bool
WriteJobStatusKnownEventToUserLog( ClassAd *job_ad )
{
int cluster, proc;
WriteUserLog *ulog = InitializeUserLog( job_ad );
if ( ulog == NULL ) {
// User doesn't want a log
return true;
}
job_ad->LookupInteger( ATTR_CLUSTER_ID, cluster );
job_ad->LookupInteger( ATTR_PROC_ID, proc );
dprintf( D_FULLDEBUG,
"(%d.%d) Writing job status known record to user logfile\n",
cluster, proc );
JobStatusKnownEvent event;
int rc = ulog->writeEvent( &event, job_ad );
delete ulog;
if ( !rc ) {
dprintf( D_ALWAYS,
"(%d.%d) Unable to log ULOG_JOB_STATUS_KNOWN event\n",
cluster, proc );
return false;
}
return true;
}
// Initialize a UserLog object for a given job and return a pointer to
// the UserLog object created. This object can then be used to write
// events and must be deleted when you're done. This returns NULL if
// the user didn't want a UserLog, so you must check for NULL before
// using the pointer you get back.
WriteUserLog*
InitializeUserLog( ClassAd *job_ad )
{
int cluster, proc;
MyString userLogFile, dagmanNodeLog;
std::string gjid;
bool use_xml = false;
std::vector<const char*> logfiles;
if( getPathToUserLog(job_ad, userLogFile) ) {
logfiles.push_back(userLogFile.Value());
}
if( getPathToUserLog(job_ad, dagmanNodeLog, ATTR_DAGMAN_WORKFLOW_LOG) ) {
logfiles.push_back(dagmanNodeLog.Value());
}
if(logfiles.empty()) {
return NULL;
}
job_ad->LookupInteger( ATTR_CLUSTER_ID, cluster );
job_ad->LookupInteger( ATTR_PROC_ID, proc );
job_ad->LookupString( ATTR_GLOBAL_JOB_ID, gjid );
job_ad->LookupBool( ATTR_ULOG_USE_XML, use_xml );
WriteUserLog *ULog = new WriteUserLog();
ULog->initialize(logfiles, cluster, proc, 0, gjid.c_str());
ULog->setUseXML( use_xml );
return ULog;
}
bool
WriteHoldEventToUserLog( ClassAd *job_ad )
{
int cluster, proc;
WriteUserLog *ulog = InitializeUserLog( job_ad );
if ( ulog == NULL ) {
// User doesn't want a log
return true;
}
job_ad->LookupInteger( ATTR_CLUSTER_ID, cluster );
job_ad->LookupInteger( ATTR_PROC_ID, proc );
dprintf( D_FULLDEBUG,
"(%d.%d) Writing hold record to user logfile\n",
cluster, proc );
JobHeldEvent event;
event.initFromClassAd(job_ad);
int rc = ulog->writeEvent(&event,job_ad);
delete ulog;
if (!rc) {
dprintf( D_ALWAYS,
"(%d.%d) Unable to log ULOG_JOB_HELD event\n",
cluster, proc );
return false;
}
return true;
}
extern "C" void
initializeUserLog ()
{
std::string logfilename,dagmanLogName;
int use_xml;
std::vector<const char*> logfiles;
if ( getPathToUserLog(JobAd, logfilename) ) {
logfiles.push_back(logfilename.c_str());
dprintf(D_FULLDEBUG, "%s = %s\n", ATTR_ULOG_FILE, logfilename.c_str());
}
if ( getPathToUserLog(JobAd, dagmanLogName, ATTR_DAGMAN_WORKFLOW_LOG) ) {
logfiles.push_back(dagmanLogName.c_str());
dprintf(D_FULLDEBUG, "%s = %s\n", ATTR_DAGMAN_WORKFLOW_LOG,
dagmanLogName.c_str());
}
if(!logfiles.empty()) {
if ( !ULog.initialize (Proc->owner, NULL, logfiles,
Proc->id.cluster, Proc->id.proc, 0)) {
EXCEPT("Failed to initialize user log!");
} else {
ULog.setUseXML(JobAd->LookupBool(ATTR_ULOG_USE_XML, use_xml) && use_xml);
}
} else {
