RootInputFileSequence.cc

Go to the documentation of this file.

/*----------------------------------------------------------------------
----------------------------------------------------------------------*/
#include "DuplicateChecker.h"
#include "PoolSource.h"
#include "RootFile.h"
#include "RootInputFileSequence.h"
#include "RootTree.h"

#include "DataFormats/Provenance/interface/BranchID.h"
#include "DataFormats/Provenance/interface/BranchIDListHelper.h"
#include "DataFormats/Provenance/interface/ProductRegistry.h"
#include "FWCore/Catalog/interface/SiteLocalConfig.h"
#include "FWCore/Framework/interface/EventPrincipal.h"
#include "FWCore/Framework/interface/FileBlock.h"
#include "FWCore/Framework/interface/LuminosityBlockPrincipal.h"
#include "FWCore/Framework/interface/RunPrincipal.h"
#include "FWCore/MessageLogger/interface/MessageLogger.h"
#include "FWCore/ParameterSet/interface/ParameterSet.h"
#include "FWCore/ParameterSet/interface/ParameterSetDescription.h"
#include "FWCore/ServiceRegistry/interface/Service.h"
#include "Utilities/StorageFactory/interface/StorageFactory.h"

#include "CLHEP/Random/RandFlat.h"
#include "InputFile.h"
#include "TSystem.h"

#include <random>

namespace edm {
  RootInputFileSequence::RootInputFileSequence(
                ParameterSet const& pset,
                PoolSource& input,
                InputFileCatalog const& catalog,
                unsigned int nStreams,
                InputType inputType) :
    input_(input),
    inputType_(inputType),
    catalog_(catalog),
    firstFile_(true),
    lfn_("unknown"),
    lfnHash_(0U),
    findFileForSpecifiedID_(nullptr),
    fileIterBegin_(fileCatalogItems().begin()),
    fileIterEnd_(fileCatalogItems().end()),
    fileIter_(fileIterEnd_),
    fileIterLastOpened_(fileIterEnd_),
    rootFile_(),
    branchesMustMatch_(BranchDescription::Permissive),
    indexesIntoFiles_(fileCatalogItems().size()),
    orderedProcessHistoryIDs_(),
    nStreams_(nStreams),
    eventSkipperByID_(inputType == InputType::Primary ? EventSkipperByID::create(pset).release() : 0),
    eventsRemainingInFile_(0),
    // The default value provided as the second argument to the getUntrackedParameter function call
    // is not used when the ParameterSet has been validated and the parameters are not optional
    // in the description.  This is currently true when PoolSource is the primary input source.
    // The modules that use PoolSource as a SecSource have not defined their fillDescriptions function
    // yet, so the ParameterSet does not get validated yet.  As soon as all the modules with a SecSource
    // have defined descriptions, the defaults in the getUntrackedParameterSet function calls can
    // and should be deleted from the code.
    initialNumberOfEventsToSkip_(inputType != InputType::SecondaryFile ? pset.getUntrackedParameter<unsigned int>("skipEvents", 0U) : 0U),
    noEventSort_(inputType == InputType::Primary ? pset.getUntrackedParameter<bool>("noEventSort", true) : false),
    skipBadFiles_(pset.getUntrackedParameter<bool>("skipBadFiles", false)),
    bypassVersionCheck_(pset.getUntrackedParameter<bool>("bypassVersionCheck", false)),
    treeCacheSize_(noEventSort_ ? pset.getUntrackedParameter<unsigned int>("cacheSize", roottree::defaultCacheSize) : 0U),
    treeMaxVirtualSize_(pset.getUntrackedParameter<int>("treeMaxVirtualSize", -1)),
    setRun_(pset.getUntrackedParameter<unsigned int>("setRunNumber", 0U)),
    productSelectorRules_(pset, "inputCommands", "InputSource"),
    duplicateChecker_(inputType == InputType::Primary ? new DuplicateChecker(pset) : 0),
    dropDescendants_(pset.getUntrackedParameter<bool>("dropDescendantsOfDroppedBranches", inputType != InputType::SecondarySource)),
    labelRawDataLikeMC_(pset.getUntrackedParameter<bool>("labelRawDataLikeMC", true)),
    usingGoToEvent_(false),
    enablePrefetching_(false),
    usedFallback_(false) {

    // The SiteLocalConfig controls the TTreeCache size and the prefetching settings.
    Service<SiteLocalConfig> pSLC;
    if(pSLC.isAvailable()) {
      if(treeCacheSize_ != 0U && pSLC->sourceTTreeCacheSize()) {
        treeCacheSize_ = *(pSLC->sourceTTreeCacheSize());
      }
      enablePrefetching_ = pSLC->enablePrefetching();
    }

    std::string branchesMustMatch = pset.getUntrackedParameter<std::string>("branchesMustMatch", std::string("permissive"));
    if(branchesMustMatch == std::string("strict")) branchesMustMatch_ = BranchDescription::Strict;

