int
InitJobAd(int cluster, int proc)
{
if (!JobAd) { // just get the job ad from the schedd once
if ( IpcFile ) {
// get the jobad file a file
priv_state priv = set_condor_priv();
bool is_stdin = false;
FILE* ipc_fp = NULL;
if( IpcFile[0] == '-' && IpcFile[1] == '\0' ) {
ipc_fp = stdin;
is_stdin = true;
} else {
ipc_fp = safe_fopen_wrapper(IpcFile,"r");
}
dprintf( D_FULLDEBUG, "Reading job ClassAd from %s\n",
is_stdin ? "STDIN" : IpcFile );
int isEOF = 0;
int isError = 0;
int isEmpty = 0;
if ( ipc_fp ) {
JobAd = new ClassAd(ipc_fp,"***",isEOF,isError,isEmpty);
if( ! is_stdin ) {
fclose( ipc_fp );
unlink( IpcFile );
}
}
set_priv(priv);
// if constructor failed, remove the JobAd, and
// fallback on getting it via sockets.
if (isError != 0 ) {
if (JobAd) delete JobAd;
JobAd = NULL;
}
if (JobAd) {
// we're done
checkForDebugging( JobAd );
return 0;
}
// if we made it here, we wanted to get the job ad
// via the file, but failed
dprintf(D_FULLDEBUG,
"Failed to read job ad file %s, using socket\n",
IpcFile);
}
if (!ConnectQ(schedd, SHADOW_QMGMT_TIMEOUT, true)) {
EXCEPT("Failed to connect to schedd!");
}
JobAd = GetJobAd( cluster, proc );
DisconnectQ(NULL);
checkForDebugging( JobAd );
}
if (!JobAd) {
EXCEPT( "failed to get job ad" );
}
return 0;
}
void
HoldJob( const char* long_reason, const char* short_reason, int reason_code,
int reason_subcode )
{
char subject[ BUFSIZ ];
FILE *mailer;
sprintf( subject, "Condor Job %d.%d put on hold\n",
Proc->id.cluster, Proc->id.proc );
if( ! JobAd ) {
dprintf( D_ALWAYS, "In HoldJob() w/ NULL JobAd!\n" );
exit( JOB_SHOULD_HOLD );
}
ExitReason = JOB_SHOULD_HOLD;
if ( !ConnectQ(schedd, SHADOW_QMGMT_TIMEOUT) ) {
dprintf( D_ALWAYS, "Failed to connect to schedd!\n" );
}
SetAttributeString( Proc->id.cluster, Proc->id.proc, ATTR_HOLD_REASON,
short_reason );
SetAttributeInt( Proc->id.cluster, Proc->id.proc, ATTR_HOLD_REASON_CODE,
reason_code );
SetAttributeInt( Proc->id.cluster, Proc->id.proc, ATTR_HOLD_REASON_SUBCODE,
reason_subcode );
if ( !DisconnectQ(0) ) {
dprintf( D_ALWAYS, "Failed to commit updated job queue status!\n" );
}
mailer = email_user_open(JobAd, subject);
if( ! mailer ) {
// User didn't want email, so just exit now with the right
// value so the schedd actually puts the job on hold.
dprintf( D_ALWAYS, "Job going into Hold state.\n");
dprintf( D_ALWAYS, "********** Shadow Exiting(%d) **********\n",
JOB_SHOULD_HOLD);
exit( JOB_SHOULD_HOLD );
}
fprintf( mailer, "Your condor job " );
if( Proc->args_v1or2[0] ) {
ArgList args;
MyString args_string;
args.AppendArgsV1or2Raw(Proc->args_v1or2[0],NULL);
args.GetArgsStringForDisplay(&args_string);
fprintf( mailer, "%s %s ", Proc->cmd[0], args_string.Value() );
} else {
fprintf( mailer, "%s ", Proc->cmd[0] );
}
fprintf( mailer, "\nis being put on hold.\n\n" );
fprintf( mailer, "%s", long_reason );
email_close(mailer);
// Now that the user knows why, exit with the right code.