dprintf(D_FULLDEBUG, "no %s found and no %s found\n", ATTR_ULOG_FILE,
ATTR_DAGMAN_WORKFLOW_LOG);
}
}
extern "C" void
log_old_starter_shadow_suspend_event_hack (char *s1, char *s2)
{
const char *magic_suspend = "TISABH Starter: Suspended user job: ";
const char *magic_unsuspend = "TISABH Starter: Unsuspended user job.";
/* This should be bug enough to hold the two string params */
char buffer[BUFSIZ * 2 + 2];
int size_suspend, size_unsuspend;
size_suspend = strlen(magic_suspend);
size_unsuspend = strlen(magic_unsuspend);
sprintf(buffer, "%s%s", s1, s2);
/* depending on if it is a suspend or unsuspend event, do something
about it. */
if (strncmp(buffer, magic_suspend, size_suspend) == 0)
{
/* matched a suspend event */
JobSuspendedEvent event;
sscanf(buffer,"TISABH Starter: Suspended user job: %d",&event.num_pids);
if (!ULog.writeEvent (&event))
{
dprintf (D_ALWAYS, "Unable to log ULOG_JOB_SUSPENDED event\n");
}
record_suspension_hack(ULOG_JOB_SUSPENDED);
return;
}
if (strncmp(buffer, magic_unsuspend, size_unsuspend) == 0)
{
/* matched an unsuspend event */
JobUnsuspendedEvent event;
if (!ULog.writeEvent (&event))
{
dprintf (D_ALWAYS, "Unable to log ULOG_JOB_UNSUSPENDED event\n");
}
record_suspension_hack(ULOG_JOB_UNSUSPENDED);
return;
}
/* otherwise, do nothing */
}
extern "C" void
log_except (const char *msg)
{
check_execute_event();
// log shadow exception event
ShadowExceptionEvent event;
if(!msg) msg = "";
snprintf(event.message, sizeof(event.message), "%s", msg);
event.message[sizeof(event.message)-1] = '\0';
// we want to log the events from the perspective of the
// user job, so if the shadow *sent* the bytes, then that
// means the user job *received* the bytes
event.recvd_bytes = BytesSent;
event.sent_bytes = BytesRecvd;
if (syscall_sock) {
event.recvd_bytes += syscall_sock->get_bytes_sent();
event.sent_bytes += syscall_sock->get_bytes_recvd();
}
if (!ULog.writeEvent (&event))
{
dprintf (D_ALWAYS, "Unable to log ULOG_SHADOW_EXCEPTION event\n");
}
}
bool writePreSkipEvent( CondorID& condorID, Job* job, const char* DAGNodeName,
const char* directory, const char *logFile )
{
TmpDir tmpDir;
MyString errMsg;
if ( !tmpDir.Cd2TmpDir( directory, errMsg ) ) {
debug_printf( DEBUG_QUIET,
"Could not change to node directory %s: %s\n",
directory, errMsg.Value() );
return false;
}
// Special HTCondorID for NOOP jobs -- actually indexed by
// otherwise-unused subprocID.
condorID._cluster = 0;
condorID._proc = Job::NOOP_NODE_PROCID;
condorID._subproc = 1+get_fake_condorID();
// Increment this value
set_fake_condorID(condorID._subproc);
if( job ) {
job->SetCondorID( condorID );
}
WriteUserLog ulog;
ulog.setEnableGlobalLog( false );
ulog.setUseXML( false );
ulog.initialize( std::vector<const char*>(1,logFile), condorID._cluster,
condorID._proc, condorID._subproc, NULL );
PreSkipEvent pEvent;
pEvent.cluster = condorID._cluster;
pEvent.proc = condorID._proc;
pEvent.subproc = condorID._subproc;
MyString pEventNotes("DAG Node: " );
pEventNotes += DAGNodeName;
// skipEventLogNotes gets deleted in PreSkipEvent destructor.