    StorageFactory *factory = StorageFactory::get();

    if(inputType == InputType::SecondarySource) {
      // For the secondary input source we do not stage in,
      // and we randomly choose the first file to open.
      // We cannot use the random number service yet.
      std::ifstream f("/dev/urandom");
      unsigned int seed;
      f.read(reinterpret_cast<char*>(&seed), sizeof(seed)); 
      std::default_random_engine dre(seed);
      size_t count = fileIterEnd_ - fileIterBegin_;
      std::uniform_int_distribution<int> distribution(0, count - 1);
      while(!rootFile_ && count != 0) {
        --count;
        int offset = distribution(dre);
        fileIter_ = fileIterBegin_ + offset;
        initFile(skipBadFiles_);
      }
    } else {
      // Prestage the files
      for(fileIter_ = fileIterBegin_; fileIter_ != fileIterEnd_; ++fileIter_) {
        factory->activateTimeout(fileIter_->fileName());
        factory->stagein(fileIter_->fileName());
        //NOTE: we do not want to stage in all secondary files since we can be given a list of
        // thousands of files and prestaging all those files can cause a site to fail.
        // So, we stage in the first secondary file only.
        if(inputType_ != InputType::Primary) {
          break;
        }
      }
      // Open the first file.
      for(fileIter_ = fileIterBegin_; fileIter_ != fileIterEnd_; ++fileIter_) {
        initFile(skipBadFiles_);
        if(rootFile_) break;
      }
    }
    if(rootFile_) {
      productRegistryUpdate().updateFromInput(rootFile_->productRegistry()->productList());
      if(inputType == InputType::Primary && initialNumberOfEventsToSkip_ != 0) {
        skipEvents(initialNumberOfEventsToSkip_);
      }
    }
  }

  std::vector<FileCatalogItem> const&
  RootInputFileSequence::fileCatalogItems() const {
    return catalog_.fileCatalogItems();
  }

  void
  RootInputFileSequence::endJob() {
    closeFile_();
  }

  std::unique_ptr<FileBlock>
  RootInputFileSequence::readFile_() {
    if(firstFile_) {
      // The first input file has already been opened.
      firstFile_ = false;
      if(!rootFile_) {
        initFile(skipBadFiles_);
      }
    } else {
      if(!nextFile()) {
        assert(0);
      }
    }
    if(!rootFile_) {
      return std::unique_ptr<FileBlock>(new FileBlock);
    }
    return rootFile_->createFileBlock();
  }

  void RootInputFileSequence::closeFile_() {
    // close the currently open file, if any, and delete the RootFile object.
    if(rootFile_) {
      if(inputType_ != InputType::SecondarySource) {
        std::unique_ptr<InputSource::FileCloseSentry>
        sentry((inputType_ == InputType::Primary) ? new InputSource::FileCloseSentry(input_, lfn_, usedFallback_) : 0);
        rootFile_->close();
        if(duplicateChecker_) duplicateChecker_->inputFileClosed();
      }
      rootFile_.reset();
    }
  }

  void RootInputFileSequence::initFile(bool skipBadFiles) {
    // We are really going to close the open file.

    // If this is the primary sequence, we are not duplicate checking across files
    // and we are not using random access to find events, then we can delete the
    // IndexIntoFile for the file we are closing. If we can't delete all of it,
    // then we can delete the parts we do not need.
    if(fileIterLastOpened_ != fileIterEnd_) {
      size_t currentIndexIntoFile = fileIterLastOpened_ - fileIterBegin_;
      bool needIndexesForDuplicateChecker = duplicateChecker_ && duplicateChecker_->checkingAllFiles() && !duplicateChecker_->checkDisabled();
      bool deleteIndexIntoFile = inputType_ == InputType::Primary &&
                                 !needIndexesForDuplicateChecker &&
                                 !usingGoToEvent_;
      if(deleteIndexIntoFile) {
        indexesIntoFiles_[currentIndexIntoFile].reset();
      } else {
        if(indexesIntoFiles_[currentIndexIntoFile]) indexesIntoFiles_[currentIndexIntoFile]->inputFileClosed();
      }
      fileIterLastOpened_ = fileIterEnd_;
    }
    closeFile_();

    if(fileIter_ == fileIterEnd_) {
      // No files specified
      return;
    }

    // Check if the logical file name was found.
    if(fileIter_->fileName().empty()) {
      // LFN not found in catalog.
      InputFile::reportSkippedFile(fileIter_->fileName(), fileIter_->logicalFileName());
      if(!skipBadFiles) {
        throw cms::Exception("LogicalFileNameNotFound", "RootInputFileSequence::initFile()\n")
          << "Logical file name '" << fileIter_->logicalFileName() << "' was not found in the file catalog.\n"
          << "If you wanted a local file, you forgot the 'file:' prefix\n"
          << "before the file name in your configuration file.\n";
      }
      LogWarning("") << "Input logical file: " << fileIter_->logicalFileName() << " was not found in the catalog, and will be skipped.\n";
      return;
    }

    lfn_ = fileIter_->logicalFileName().empty() ? fileIter_->fileName() : fileIter_->logicalFileName();
    lfnHash_ = std::hash<std::string>()(lfn_);
    usedFallback_ = false;

