|
|
|
#include <Vulture.h>
Public Member Functions | |
| int | hasStarted () |
| int | hasStopped () |
| pid_t | kill () |
| pid_t | makeProcess () |
| void | retrieve_corefile (char *, char *, uint64_t) |
| int | start (std::string, int=0) |
| int | stop () |
| Vulture (bool) | |
| virtual | ~Vulture () |
Private Member Functions | |
| void | analyze () |
| bool | control (toolbox::task::WorkLoop *) |
| bool | prowling (toolbox::task::WorkLoop *) |
| void | startProwling () |
Private Attributes | |
| toolbox::task::ActionSignature * | asCtrl_ |
| toolbox::task::ActionSignature * | asProwl_ |
| std::vector< std::string > | currentCoreList_ |
| bool | handicapped_ |
| std::string | iDieUrl_ |
| time_t | lastUpdate_ |
| MasterQueue * | mq_ |
| unsigned int | newCores_ |
| CurlPoster * | poster_ |
| bool | prowling_ |
| bool | running_ |
| SlaveQueue * | sq_ |
| int | started_ |
| int | stopped_ |
| DIR * | tmp_ |
| bool | updateMode_ |
| pid_t | vulturePid_ |
| toolbox::task::WorkLoop * | wlCtrl_ |
| toolbox::task::WorkLoop * | wlProwl_ |
Static Private Attributes | |
| static const std::string | FS = "/tmp" |
| static const int | vulture_queue_offset = 400 |
| evf::Vulture::Vulture | ( | bool | push | ) |
Definition at line 117 of file Vulture.cc.
: wlCtrl_(0) , asCtrl_(0) , running_(false) , wlProwl_(0) , asProwl_(0) , prowling_(false) , iDieUrl_("") , updateMode_(push) , vulturePid_(0) , tmp_(0) , newCores_(0) , poster_(0) , mq_(new MasterQueue(vulture_queue_offset)) , sq_(0) // this is only defined in the forked process , started_(-1) , stopped_(-1) , handicapped_(false) { // create command file for gdb, if not already there std::ifstream vulture("/tmp/vulture.cmd"); if (!vulture.good()) { FILE *outf = fopen("/tmp/vulture.cmd","w"); fprintf(outf,"where\n"); fclose(outf); } }
| evf::Vulture::~Vulture | ( | ) | [virtual] |
| void evf::Vulture::analyze | ( | void | ) | [private] |
Definition at line 398 of file Vulture.cc.
References edmPickEvents::command, gather_cfg::cout, currentCoreList_, FS, i, newCores_, AlCaHLTBitMon_ParallelJobs::p, evf::utils::pid, poster_, evf::CurlPoster::postString(), alignCSCRings::s, and svgfig::stack.
Referenced by prowling().
{
// do a first analysis of the coredump
if(newCores_==0) return;
for(unsigned int i = currentCoreList_.size()-newCores_;
i < currentCoreList_.size();
i++){
std::string command = "gdb /opt/xdaq/bin/xdaq.exe -batch -x /tmp/vulture.cmd -c /tmp/";
std::string cmdout;
command += currentCoreList_[i];
std::string filePathAndName = FS + "/";
filePathAndName += currentCoreList_[i];
std::string pid =
currentCoreList_[i].substr(currentCoreList_[i].find_first_of(".")+1,
currentCoreList_[i].length());
FILE *ps = popen(command.c_str(),"r");
size_t s = 256;
char *p=new char[s];
bool filter = false;
while(getline(&p,&s,ps) != -1){
if(strncmp("Core",p,4)==0) filter = true;
if(filter)cmdout += p;
}
delete[] p;
pclose(ps);
int errsv = 0;
int rch = chmod(filePathAndName.c_str(),0777);
if(rch != 0){
errsv = errno;
std::cout << "ERROR: couldn't change corefile access privileges -"
<< strerror(errsv)<< std::endl;
}
unsigned int ipid = (unsigned int)atoi(pid.c_str());
poster_->postString(cmdout.c_str(),cmdout.length(),ipid, CurlPoster::stack);
}
}
| bool evf::Vulture::control | ( | toolbox::task::WorkLoop * | wl | ) | [private] |
Definition at line 295 of file Vulture.cc.