pEvent.skipEventLogNotes = strnewp( pEventNotes.Value() );
if ( !ulog.writeEvent( &pEvent ) ) {
EXCEPT( "Error: writing PRESKIP event failed!" );
return false;
}
return true;
}
// The GlobusSubmitEvent is now deprecated and should be removed at
// some point in the future (6.9?).
bool
WriteGlobusSubmitEventToUserLog( ClassAd *job_ad )
{
int cluster, proc;
std::string contact;
WriteUserLog *ulog = InitializeUserLog( job_ad );
if ( ulog == NULL ) {
// User doesn't want a log
return true;
}
job_ad->LookupInteger( ATTR_CLUSTER_ID, cluster );
job_ad->LookupInteger( ATTR_PROC_ID, proc );
dprintf( D_FULLDEBUG,
"(%d.%d) Writing globus submit record to user logfile\n",
cluster, proc );
GlobusSubmitEvent event;
job_ad->LookupString( ATTR_GRID_RESOURCE, contact );
Tokenize( contact );
GetNextToken( " ", false );
event.rmContact = strnewp(GetNextToken( " ", false ));
job_ad->LookupString( ATTR_GRID_JOB_ID, contact );
Tokenize( contact );
if ( strcasecmp( GetNextToken( " ", false ), "gt2" ) == 0 ) {
GetNextToken( " ", false );
}
event.jmContact = strnewp(GetNextToken( " ", false ));
event.restartableJM = true;
int rc = ulog->writeEvent(&event,job_ad);
delete ulog;
if (!rc) {
dprintf( D_ALWAYS,
"(%d.%d) Unable to log ULOG_GLOBUS_SUBMIT event\n",
cluster, proc );
return false;
}
return true;
}
int writeCheckpointedEvent()
{
CheckpointedEvent checkpoint;
if ( !logFile.writeEvent(&checkpoint) ) {
printf("Complain about bad checkpoint write\n");
exit(1);
}
return(0);
}
int writeExecutableErrorEvent()
{
ExecutableErrorEvent executeerror;
executeerror.errType = CONDOR_EVENT_BAD_LINK;
if ( !logFile.writeEvent(&executeerror) ) {
printf("Complain about bad executeerror write\n");
exit(1);
}
return(0);
}
int writeJobReleasedEvent()
{
JobReleasedEvent jobreleasedevent;
jobreleasedevent.setReason("MessinWithYou");
if ( !logFile.writeEvent(&jobreleasedevent) ) {
printf("Complain about bad jobreleasedevent write\n");
exit(1);
}
return(0);
}
int writeJobUnsuspendedEvent()
{
JobUnsuspendedEvent jobunsuspendevent;
//jobunsuspendevent.num_pids = 99;
if ( !logFile.writeEvent(&jobunsuspendevent) ) {
printf("Complain about bad jobunsuspendevent write\n");
exit(1);
}
return(0);
}
int writeJobImageSizeEvent()
{
JobImageSizeEvent jobimagesizeevent;
jobimagesizeevent.image_size_kb = 128;
if ( !logFile.writeEvent(&jobimagesizeevent) ) {
printf("Complain about bad jobimagesizeevent write\n");
exit(1);
}
return(0);
}
int writeGlobusResourceDownEvent()
{
GlobusResourceDownEvent globusresourcedownevent;
globusresourcedownevent.rmContact = strdup("ResourceDown");;
if ( !logFile.writeEvent(&globusresourcedownevent) ) {
printf("Complain about bad globusresourcedownevent write\n");
exit(1);
}
return(0);
}
int writeGlobusSubmitFailedEvent()
{
GlobusSubmitFailedEvent globussubmitfailedevent;
globussubmitfailedevent.reason = strdup("Cause it could");;
if ( !logFile.writeEvent(&globussubmitfailedevent) ) {
printf("Complain about bad globussubmitfailedevent write\n");
exit(1);
}
return(0);
}
int writeJobAbortedEvent()
{
JobAbortedEvent jobabort;
jobabort.setReason("cause I said so!");
if ( !logFile.writeEvent(&jobabort) ) {
printf("Complain about bad jobabort write\n");
exit(1);
}
return(0);
}
int writeExecuteEvent()
{
ExecuteEvent execute;
execute.setExecuteHost("<128.105.165.12:32779>");
if ( !logFile.writeEvent(&execute) ) {
printf("Complain about bad execute write\n");
exit(1);
}
return(0);
}
// The GlobusResourceDownEvent is now deprecated and should be removed at
// some point in the future (6.9?).