dprintf( D_ALWAYS, "Job going into Hold state.\n");
dprintf( D_ALWAYS, "********** Shadow Exiting(%d) **********\n",
JOB_SHOULD_HOLD);
exit( JOB_SHOULD_HOLD );
}
bool ActualScheddQ::disconnect(bool commit_transaction, CondorError & errstack) {
bool rval = false;
if (qmgr) {
rval = DisconnectQ(qmgr, commit_transaction, &errstack);
}
qmgr = NULL;
return rval;
}
//---------------------------------------------------------------------------
void
DagmanClassad::CloseConnection( Qmgr_connection *queue )
{
if ( !DisconnectQ( queue ) ) {
debug_printf( DEBUG_QUIET,
"WARNING: queue transaction failed. No attributes were set.\n" );
check_warning_strictness( DAG_STRICT_3 );
}
}
int
CondorQ::fetchQueueFromHostAndProcess ( const char *host,
StringList &attrs,
int fetch_opts,
int match_limit,
condor_q_process_func process_func,
void * process_func_data,
int useFastPath,
CondorError* errstack,
ClassAd ** psummary_ad)
{
Qmgr_connection *qmgr;
ExprTree *tree;
char *constraint;
int result;
// make the query ad
if ((result = query.makeQuery (tree)) != Q_OK)
return result;
constraint = strdup( ExprTreeToString( tree ) );
delete tree;
if (useFastPath > 1) {
int result = fetchQueueFromHostAndProcessV2(host, constraint, attrs, fetch_opts, match_limit, process_func, process_func_data, connect_timeout, useFastPath, errstack, psummary_ad);
free( constraint);
return result;
}
if (fetch_opts != fetch_Jobs) {
free( constraint );
return Q_UNSUPPORTED_OPTION_ERROR;
}
/*
connect to the Q manager.
use a timeout of 20 seconds, and a read-only connection.
why 20 seconds? because careful research by Derek has shown
that whenever one needs a periodic time value, 20 is always
optimal. :^).
*/
init(); // needed to get default connect_timeout
if( !(qmgr = ConnectQ( host, connect_timeout, true, errstack)) ) {
free( constraint );
return Q_SCHEDD_COMMUNICATION_ERROR;
}
// get the ads and filter them
result = getFilterAndProcessAds (constraint, attrs, match_limit, process_func, process_func_data, useFastPath);
DisconnectQ (qmgr);
free( constraint );
return result;
}
int CondorQ::
fetchQueue (ClassAdList &list, StringList &attrs, ClassAd *ad, CondorError* errstack)
{
Qmgr_connection *qmgr;
ExprTree *tree;
int result;
char scheddString [32];
const char *constraint;
bool useFastPath = false;
// make the query ad
if ((result = query.makeQuery (tree)) != Q_OK)
return result;
constraint = ExprTreeToString( tree );
delete tree;
// connect to the Q manager
init(); // needed to get default connect_timeout
if (ad == 0)
{
// local case
if( !(qmgr = ConnectQ( 0, connect_timeout, true, errstack)) ) {
errstack->push("TEST", 0, "FOO");
return Q_SCHEDD_COMMUNICATION_ERROR;
}
useFastPath = true;
}
else
{
// remote case to handle condor_globalq
if (!ad->LookupString (ATTR_SCHEDD_IP_ADDR, scheddString, sizeof(scheddString)))
return Q_NO_SCHEDD_IP_ADDR;
if( !(qmgr = ConnectQ( scheddString, connect_timeout, true, errstack)) )
return Q_SCHEDD_COMMUNICATION_ERROR;
}
// get the ads and filter them
getAndFilterAds (constraint, attrs, list, useFastPath);
DisconnectQ (qmgr);
return Q_OK;
}
int CondorQ::
fetchQueueFromHost (ClassAdList &list, StringList &attrs, const char *host, char const *schedd_version, CondorError* errstack)
{
Qmgr_connection *qmgr;
ExprTree *tree;
const char *constraint;
int result;
// make the query ad
if ((result = query.makeQuery (tree)) != Q_OK)
return result;
constraint = ExprTreeToString( tree );
delete tree;
/*
connect to the Q manager.
use a timeout of 20 seconds, and a read-only connection.
why 20 seconds? because careful research by Derek has shown
that whenever one needs a periodic time value, 20 is always
optimal. :^).
*/
init(); // needed to get default connect_timeout
if( !(qmgr = ConnectQ( host, connect_timeout, true, errstack)) )
return Q_SCHEDD_COMMUNICATION_ERROR;
int useFastPath = 0;
if( schedd_version && *schedd_version ) {
CondorVersionInfo v(schedd_version);
useFastPath = v.built_since_version(6,9,3) ? 1 : 0;
if (v.built_since_version(8, 1, 5)) {
useFastPath = 2;
}
}
// get the ads and filter them
result = getAndFilterAds (constraint, attrs, -1, list, useFastPath);
DisconnectQ (qmgr);
return result;
}
int
CondorQ::fetchQueueFromHostAndProcess ( const char *host,
StringList &attrs,
process_function process_func,
bool useFastPath,
CondorError* errstack)
{
Qmgr_connection *qmgr;
ExprTree *tree;
char *constraint;
int result;
// make the query ad
if ((result = query.makeQuery (tree)) != Q_OK)
return result;
constraint = strdup( ExprTreeToString( tree ) );
delete tree;
/*
connect to the Q manager.
use a timeout of 20 seconds, and a read-only connection.
why 20 seconds? because careful research by Derek has shown
that whenever one needs a periodic time value, 20 is always
optimal. :^).