    // Determine whether we have a fallback URL specified; if so, prepare it;
    // Only valid if it is non-empty and differs from the original filename.
    std::string fallbackName = fileIter_->fallbackFileName();
    bool hasFallbackUrl = !fallbackName.empty() && fallbackName != fileIter_->fileName();

    std::shared_ptr<InputFile> filePtr;
    std::list<std::string> originalInfo;
    try {
      std::unique_ptr<InputSource::FileOpenSentry>
        sentry(inputType_ == InputType::Primary ? new InputSource::FileOpenSentry(input_, lfn_, usedFallback_) : 0);
      filePtr.reset(new InputFile(gSystem->ExpandPathName(fileIter_->fileName().c_str()), "  Initiating request to open file ", inputType_));
    }
    catch (cms::Exception const& e) {
      if(!skipBadFiles) {
        if(hasFallbackUrl) {
          std::ostringstream out;
          out << e.explainSelf();
          std::string pfn(gSystem->ExpandPathName(fallbackName.c_str()));
          InputFile::reportFallbackAttempt(pfn, fileIter_->logicalFileName(), out.str());
          originalInfo = e.additionalInfo();
        } else {
          InputFile::reportSkippedFile(fileIter_->fileName(), fileIter_->logicalFileName());
          Exception ex(errors::FileOpenError, "", e);
          ex.addContext("Calling RootInputFileSequence::initFile()");
          std::ostringstream out;
          out << "Input file " << fileIter_->fileName() << " could not be opened.";
          ex.addAdditionalInfo(out.str());
          throw ex;
        }
      }
    }
    if(!filePtr && (hasFallbackUrl)) {
      try {
        usedFallback_ = true;
        std::unique_ptr<InputSource::FileOpenSentry>
          sentry(inputType_ == InputType::Primary ? new InputSource::FileOpenSentry(input_, lfn_, usedFallback_) : 0);
        std::string fallbackFullName = gSystem->ExpandPathName(fallbackName.c_str());
        StorageFactory *factory = StorageFactory::get();
        if (factory) {factory->activateTimeout(fallbackFullName);}
        filePtr.reset(new InputFile(fallbackFullName.c_str(), "  Fallback request to file ", inputType_));
      }
      catch (cms::Exception const& e) {
        if(!skipBadFiles) {
          InputFile::reportSkippedFile(fileIter_->fileName(), fileIter_->logicalFileName());
          Exception ex(errors::FallbackFileOpenError, "", e);
          ex.addContext("Calling RootInputFileSequence::initFile()");
          std::ostringstream out;
          out << "Input file " << fileIter_->fileName() << " could not be opened.\n";
          out << "Fallback Input file " << fallbackName << " also could not be opened.";
          if (originalInfo.size()) {
            out << std::endl << "Original exception info is above; fallback exception info is below.";
            ex.addAdditionalInfo(out.str());
            for (auto const & s : originalInfo) {
              ex.addAdditionalInfo(s);
            }
          } else {
            ex.addAdditionalInfo(out.str());
          }
          throw ex;
        }
      }
    }
    if(filePtr) {
      std::vector<std::shared_ptr<IndexIntoFile> >::size_type currentIndexIntoFile = fileIter_ - fileIterBegin_;
      rootFile_ = RootFileSharedPtr(new RootFile(
          fileIter_->fileName(),
          processConfiguration(),
          fileIter_->logicalFileName(),
          filePtr,
          eventSkipperByID_,
          initialNumberOfEventsToSkip_ != 0,
          remainingEvents(),
          remainingLuminosityBlocks(),
      nStreams_,
          treeCacheSize_,
          treeMaxVirtualSize_,
          input_.processingMode(),
          setRun_,
          noEventSort_,
          productSelectorRules_,
          inputType_,
          (inputType_ == InputType::SecondarySource ?  std::make_shared<BranchIDListHelper>() :  input_.branchIDListHelper()),
          (inputType_ == InputType::SecondarySource ?  std::shared_ptr<ThinnedAssociationsHelper>() : input_.thinnedAssociationsHelper()),
          associationsFromSecondary_,
          duplicateChecker_,
          dropDescendants_,
          processHistoryRegistryForUpdate(),
          indexesIntoFiles_,
          currentIndexIntoFile,
          orderedProcessHistoryIDs_,
          bypassVersionCheck_,
          labelRawDataLikeMC_,
          usingGoToEvent_,
          enablePrefetching_));

      assert(rootFile_);
      fileIterLastOpened_ = fileIter_;
      indexesIntoFiles_[currentIndexIntoFile] = rootFile_->indexIntoFileSharedPtr();
      char const* inputType = 0;
      switch(inputType_) {
      case InputType::Primary: inputType = "primaryFiles"; break;
      case InputType::SecondaryFile: inputType = "secondaryFiles"; break;
      case InputType::SecondarySource: inputType = "mixingFiles"; break;
      }
      rootFile_->reportOpened(inputType);
    } else {
      InputFile::reportSkippedFile(fileIter_->fileName(), fileIter_->logicalFileName());
      if(!skipBadFiles) {
        throw Exception(errors::FileOpenError) <<
           "RootInputFileSequence::initFile(): Input file " << fileIter_->fileName() << " was not found or could not be opened.\n";
      }
      LogWarning("") << "Input file: " << fileIter_->fileName() << " was not found or could not be opened, and will be skipped.\n";
    }
  }

