
                                                                   castor_paths

        # End of build_datasets_information.

    ##########

    def show_exit_message(self):
        """Tell the user what to do now, after this part is done.

        This should provide the user with some (preferably
        copy-pasteable) instructions on what to do now with the setups
        and files that have been created.

        """

        # TODO TODO TODO
        # This could be improved a bit.
        # TODO TODO TODO end

        sep_line = "-" * 60

        self.logger.info("")
        self.logger.info(sep_line)
        self.logger.info("  Configuration files have been created.")
        self.logger.info("  From here on please follow the usual CRAB instructions.")
        self.logger.info("  Quick copy-paste instructions are shown below.")
        self.logger.info(sep_line)

        self.logger.info("")
        self.logger.info("    Create all CRAB jobs:")
        self.logger.info("      multicrab -create")
        self.logger.info("")
        self.logger.info("    Submit all CRAB jobs:")
        self.logger.info("      multicrab -submit")
        self.logger.info("")
        self.logger.info("    Check CRAB status:")
        self.logger.info("      multicrab -status")
        self.logger.info("")

        self.logger.info("")
        self.logger.info("  For more information please see the CMS Twiki:")
        self.logger.info("    %s" % twiki_url)
        self.logger.info(sep_line)

        # If there were any jobs for which we could not find a
        # matching site show a warning message about that.
        if not self.all_sites_found:
            self.logger.warning("  For some of the jobs no matching " \
                                "site could be found")
            self.logger.warning("  --> please scan your multicrab.cfg" \
                                "for occurrences of `%s'." % \
                                self.no_matching_site_found_str)
            self.logger.warning("      You will have to fix those " \
                                "by hand, sorry.")

        # End of show_exit_message.

    ##########

    def run(self):
        "Main entry point of the CMS harvester."

        # Start with a positive thought.
        exit_code = 0

        try:

            try:

                # Parse all command line options and arguments
                self.parse_cmd_line_options()
                # and check that they make sense.
                self.check_input_status()

                # Check if CMSSW is setup.
                self.check_cmssw()

                # Check if DBS is setup,
                self.check_dbs()
                # and if all is fine setup the Python side.
                self.setup_dbs()

                # Fill our dictionary with all the required info we
                # need to understand harvesting jobs. This needs to be
                # done after the CMSSW version is known.
                self.setup_harvesting_info()

                # Obtain list of dataset names to consider
                self.build_dataset_use_list()
                # and the list of dataset names to ignore.
                self.build_dataset_ignore_list()

                # The same for the runs lists (if specified).
                self.build_runs_use_list()
                self.build_runs_ignore_list()

                # Process the list of datasets to ignore and fold that
                # into the list of datasets to consider.
                # NOTE: The run-based selection is done later since
                # right now we don't know yet which runs a dataset
                # contains.
                self.process_dataset_ignore_list()

                # Obtain all required information on the datasets,
                # like run numbers and GlobalTags.
                self.build_datasets_information()

                if self.use_ref_hists and \
                       self.ref_hist_mappings_needed():
                    # Load the dataset name to reference histogram
                    # name mappings from file.
                    self.load_ref_hist_mappings()
                    # Now make sure that for all datasets we want to
                    # process there is a reference defined. Otherwise
                    # just bomb out before wasting any more time.
                    self.check_ref_hist_mappings()
                else:
                    self.logger.info("No need to load reference " \
                                     "histogram mappings file")

                # OBSOLETE OBSOLETE OBSOLETE
##                # TODO TODO TODO
##                # Need to think about where this should go, but
##                # somewhere we have to move over the fact that we want
##                # to process all runs for each dataset that we're
##                # considering. This basically means copying over the
##                # information from self.datasets_information[]["runs"]
##                # to self.datasets_to_use[].
##                for dataset_name in self.datasets_to_use.keys():
##                    self.datasets_to_use[dataset_name] = self.datasets_information[dataset_name]["runs"]
##                # TODO TODO TODO end
                # OBSOLETE OBSOLETE OBSOLETE end

                self.process_runs_use_and_ignore_lists()

                # If we've been asked to sacrifice some parts of
                # spread-out samples in order to be able to partially
                # harvest them, we'll do that here.
                if self.harvesting_mode == "single-step-allow-partial":
                    self.singlify_datasets()

                # Check dataset name(s)
                self.check_dataset_list()
                # and see if there is anything left to do.
                if len(self.datasets_to_use) < 1:
                    self.logger.info("After all checks etc. " \
                                     "there are no datasets (left?) " \
                                     "to process")
                else:

                    self.logger.info("After all checks etc. we are left " \
                                     "with %d dataset(s) to process " \
                                     "for a total of %d runs" % \
                                     (len(self.datasets_to_use),
                                      sum([len(i) for i in \
                                           self.datasets_to_use.values()])))

                    # NOTE: The order in which things are done here is
                    # important. At the end of the job, independent on
                    # how it ends (exception, CTRL-C, normal end) the
                    # book keeping is written to file. At that time it
                    # should be clear which jobs are done and can be
                    # submitted. This means we first create the
                    # general files, and then the per-job config
                    # files.