bool
WriteGlobusResourceDownEventToUserLog( ClassAd *job_ad )
{
int cluster, proc;
std::string contact;
WriteUserLog *ulog = InitializeUserLog( job_ad );
if ( ulog == NULL ) {
// User doesn't want a log
return true;
}
job_ad->LookupInteger( ATTR_CLUSTER_ID, cluster );
job_ad->LookupInteger( ATTR_PROC_ID, proc );
dprintf( D_FULLDEBUG,
"(%d.%d) Writing globus down record to user logfile\n",
cluster, proc );
GlobusResourceDownEvent event;
job_ad->LookupString( ATTR_GRID_RESOURCE, contact );
if ( contact.empty() ) {
// Not a Globus job, don't log the event
delete ulog;
return true;
}
Tokenize( contact );
GetNextToken( " ", false );
event.rmContact = strnewp(GetNextToken( " ", false ));
int rc = ulog->writeEvent(&event,job_ad);
delete ulog;
if (!rc) {
dprintf( D_ALWAYS,
"(%d.%d) Unable to log ULOG_GLOBUS_RESOURCE_DOWN event\n",
cluster, proc );
return false;
}
return true;
}
bool
WriteEvictEventToUserLog( ClassAd *job_ad )
{
int cluster, proc;
WriteUserLog *ulog = InitializeUserLog( job_ad );
if ( ulog == NULL ) {
// User doesn't want a log
return true;
}
job_ad->LookupInteger( ATTR_CLUSTER_ID, cluster );
job_ad->LookupInteger( ATTR_PROC_ID, proc );
dprintf( D_FULLDEBUG,
"(%d.%d) Writing evict record to user logfile\n",
cluster, proc );
JobEvictedEvent event;
struct rusage r;
memset( &r, 0, sizeof( struct rusage ) );
#if !defined(WIN32)
event.run_local_rusage = r;
event.run_remote_rusage = r;
#endif /* WIN32 */
event.sent_bytes = 0;
event.recvd_bytes = 0;
event.checkpointed = false;
int rc = ulog->writeEvent(&event,job_ad);
delete ulog;
if (!rc) {
dprintf( D_ALWAYS,
"(%d.%d) Unable to log ULOG_JOB_EVICTED event\n",
cluster, proc );
return false;
}
return true;
}
int writeNodeExecuteEvent()
{
NodeExecuteEvent nodeexecuteevent;
nodeexecuteevent.node = 49;
nodeexecuteevent.setExecuteHost("<128.105.165.12:32779>");
if ( !logFile.writeEvent(&nodeexecuteevent) ) {
printf("Complain about bad nodeexecuteevent write\n");
exit(1);
}
return(0);
}
int writeJobEvictedEvent()
{
JobEvictedEvent jobevicted;
jobevicted.setReason("It misbehaved!");
jobevicted.setCoreFile("corefile");
if ( !logFile.writeEvent(&jobevicted) ) {
printf("Complain about bad jobevicted write\n");
exit(1);
}
return(0);
}
int writePostScriptTerminatedEvent()
{
PostScriptTerminatedEvent postscriptterminated;
postscriptterminated.normal = false;
postscriptterminated.signalNumber = 9;
postscriptterminated.returnValue = 4;
if ( !logFile.writeEvent(&postscriptterminated) ) {
printf("Complain about bad postscriptterminated write\n");
exit(1);
}
return(0);
}
int writeGlobusSubmitEvent()
{
GlobusSubmitEvent globussubmitevent;
globussubmitevent.rmContact = strdup("ResourceManager");;
globussubmitevent.jmContact = strdup("JobManager");;
globussubmitevent.restartableJM = true;
if ( !logFile.writeEvent(&globussubmitevent) ) {