  std::shared_ptr<ProductRegistry const>
  RootInputFileSequence::fileProductRegistry() const {
    assert(rootFile_);
    return rootFile_->productRegistry();
  }

  std::shared_ptr<BranchIDListHelper const>
  RootInputFileSequence::fileBranchIDListHelper() const {
    assert(rootFile_);
    return rootFile_->branchIDListHelper();
  }

  bool RootInputFileSequence::nextFile() {
    if(fileIter_ != fileIterEnd_) ++fileIter_;
    if(fileIter_ == fileIterEnd_) {
      if(inputType_ == InputType::Primary) {
        return false;
      } else {
        fileIter_ = fileIterBegin_;
      }
    }

    initFile(skipBadFiles_);

    if(inputType_ == InputType::Primary && rootFile_) {
      // make sure the new product registry is compatible with the main one
      std::string mergeInfo = productRegistryUpdate().merge(*rootFile_->productRegistry(),
                                                            fileIter_->fileName(),
                                                            branchesMustMatch_);
      if(!mergeInfo.empty()) {
        throw Exception(errors::MismatchedInputFiles,"RootInputFileSequence::nextFile()") << mergeInfo;
      }
    }
    return true;
  }

  bool RootInputFileSequence::previousFile() {
    if(fileIter_ == fileIterBegin_) {
      if(inputType_ == InputType::Primary) {
        return false;
      } else {
        fileIter_ = fileIterEnd_;
      }
    }
    --fileIter_;

    initFile(false);

    if(inputType_ == InputType::Primary && rootFile_) {
      // make sure the new product registry is compatible to the main one
      std::string mergeInfo = productRegistryUpdate().merge(*rootFile_->productRegistry(),
                                                            fileIter_->fileName(),
                                                            branchesMustMatch_);
      if(!mergeInfo.empty()) {
        throw Exception(errors::MismatchedInputFiles,"RootInputFileSequence::previousEvent()") << mergeInfo;
      }
    }
    if(rootFile_) rootFile_->setToLastEntry();
    return true;
  }

  RootInputFileSequence::~RootInputFileSequence() {
  }

  std::shared_ptr<RunAuxiliary>
  RootInputFileSequence::readRunAuxiliary_() {
    assert(rootFile_);
    return rootFile_->readRunAuxiliary_();
  }

  std::shared_ptr<LuminosityBlockAuxiliary>
  RootInputFileSequence::readLuminosityBlockAuxiliary_() {
    assert(rootFile_);
    return rootFile_->readLuminosityBlockAuxiliary_();
  }

  void
  RootInputFileSequence::readRun_(RunPrincipal& runPrincipal) {
    assert(rootFile_);
    rootFile_->readRun_(runPrincipal);
  }

  void
  RootInputFileSequence::readLuminosityBlock_(LuminosityBlockPrincipal& lumiPrincipal) {
    assert(rootFile_);
    rootFile_->readLuminosityBlock_(lumiPrincipal);
  }

  // readEvent() is responsible for setting up the EventPrincipal.
  //
  //   1. fill an EventPrincipal with a unique EventID
  //   2. For each entry in the provenance, put in one ProductHolder,
  //      holding the Provenance for the corresponding EDProduct.
  //   3. set up the caches in the EventPrincipal to know about this
  //      ProductHolder.
  //
  // We do *not* create the EDProduct instance (the equivalent of reading
  // the branch containing this EDProduct. That will be done by the Delayed Reader,
  //  when it is asked to do so.
  //

  void
  RootInputFileSequence::readEvent(EventPrincipal& eventPrincipal) {
    assert(rootFile_);
    rootFile_->readEvent(eventPrincipal);
  }

  InputSource::ItemType
  RootInputFileSequence::getNextItemType(RunNumber_t& run, LuminosityBlockNumber_t& lumi, EventNumber_t& event) {
    if(fileIter_ == fileIterEnd_) {
      return InputSource::IsStop;
    }
    if(firstFile_) {
      return InputSource::IsFile;
    }
    if(rootFile_) {
      IndexIntoFile::EntryType entryType = rootFile_->getNextItemType(run, lumi, event);
      if(entryType == IndexIntoFile::kEvent) {
        return InputSource::IsEvent;
      } else if(entryType == IndexIntoFile::kLumi) {
        return InputSource::IsLumi;
      } else if(entryType == IndexIntoFile::kRun) {
        return InputSource::IsRun;
      }
      assert(entryType == IndexIntoFile::kEnd);
    }
    if(fileIter_ + 1 == fileIterEnd_) {
      return InputSource::IsStop;
    }
    return InputSource::IsFile;
  }

  bool
  RootInputFileSequence::containedInCurrentFile(RunNumber_t run, LuminosityBlockNumber_t lumi, EventNumber_t event) const {
    if(!rootFile_) return false;
    return rootFile_->containsItem(run, lumi, event);
  }

  // Rewind to before the first event that was read.
  void
  RootInputFileSequence::rewind_() {
    if(fileIter_ != fileIterBegin_) {
      closeFile_();
      fileIter_ = fileIterBegin_;
    }
    if(!rootFile_) {
      initFile(false);
    }
    rewindFile();
    firstFile_ = true;
    if(rootFile_) {
      if(initialNumberOfEventsToSkip_ != 0) {
        skipEvents(initialNumberOfEventsToSkip_);
      }
    }
  }