                    # TODO TODO TODO
                    # It would be good to modify the book keeping a
                    # bit. Now we write the crab.cfg (which is the
                    # same for all samples and runs) and the
                    # multicrab.cfg (which contains blocks for all
                    # runs of all samples) without updating our book
                    # keeping. The only place we update the book
                    # keeping is after writing the harvesting config
                    # file for a given dataset. Since there is only
                    # one single harvesting configuration for each
                    # dataset, we have no book keeping information on
                    # a per-run basis.
                    # TODO TODO TODO end

                    # Check if the CASTOR output area exists. If
                    # necessary create it.
                    self.create_and_check_castor_dirs()

                    # Create one crab and one multicrab configuration
                    # for all jobs together.
                    self.write_crab_config()
                    self.write_multicrab_config()

                    # Loop over all datasets and create harvesting
                    # config files for all of them. One harvesting
                    # config per dataset is enough. The same file will
                    # be re-used by CRAB for each run.
                    # NOTE: We always need a harvesting
                    # configuration. For the two-step harvesting we
                    # also need a configuration file for the first
                    # step: the monitoring element extraction.
                    for dataset_name in self.datasets_to_use.keys():
                        try:
                            self.write_harvesting_config(dataset_name)
                            if self.harvesting_mode == "two-step":
                                self.write_me_extraction_config(dataset_name)
                        except:
                            # Doh! Just re-raise the damn thing.
                            raise
                        else:
##                            tmp = self.datasets_information[dataset_name] \
##                                  ["num_events"]
                            tmp = {}
                            for run_number in self.datasets_to_use[dataset_name]:
                                tmp[run_number] = self.datasets_information \
                                                  [dataset_name]["num_events"][run_number]
                            if dataset_name in self.book_keeping_information:
                                self.book_keeping_information[dataset_name].update(tmp)
                            else:
                                self.book_keeping_information[dataset_name] = tmp

                    # Explain the user what to do now.
                    self.show_exit_message()

            except Usage as err:
                # self.logger.fatal(err.msg)
                # self.option_parser.print_help()
                pass

            except Error as err:
                # self.logger.fatal(err.msg)
                exit_code = 1

            except Exception as err:
                # Hmmm, ignore keyboard interrupts from the
                # user. These are not a `serious problem'. We also
                # skip SystemExit, which is the exception thrown when
                # one calls sys.exit(). This, for example, is done by
                # the option parser after calling print_help(). We
                # also have to catch all `no such option'
                # complaints. Everything else we catch here is a
                # `serious problem'.
                if isinstance(err, SystemExit):
                    self.logger.fatal(err.code)
                elif not isinstance(err, KeyboardInterrupt):
                    self.logger.fatal("!" * 50)
                    self.logger.fatal("  This looks like a serious problem.")
                    self.logger.fatal("  If you are sure you followed all " \
                                      "instructions")
                    self.logger.fatal("  please copy the below stack trace together")
                    self.logger.fatal("  with a description of what you were doing to")
                    self.logger.fatal("  jeroen.hegeman@cern.ch.")
                    self.logger.fatal("  %s" % self.ident_string())
                    self.logger.fatal("!" * 50)
                    self.logger.fatal(str(err))
                    import traceback
                    traceback_string = traceback.format_exc()
                    for line in traceback_string.split("\n"):
                        self.logger.fatal(line)
                    self.logger.fatal("!" * 50)
                    exit_code = 2

        # This is the stuff that we should really do, no matter
        # what. Of course cleaning up after ourselves is also done
        # from this place.  This alsokeeps track of the book keeping
        # so far. (This means that if half of the configuration files
        # were created before e.g. the disk was full, we should still
        # have a consistent book keeping file.
        finally:

            self.cleanup()

        ###

        if self.crab_submission == True:
            os.system("multicrab -create")
            os.system("multicrab -submit")

        # End of run.
        return exit_code

    # End of CMSHarvester.

###########################################################################
## Main entry point.
###########################################################################

if __name__ == "__main__":
    "Main entry point for harvesting."

    CMSHarvester().run()

    # Done.

###########################################################################