printf("Complain about bad globussubmitevent write\n");
exit(1);
}
return(0);
}
int writeJobHeldEvent()
{
JobHeldEvent jobheldevent;
jobheldevent.setReason("CauseWeCan");
jobheldevent.setReasonCode(404);
jobheldevent.setReasonSubCode(0xff);
if ( !logFile.writeEvent(&jobheldevent) ) {
printf("Complain about bad jobheldevent write\n");
exit(1);
}
return(0);
}
int writeSubmitEvent()
{
SubmitEvent submit;
submit.setSubmitHost("<128.105.165.12:32779>");
submit.submitEventLogNotes = strdup("DAGMan info");
submit.submitEventUserNotes = strdup("User info");
if ( !logFile.writeEvent(&submit) ) {
printf("Complain about bad submit write\n");
exit(1);
}
return(0);
}
bool
WriteGridResourceDownEventToUserLog( ClassAd *job_ad )
{
int cluster, proc;
std::string contact;
WriteUserLog *ulog = InitializeUserLog( job_ad );
if ( ulog == NULL ) {
// User doesn't want a log
return true;
}
job_ad->LookupInteger( ATTR_CLUSTER_ID, cluster );
job_ad->LookupInteger( ATTR_PROC_ID, proc );
dprintf( D_FULLDEBUG,
"(%d.%d) Writing grid source down record to user logfile\n",
cluster, proc );
GridResourceDownEvent event;
job_ad->LookupString( ATTR_GRID_RESOURCE, contact );
if ( contact.empty() ) {
dprintf( D_ALWAYS,
"(%d.%d) %s attribute missing in job ad\n",
cluster, proc, ATTR_GRID_RESOURCE );
}
event.resourceName = strnewp( contact.c_str() );
int rc = ulog->writeEvent(&event,job_ad);
delete ulog;
if (!rc) {
dprintf( D_ALWAYS,
"(%d.%d) Unable to log ULOG_GRID_RESOURCE_DOWN event\n",
cluster, proc );
return false;
}
return true;
}
int writeRemoteErrorEvent()
{
RemoteErrorEvent remoteerror;
remoteerror.setExecuteHost("<128.105.165.12:32779>");
remoteerror.setDaemonName("<write job log test>");
remoteerror.setErrorText("this is the write test error string");
remoteerror.setCriticalError(true);
if ( !logFile.writeEvent(&remoteerror) ) {
printf("Complain about bad remoteerror write\n");
exit(1);
}
return(0);
}
extern "C" void
log_image_size (int size)
{
check_execute_event();
// log the event
JobImageSizeEvent event;
event.image_size_kb = size;
if (!ULog.writeEvent (&event))
{
dprintf (D_ALWAYS, "Unable to log ULOG_IMAGE_SIZE event\n");
}
}
extern "C" void
log_checkpoint (struct rusage *localr, struct rusage *remoter)
{
check_execute_event();
CheckpointedEvent event;
event.run_local_rusage = *localr;
event.run_remote_rusage = *remoter;
if (!ULog.writeEvent (&event))
{
dprintf (D_ALWAYS, "Could not log ULOG_CHECKPOINTED event\n");
}
}
int writeShadowExceptionEvent()
{
ShadowExceptionEvent shadowexceptionevent;
shadowexceptionevent.sent_bytes = 4096;
shadowexceptionevent.recvd_bytes = 4096;
shadowexceptionevent.message[0] = '\0';
strncat(shadowexceptionevent.message,"shadow message", 15);
if ( !logFile.writeEvent(&shadowexceptionevent) ) {
printf("Complain about bad shadowexceptionevent write\n");
exit(1);
}
return(0);
}