  // Rewind to the beginning of the current file
  void
  RootInputFileSequence::rewindFile() {
    if(rootFile_) rootFile_->rewind();
  }

  // Advance "offset" events.  Offset can be positive or negative (or zero).
  bool
  RootInputFileSequence::skipEvents(int offset) {
    // We never call skipEvents for secondary input files.  If we did,
    // we would have to implement synchronization if a new file is opened.
    // To avoid this, just assert.
    assert(inputType_ != InputType::SecondaryFile);
    assert(rootFile_);
    while(offset != 0) {
      bool atEnd = rootFile_->skipEvents(offset);
      if((offset > 0 || atEnd) && !nextFile()) {
        return false;
      }
      if(offset < 0 && !previousFile()) {
        fileIter_ = fileIterEnd_;
        return false;
      }
    }
    return true;
  }

  bool
  RootInputFileSequence::goToEvent(EventID const& eventID) {
    usingGoToEvent_ = true;
    if(rootFile_) {
      if(rootFile_->goToEvent(eventID)) {
        return true;
      }
      // If only one input file, give up now, to save time.
      if(rootFile_ && indexesIntoFiles_.size() == 1) {
        return false;
      }
      // Save the current file and position so that we can restore them
      // if we fail to restore the desired event
      bool closedOriginalFile = false;
      std::vector<FileCatalogItem>::const_iterator originalFile = fileIter_;
      IndexIntoFile::IndexIntoFileItr originalPosition = rootFile_->indexIntoFileIter();

      // Look for item (run/lumi/event) in files previously opened without reopening unnecessary files.
      typedef std::vector<std::shared_ptr<IndexIntoFile> >::const_iterator Iter;
      for(Iter it = indexesIntoFiles_.begin(), itEnd = indexesIntoFiles_.end(); it != itEnd; ++it) {
        if(*it && (*it)->containsItem(eventID.run(), eventID.luminosityBlock(), eventID.event())) {
          // We found it. Close the currently open file, and open the correct one.
          fileIter_ = fileIterBegin_ + (it - indexesIntoFiles_.begin());
          initFile(false);
          // Now get the item from the correct file.
          assert(rootFile_);
          bool found = rootFile_->goToEvent(eventID);
          assert(found);
          return true;
        }
      }
      // Look for item in files not yet opened.
      for(Iter it = indexesIntoFiles_.begin(), itEnd = indexesIntoFiles_.end(); it != itEnd; ++it) {
        if(!*it) {
          fileIter_ = fileIterBegin_ + (it - indexesIntoFiles_.begin());
          initFile(false);
          closedOriginalFile = true;
          if((*it)->containsItem(eventID.run(), eventID.luminosityBlock(), eventID.event())) {
            assert(rootFile_);
            if(rootFile_->goToEvent(eventID)) {
              return true;
            }
          }
        }
      }
      if(closedOriginalFile) {
        fileIter_ = originalFile;
        initFile(false);
        assert(rootFile_);
        rootFile_->setPosition(originalPosition);
      }
    }
    return false;
  }

  bool
  RootInputFileSequence::skipToItemInNewFile(RunNumber_t run, LuminosityBlockNumber_t lumi, EventNumber_t event, size_t fileNameHash) {
    // Look for item in files not yet opened. We have a hash of the logical file name
    assert(fileNameHash != 0U);
    // If the lookup table is not yet filled in, fill it. 
    if(!findFileForSpecifiedID_) {
      // We use a multimap because there may be hash collisions (Two different LFNs could have the same hash).
      // We map the hash of the LFN to the index into the list of files.
      findFileForSpecifiedID_.reset(new std::unordered_multimap<size_t, size_t>);
      auto hasher = std::hash<std::string>();
      for(auto fileIter = fileIterBegin_; fileIter != fileIterEnd_; ++fileIter) {
        findFileForSpecifiedID_->insert(std::make_pair(hasher(fileIter->logicalFileName()), fileIter - fileIterBegin_));
      }
    }
    // Look up the logical file name in the table
    auto range = findFileForSpecifiedID_->equal_range(fileNameHash);
    for(auto iter = range.first; iter != range.second; ++iter) {
      // Don't look in files previously opened, because those have already been searched.
      if(!indexesIntoFiles_[iter->second]) {
        fileIter_ = fileIterBegin_ + iter->second;
        initFile(false);
        assert(rootFile_);
        bool found = rootFile_->setEntryAtItem(run, lumi, event);
        if(found) {
          return true;
        }
      }
    }
    // Not found
    return false;
  }

  bool
  RootInputFileSequence::skipToItemInNewFile(RunNumber_t run, LuminosityBlockNumber_t lumi, EventNumber_t event) {
    // Look for item in files not yet opened.  We do not have a valid hash of the logical file name.
    typedef std::vector<std::shared_ptr<IndexIntoFile> >::const_iterator Iter;
    for(Iter it = indexesIntoFiles_.begin(), itEnd = indexesIntoFiles_.end(); it != itEnd; ++it) {
      if(!*it) {
        // File not yet opened.
        fileIter_ = fileIterBegin_ + (it - indexesIntoFiles_.begin());
        initFile(false);
        assert(rootFile_);
        bool found = rootFile_->setEntryAtItem(run, lumi, event);
        if(found) {
          return true;
        }
      }
    }
    // Not found
    return false;
  }