References evf::CurlPoster::check(), gather_cfg::cout, alignCSCRings::e, lumiQueryAPI::msg, MSQM_MESSAGE_TYPE_NOP, MSQM_VULTURE_TYPE_STA, MSQM_VULTURE_TYPE_STP, MSQS_VULTURE_TYPE_ACK, evf::SlaveQueue::post(), poster_, prowling_, evf::SlaveQueue::rcv(), evf::vulture_start_message::run_, sq_, startProwling(), and evf::vulture_start_message::url_.
Referenced by makeProcess().
{
MsgBuf msg;
unsigned long mtype = MSQM_MESSAGE_TYPE_NOP;
try{mtype = sq_->rcv(msg);}catch(evf::Exception &e){
std::cout << "Vulture::exception on msgrcv for control, bailing out of control workloop - good bye" << std::endl;
return false;
}
mtype = msg->mtype;
switch(mtype){
case MSQM_VULTURE_TYPE_STA:
{
vulture_start_message *sta = (vulture_start_message*)msg->mtext;
if(poster_ == 0) poster_ = new CurlPoster(sta->url_);
if(poster_->check(sta->run_)){
try{
startProwling();
MsgBuf msg1(0,MSQS_VULTURE_TYPE_ACK) ;
sq_->post(msg1);
}
catch(evf::Exception &e)
{
std::cout << "Vulture::start - exception in starting prowling workloop " << e.what() << std::endl;
//@EM ToDo generate some message here
}
}else{
std::cout << "Vulture::start - could not contact iDie - chech Url - will not start prowling loop" << std::endl;
prowling_ = false;
}
break;
}
case MSQM_VULTURE_TYPE_STP:
{
prowling_ = false;
break;
}
default:
{
// do nothing @EM ToDo generate an appropriate error message
}
}
return true;
}
| int evf::Vulture::hasStarted | ( | ) |
Definition at line 215 of file Vulture.cc.
References alignCSCRings::e, MAX_MSG_SIZE, mq_, MSQS_VULTURE_TYPE_ACK, evf::MasterQueue::rcvNonBlocking(), and started_.
Referenced by evf::FUEventProcessor::updater().
{
if(started_<0){
MsgBuf msg2(MAX_MSG_SIZE,MSQS_VULTURE_TYPE_ACK);
try{
mq_->rcvNonBlocking(msg2);
started_ = 0;
}
catch(evf::Exception &e){
}
} else {started_ = 1;}
return started_;
}
| int evf::Vulture::hasStopped | ( | ) |
Definition at line 228 of file Vulture.cc.
References alignCSCRings::e, MAX_MSG_SIZE, mq_, MSQS_VULTURE_TYPE_ACK, evf::MasterQueue::rcvNonBlocking(), and stopped_.
{
if(stopped_<0){
MsgBuf msg2(MAX_MSG_SIZE,MSQS_VULTURE_TYPE_ACK);
try{
mq_->rcvNonBlocking(msg2);
stopped_ = 0;
}
catch(evf::Exception &e){
}
} else {stopped_ = 1;}
return stopped_;
}
| pid_t evf::Vulture::kill | ( | ) |
Definition at line 263 of file Vulture.cc.
References vulturePid_.
{
::kill (vulturePid_, SIGKILL);
int sl;
pid_t killedOrNot = waitpid(vulturePid_,&sl,WNOHANG);
vulturePid_ = 0;
return killedOrNot;
}
| pid_t evf::Vulture::makeProcess | ( | ) |
Definition at line 154 of file Vulture.cc.
References asCtrl_, control(), gather_cfg::cout, alignCSCRings::e, Exception, FS, handicapped_, sq_, summarizeEdmComparisonLogfiles::success, tmp_, vulture_queue_offset, vulturePid_, and wlCtrl_.
Referenced by evf::FUEventProcessor::configuring().