  bool
  RootInputFileSequence::skipToItem(RunNumber_t run, LuminosityBlockNumber_t lumi, EventNumber_t event, size_t fileNameHash, bool currentFileFirst) {
    // Attempt to find item in currently open input file.
    bool found = currentFileFirst && rootFile_ && rootFile_->setEntryAtItem(run, lumi, event);
    if(!found) {
      // If only one input file, give up now, to save time.
      if(currentFileFirst && rootFile_ && indexesIntoFiles_.size() == 1) {
        return false;
      }
      // Look for item (run/lumi/event) in files previously opened without reopening unnecessary files.
      typedef std::vector<std::shared_ptr<IndexIntoFile> >::const_iterator Iter;
      for(Iter it = indexesIntoFiles_.begin(), itEnd = indexesIntoFiles_.end(); it != itEnd; ++it) {
        if(*it && (*it)->containsItem(run, lumi, event)) {
          // We found it. Close the currently open file, and open the correct one.
          std::vector<FileCatalogItem>::const_iterator currentIter = fileIter_;
          fileIter_ = fileIterBegin_ + (it - indexesIntoFiles_.begin());
          if(fileIter_ != currentIter) {
            initFile(false);
          }
          // Now get the item from the correct file.
          assert(rootFile_);
          found = rootFile_->setEntryAtItem(run, lumi, event);
          assert(found);
          return true;
        }
      }
      return (fileNameHash != 0U && skipToItemInNewFile(run, lumi, event, fileNameHash)) || skipToItemInNewFile(run, lumi, event);
    }
    return true;
  }

  ProcessHistoryRegistry const&
  RootInputFileSequence::processHistoryRegistry() const {
    return input_.processHistoryRegistry();
  }

  ProcessHistoryRegistry&
  RootInputFileSequence::processHistoryRegistryForUpdate() {
    return input_.processHistoryRegistryForUpdate();
  }

  ProcessConfiguration const&
  RootInputFileSequence::processConfiguration() const {
    return input_.processConfiguration();
  }

  int
  RootInputFileSequence::remainingEvents() const {
    return input_.remainingEvents();
  }

  int
  RootInputFileSequence::remainingLuminosityBlocks() const {
    return input_.remainingLuminosityBlocks();
  }

  ProductRegistry &
  RootInputFileSequence::productRegistryUpdate() const{
    return input_.productRegistryUpdate();
  }

  std::shared_ptr<ProductRegistry const>
  RootInputFileSequence::productRegistry() const{
    return input_.productRegistry();
  }

  void
  RootInputFileSequence::dropUnwantedBranches_(std::vector<std::string> const& wantedBranches) {
    std::vector<std::string> rules;
    rules.reserve(wantedBranches.size() + 1);
    rules.emplace_back("drop *");
    for(std::string const& branch : wantedBranches) {
      rules.push_back("keep " + branch + "_*");
    }
    ParameterSet pset;
    pset.addUntrackedParameter("inputCommands", rules);
    productSelectorRules_ = ProductSelectorRules(pset, "inputCommands", "InputSource");
  }

  void
  RootInputFileSequence::skipEntries(unsigned int offset) {
    // offset is decremented by the number of events actually skipped.
    bool completed = rootFile_->skipEntries(offset);
    while(!completed) {
      ++fileIter_;
      if(fileIter_ == fileIterEnd_) {
        fileIter_ = fileIterBegin_;
      }
      initFile(false);
      assert(rootFile_);
      rootFile_->setAtEventEntry(IndexIntoFile::invalidEntry);
      completed = rootFile_->skipEntries(offset);
    }
  }

  bool
  RootInputFileSequence::readOneSequential(EventPrincipal& cache, size_t& fileNameHash) {
    skipBadFiles_ = false;
    if(firstFile_) {
      firstFile_ = false;
      if(fileIterEnd_ == fileIterBegin_) {
        throw Exception(errors::Configuration) << "RootInputFileSequence::readOneSequential(): no input files specified for secondary input source.\n";
      }
      if(fileIter_ != fileIterBegin_) {
        fileIter_ = fileIterBegin_;
        initFile(false);
      }
      assert(rootFile_);
      rootFile_->setAtEventEntry(IndexIntoFile::invalidEntry);
      skipEntries(initialNumberOfEventsToSkip_);
    }
    assert(rootFile_);
    rootFile_->nextEventEntry();
    bool found = rootFile_->readCurrentEvent(cache);
    if(!found) {
      ++fileIter_;
      if(fileIter_ == fileIterEnd_) {
        fileIter_ = fileIterBegin_;
      }
      initFile(false);
      assert(rootFile_);
      rootFile_->setAtEventEntry(IndexIntoFile::invalidEntry);
      return readOneSequential(cache, fileNameHash);
    }
    fileNameHash = lfnHash_;
    return true;
  }