{
pid_t retval = fork();
if(retval==0){ // we are in the forked process
int success = -1;
// #ifdef linux
// success = prctl( PR_SET_DUMPABLE, 0 );
// #endif
if(success != 0){
std::cout << "Vulture::could not set process undumpable" << std::endl;
handicapped_ = true;
}
#ifdef linux
success = prctl( PR_SET_PDEATHSIG, SIGKILL );
#endif
if(success != 0){
std::cout << "Vulture::could not set process death signal" << std::endl;
handicapped_ = true;
}
tmp_ = opendir(FS.c_str());
#ifdef linux
success = prctl ( PR_SET_NAME , "vulture");
#endif
if(success != 0){
std::cout << "Vulture::could not set process name" << std::endl;
handicapped_ = true;
}
try{
pt::PeerTransport * ptr =
pt::getPeerTransportAgent()->getPeerTransport("https","soap",pt::Receiver);
delete ptr;
}
catch (pt::exception::PeerTransportNotFound & e ){
//do nothing here since we don't know what to do... ?
}
// freopen("/dev/null","w",stderr);
sq_ = new SlaveQueue(vulture_queue_offset);
// start the ctrl workloop
try {
wlCtrl_=
toolbox::task::getWorkLoopFactory()->getWorkLoop("Ctrll",
"waiting");
if (!wlCtrl_->isActive()) wlCtrl_->activate();
asCtrl_ = toolbox::task::bind(this,&Vulture::control,
"Ctrl");
wlCtrl_->submit(asCtrl_);
}
catch (xcept::Exception& e) {
std::cout << "Vulture:constructor - could not start workloop 'Ctrl' for process " << retval << std::endl;
}
}
else{
vulturePid_ = retval;
}
return retval;
}
| bool evf::Vulture::prowling | ( | toolbox::task::WorkLoop * | wl | ) | [private] |
Definition at line 343 of file Vulture.cc.
References analyze(), gather_cfg::cout, currentCoreList_, alignCSCRings::e, lastUpdate_, newCores_, cmsPerfSuiteHarvest::now, poster_, evf::CurlPoster::postString(), prowling_, stor::utils::sleep(), evf::CurlPoster::stack, testRegression::stat, and tmp_.
Referenced by startProwling().
{
if(!prowling_){
char messageDie[5];
sprintf(messageDie,"Dead");
if(poster_==0){
std::cout << "Vulture: asked to stop prowling but no poster "
<< std::endl;
return false;
}
try{
poster_->postString(messageDie,5,0,CurlPoster::stack);
}
catch(evf::Exception &e){
//do nothing just swallow the exception
}
std::cout << "Received STOP message, going to delete poster " << std::endl;
// delete poster_;
// poster_=0;
return false;
}
newCores_ = 0;
struct stat filestat;
timeval now;
gettimeofday(&now,0);
// examine /tmp looking for new coredumps
dirent *dirp;
while((dirp = readdir(tmp_))!=0){
if(strncmp(dirp->d_name,"core",4)==0){
stat(dirp->d_name,&filestat);
if(filestat.st_mtime > lastUpdate_){
currentCoreList_.push_back(dirp->d_name);
newCores_++;
}
}
}
rewinddir(tmp_);
lastUpdate_ = now.tv_sec;
try{
analyze();
}
catch(evf::Exception &e){
std::cout << "Vulture cannot send to iDie server, bail out " << std::endl;
return false;
}
::sleep(60);
return true;
}
| void evf::Vulture::retrieve_corefile | ( | char * | , |
| char * | , | ||
| uint64_t | |||
| ) |
| pid_t evf::Vulture::start | ( | std::string | url, |
| int | run = 0 |
||
| ) |
Definition at line 241 of file Vulture.cc.
References mq_, MSQM_VULTURE_TYPE_STA, evf::MasterQueue::post(), DTTTrigCorrFirst::run, evf::vulture_start_message::run_, stopped_, evf::vulture_start_message::url_, and vulturePid_.
Referenced by evf::FUEventProcessor::enabling().
{
//communicate start-of-run to Vulture
vulture_start_message stamsg;
strcpy(stamsg.url_,url.c_str());
stamsg.run_ = run;
MsgBuf msg1(sizeof(vulture_start_message),MSQM_VULTURE_TYPE_STA);
memcpy(msg1->mtext,&stamsg,sizeof(vulture_start_message));
mq_->post(msg1);
stopped_ = -1;
return vulturePid_;
}
| void evf::Vulture::startProwling | ( | ) | [private] |
Definition at line 272 of file Vulture.cc.