  bool
  RootInputFileSequence::readOneSequentialWithID(EventPrincipal& cache, size_t& fileNameHash, LuminosityBlockID const& id) {
    skipBadFiles_ = false;
    if(firstFile_) {
      firstFile_ = false;
      if(fileIterEnd_ == fileIterBegin_) {
        throw Exception(errors::Configuration) << "RootInputFileSequence::readOneSequential(): no input files specified for secondary input source.\n";
      }
      if(fileIter_ != fileIterBegin_) {
        fileIter_ = fileIterBegin_;
        initFile(false);
      }
      assert(rootFile_);
      rootFile_->setAtEventEntry(IndexIntoFile::invalidEntry);
      int offset = initialNumberOfEventsToSkip_;
      while(offset > 0) {
        bool found = readOneSequentialWithID(cache, fileNameHash, id);
        if(!found) {
          return false;
        }
        --offset;
      }
    }
    assert(rootFile_);
    if(fileIter_ == fileIterEnd_ ||
        rootFile_->indexIntoFileIter().run() != id.run() ||
        rootFile_->indexIntoFileIter().lumi() != id.luminosityBlock()) {
      bool found = skipToItem(id.run(), id.luminosityBlock(), 0, 0, false);
      if(!found) {
        return false;
      }
    }
    assert(rootFile_);
    bool found = rootFile_->setEntryAtNextEventInLumi(id.run(), id.luminosityBlock());
    if(found) {
      found = rootFile_->readCurrentEvent(cache);
    }
    if(!found) {
      found = skipToItemInNewFile(id.run(), id.luminosityBlock(), 0);
      if(!found) {
        return false;
      }
      return readOneSequentialWithID(cache, fileNameHash, id);
    }
    fileNameHash = lfnHash_;
    return true;
  }

  void
  RootInputFileSequence::readOneSpecified(EventPrincipal& cache, size_t& fileNameHash, SecondaryEventIDAndFileInfo const& idx) {
    firstFile_ = false;
    if(fileIterEnd_ == fileIterBegin_) {
      throw Exception(errors::Configuration) << "RootInputFileSequence::readOneSpecified(): no input files specified for secondary input source.\n";
    }
    skipBadFiles_ = false;
    EventID const& id = idx.eventID();
    bool found = skipToItem(id.run(), id.luminosityBlock(), id.event(), idx.fileNameHash());
    if(!found) {
       throw Exception(errors::NotFound) <<
         "RootInputFileSequence::readOneSpecified(): Secondary Input files" <<
         " do not contain specified event:\n" << id << "\n";
    }
    assert(rootFile_);
    found = rootFile_->readCurrentEvent(cache);
    assert(found);
    fileNameHash = idx.fileNameHash();
    if(fileNameHash == 0U)  {
      fileNameHash = lfnHash_;
    }
  }

  void
  RootInputFileSequence::readOneRandom(EventPrincipal& cache, size_t& fileNameHash, CLHEP::HepRandomEngine* engine) {
    firstFile_ = false;
    if(fileIterEnd_ == fileIterBegin_) {
      throw Exception(errors::Configuration) << "RootInputFileSequence::readOneRandom(): no input files specified for secondary input source.\n";
    }
    assert(rootFile_);
    skipBadFiles_ = false;
    unsigned int currentSeqNumber = fileIter_ - fileIterBegin_;
    while(eventsRemainingInFile_ == 0) {

      fileIter_ = fileIterBegin_ + CLHEP::RandFlat::shootInt(engine, fileCatalogItems().size());
      unsigned int newSeqNumber = fileIter_ - fileIterBegin_;
      if(newSeqNumber != currentSeqNumber) {
        initFile(false);
        currentSeqNumber = newSeqNumber;
      }
      eventsRemainingInFile_ = rootFile_->eventTree().entries();
      if(eventsRemainingInFile_ == 0) {
        throw Exception(errors::NotFound) <<
           "RootInputFileSequence::readOneRandom(): Secondary Input file " << fileIter_->fileName() << " contains no events.\n";
      }
      rootFile_->setAtEventEntry(CLHEP::RandFlat::shootInt(engine, eventsRemainingInFile_) - 1);
    }
    rootFile_->nextEventEntry();

    bool found = rootFile_->readCurrentEvent(cache);
    if(!found) {
      rootFile_->setAtEventEntry(0);
      bool found = rootFile_->readCurrentEvent(cache);
      assert(found);
    }
    fileNameHash = lfnHash_;
    --eventsRemainingInFile_;
  }