References asProwl_, alignCSCRings::e, Exception, lastUpdate_, lumiQueryAPI::msg, cmsPerfSuiteHarvest::now, prowling(), prowling_, and wlProwl_.
Referenced by control().
{
timeval now;
gettimeofday(&now,0);
lastUpdate_ = now.tv_sec;
prowling_ = true;
try {
wlProwl_=
toolbox::task::getWorkLoopFactory()->getWorkLoop("Prowl",
"waiting");
if (!wlProwl_->isActive()) wlProwl_->activate();
asProwl_ = toolbox::task::bind(this,&Vulture::prowling,
"Prowl");
wlProwl_->submit(asProwl_);
}
catch (xcept::Exception& e) {
std::string msg = "Failed to start workloop 'Prowl'.";
XCEPT_RETHROW(evf::Exception,msg,e);
}
}
| pid_t evf::Vulture::stop | ( | ) |
Definition at line 254 of file Vulture.cc.
References mq_, MSQM_VULTURE_TYPE_STP, NUMERIC_MESSAGE_SIZE, evf::MasterQueue::post(), started_, and vulturePid_.
Referenced by evf::FUEventProcessor::stopping().
{
MsgBuf msg1(NUMERIC_MESSAGE_SIZE,MSQM_VULTURE_TYPE_STP);
mq_->post(msg1);
started_ = -1;
return vulturePid_;
}
toolbox::task::ActionSignature* evf::Vulture::asCtrl_ [private] |
Definition at line 61 of file Vulture.h.
Referenced by makeProcess().
toolbox::task::ActionSignature* evf::Vulture::asProwl_ [private] |
Definition at line 64 of file Vulture.h.
Referenced by startProwling().
std::vector<std::string> evf::Vulture::currentCoreList_ [private] |
Definition at line 70 of file Vulture.h.
Referenced by analyze(), and prowling().
const std::string evf::Vulture::FS = "/tmp" [static, private] |
Definition at line 59 of file Vulture.h.
Referenced by analyze(), and makeProcess().
bool evf::Vulture::handicapped_ [private] |
Definition at line 78 of file Vulture.h.
Referenced by makeProcess().
std::string evf::Vulture::iDieUrl_ [private] |
time_t evf::Vulture::lastUpdate_ [private] |
Definition at line 71 of file Vulture.h.
Referenced by prowling(), and startProwling().
MasterQueue* evf::Vulture::mq_ [private] |
Definition at line 74 of file Vulture.h.
Referenced by hasStarted(), hasStopped(), start(), stop(), and ~Vulture().
unsigned int evf::Vulture::newCores_ [private] |
Definition at line 72 of file Vulture.h.
Referenced by analyze(), and prowling().
CurlPoster* evf::Vulture::poster_ [private] |
Definition at line 73 of file Vulture.h.
Referenced by analyze(), control(), prowling(), and ~Vulture().
bool evf::Vulture::prowling_ [private] |
Definition at line 65 of file Vulture.h.
Referenced by control(), prowling(), and startProwling().
bool evf::Vulture::running_ [private] |
SlaveQueue* evf::Vulture::sq_ [private] |
Definition at line 75 of file Vulture.h.
Referenced by control(), makeProcess(), and ~Vulture().
int evf::Vulture::started_ [private] |
Definition at line 76 of file Vulture.h.
Referenced by hasStarted(), and stop().
int evf::Vulture::stopped_ [private] |
Definition at line 77 of file Vulture.h.
Referenced by hasStopped(), and start().
DIR* evf::Vulture::tmp_ [private] |
Definition at line 69 of file Vulture.h.
Referenced by makeProcess(), and prowling().
bool evf::Vulture::updateMode_ [private] |
const int evf::Vulture::vulture_queue_offset = 400 [static, private] |
Definition at line 53 of file Vulture.h.
Referenced by makeProcess().
pid_t evf::Vulture::vulturePid_ [private] |
toolbox::task::WorkLoop* evf::Vulture::wlCtrl_ [private] |
Definition at line 60 of file Vulture.h.
Referenced by makeProcess().
toolbox::task::WorkLoop* evf::Vulture::wlProwl_ [private] |
Definition at line 63 of file Vulture.h.
Referenced by startProwling().
1.7.3