  // bool RootFile::setEntryAtNextEventInLumi(RunNumber_t run, LuminosityBlockNumber_t lumi) {

  bool
  RootInputFileSequence::readOneRandomWithID(EventPrincipal& cache, size_t& fileNameHash, LuminosityBlockID const& id, CLHEP::HepRandomEngine* engine) {
    firstFile_ = false;
    if(fileIterEnd_ == fileIterBegin_) {
      throw Exception(errors::Configuration) << "RootInputFileSequence::readOneRandomWithID(): no input files specified for secondary input source.\n";
    }
    skipBadFiles_ = false;
    if(fileIter_ == fileIterEnd_ || !rootFile_ ||
        rootFile_->indexIntoFileIter().run() != id.run() ||
        rootFile_->indexIntoFileIter().lumi() != id.luminosityBlock()) {
      bool found = skipToItem(id.run(), id.luminosityBlock(), 0);
      if(!found) {
        return false;
      }
      int eventsInLumi = 0;
      assert(rootFile_);
      while(rootFile_->setEntryAtNextEventInLumi(id.run(), id.luminosityBlock())) ++eventsInLumi;
      found = skipToItem(id.run(), id.luminosityBlock(), 0);
      assert(found);
      int eventInLumi = CLHEP::RandFlat::shootInt(engine, eventsInLumi);
      for(int i = 0; i < eventInLumi; ++i) {
        bool found = rootFile_->setEntryAtNextEventInLumi(id.run(), id.luminosityBlock());
        assert(found);
      }
    }
    assert(rootFile_);
    bool found = rootFile_->setEntryAtNextEventInLumi(id.run(), id.luminosityBlock());
    if(found) {
      found = rootFile_->readCurrentEvent(cache);
    }
    if(!found) {
      bool found = rootFile_->setEntryAtItem(id.run(), id.luminosityBlock(), 0);
      if(!found) {
        return false;
      }
      return readOneRandomWithID(cache, fileNameHash, id, engine);
    }
    fileNameHash = lfnHash_;
    return true;
  }

  void
  RootInputFileSequence::fillDescription(ParameterSetDescription & desc) {
    desc.addUntracked<unsigned int>("skipEvents", 0U)
        ->setComment("Skip the first 'skipEvents' events that otherwise would have been processed.");
    desc.addUntracked<bool>("noEventSort", true)
        ->setComment("True:  Process runs, lumis and events in the order they appear in the file (but see notes 1 and 2).\n"
                     "False: Process runs, lumis and events in each file in numerical order (run#, lumi#, event#) (but see note 3).\n"
                     "Note 1: Events within the same lumi will always be processed contiguously.\n"
                     "Note 2: Lumis within the same run will always be processed contiguously.\n"
                     "Note 3: Any sorting occurs independently in each input file (no sorting across input files).");
    desc.addUntracked<bool>("skipBadFiles", false)
        ->setComment("True:  Ignore any missing or unopenable input file.\n"
                     "False: Throw exception if missing or unopenable input file.");
    desc.addUntracked<bool>("bypassVersionCheck", false)
        ->setComment("True:  Bypass release version check.\n"
                     "False: Throw exception if reading file in a release prior to the release in which the file was written.");
    desc.addUntracked<unsigned int>("cacheSize", roottree::defaultCacheSize)
        ->setComment("Size of ROOT TTree prefetch cache.  Affects performance.");
    desc.addUntracked<int>("treeMaxVirtualSize", -1)
        ->setComment("Size of ROOT TTree TBasket cache.  Affects performance.");
    desc.addUntracked<unsigned int>("setRunNumber", 0U)
        ->setComment("If non-zero, change number of first run to this number. Apply same offset to all runs.  Allowed only for simulation.");
    desc.addUntracked<bool>("dropDescendantsOfDroppedBranches", true)
        ->setComment("If True, also drop on input any descendent of any branch dropped on input.");
    std::string defaultString("permissive");
    desc.addUntracked<std::string>("branchesMustMatch", defaultString)
        ->setComment("'strict':     Branches in each input file must match those in the first file.\n"
                     "'permissive': Branches in each input file may be any subset of those in the first file.");
    desc.addUntracked<bool>("labelRawDataLikeMC", true)
        ->setComment("If True: replace module label for raw data to match MC. Also use 'LHC' as process.");

    ProductSelectorRules::fillDescription(desc, "inputCommands");
    EventSkipperByID::fillDescription(desc);
    DuplicateChecker::fillDescription(desc);
  }

  ProcessingController::ForwardState
  RootInputFileSequence::forwardState() const {
    if(rootFile_) {
      if(!rootFile_->wasLastEventJustRead()) {
        return ProcessingController::kEventsAheadInFile;
      }
      std::vector<FileCatalogItem>::const_iterator itr(fileIter_);
      if(itr != fileIterEnd_) ++itr;
      if(itr != fileIterEnd_) {
        return ProcessingController::kNextFileExists;
      }
      return ProcessingController::kAtLastEvent;
    }
    return ProcessingController::kUnknownForward;
  }

  ProcessingController::ReverseState
  RootInputFileSequence::reverseState() const {
    if(rootFile_) {
      if(!rootFile_->wasFirstEventJustRead()) {
        return ProcessingController::kEventsBackwardsInFile;
      }
      if(fileIter_ != fileIterBegin_) {
        return ProcessingController::kPreviousFileExists;
      }
      return ProcessingController::kAtFirstEvent;
    }
    return ProcessingController::kUnknownReverse;
  }

  void RootInputFileSequence::initAssociationsFromSecondary(std::set<BranchID> const& associationsFromSecondary) {
    for(auto const& branchID : associationsFromSecondary) {
      associationsFromSecondary_.push_back(branchID);
    }
    rootFile_->initAssociationsFromSecondary(associationsFromSecondary_);
  }
}