*/
init(); // needed to get default connect_timeout
if( !(qmgr = ConnectQ( host, connect_timeout, true, errstack)) ) {
free( constraint );
return Q_SCHEDD_COMMUNICATION_ERROR;
}
// get the ads and filter them
result = getFilterAndProcessAds (constraint, attrs, process_func, useFastPath);
DisconnectQ (qmgr);
free( constraint );
return result;
}
void
doContactSchedd()
{
int rc;
Qmgr_connection *schedd;
BaseJob *curr_job;
ClassAd *next_ad;
char expr_buf[12000];
bool schedd_updates_complete = false;
bool schedd_deletes_complete = false;
bool add_remove_jobs_complete = false;
bool update_jobs_complete = false;
bool commit_transaction = true;
int failure_line_num = 0;
bool send_reschedule = false;
std::string error_str = "";
StringList dirty_job_ids;
char *job_id_str;
PROC_ID job_id;
CondorError errstack;
dprintf(D_FULLDEBUG,"in doContactSchedd()\n");
initJobExprs();
contactScheddTid = TIMER_UNSET;
// vacateJobs
/////////////////////////////////////////////////////
if ( pendingScheddVacates.getNumElements() != 0 ) {
std::string buff;
StringList job_ids;
VacateRequest curr_request;
int result;
ClassAd* rval;
pendingScheddVacates.startIterations();
while ( pendingScheddVacates.iterate( curr_request ) != 0 ) {
formatstr( buff, "%d.%d", curr_request.job->procID.cluster,
curr_request.job->procID.proc );
job_ids.append( buff.c_str() );
}
char *tmp = job_ids.print_to_string();
if ( tmp ) {
dprintf( D_FULLDEBUG, "Calling vacateJobs on %s\n", tmp );
free(tmp);
tmp = NULL;
}
rval = ScheddObj->vacateJobs( &job_ids, VACATE_FAST, &errstack );
if ( rval == NULL ) {
formatstr( error_str, "vacateJobs returned NULL, CondorError: %s!",
errstack.getFullText().c_str() );
goto contact_schedd_failure;
} else {
pendingScheddVacates.startIterations();
while ( pendingScheddVacates.iterate( curr_request ) != 0 ) {
formatstr( buff, "job_%d_%d", curr_request.job->procID.cluster,
curr_request.job->procID.proc );
if ( !rval->LookupInteger( buff.c_str(), result ) ) {
dprintf( D_FULLDEBUG, "vacateJobs returned malformed ad\n" );
EXCEPT( "vacateJobs returned malformed ad" );
} else {
dprintf( D_FULLDEBUG, " %d.%d vacate result: %d\n",
curr_request.job->procID.cluster,
curr_request.job->procID.proc,result);
pendingScheddVacates.remove( curr_request.job->procID );
curr_request.result = (action_result_t)result;
curr_request.job->SetEvaluateState();
completedScheddVacates.insert( curr_request.job->procID,
curr_request );
}
}
delete rval;
}
}
schedd = ConnectQ( ScheddAddr, QMGMT_TIMEOUT, false, NULL, myUserName, CondorVersion() );
if ( !schedd ) {
error_str = "Failed to connect to schedd!";
goto contact_schedd_failure;
}
// CheckLeases
/////////////////////////////////////////////////////
if ( checkLeasesSignaled ) {
dprintf( D_FULLDEBUG, "querying for renewed leases\n" );
// Grab the lease attributes of all the jobs in our global hashtable.
BaseJob::JobsByProcId.startIterations();
while ( BaseJob::JobsByProcId.iterate( curr_job ) != 0 ) {
int new_expiration;
rc = GetAttributeInt( curr_job->procID.cluster,
curr_job->procID.proc,
ATTR_TIMER_REMOVE_CHECK,
&new_expiration );
if ( rc < 0 ) {
if ( errno == ETIMEDOUT ) {
failure_line_num = __LINE__;
commit_transaction = false;
goto contact_schedd_disconnect;
} else {
// This job doesn't have doesn't have a lease from
// the submitter. Skip it.
continue;
}
}
curr_job->UpdateJobLeaseReceived( new_expiration );
}
checkLeasesSignaled = false;
} // end of handling check leases
// AddJobs
/////////////////////////////////////////////////////
if ( addJobsSignaled || firstScheddContact ) {
int num_ads = 0;
dprintf( D_FULLDEBUG, "querying for new jobs\n" );
// Make sure we grab all Globus Universe jobs (except held ones
// that we previously indicated we were done with)
// when we first start up in case we're recovering from a
// shutdown/meltdown.
// Otherwise, grab all jobs that are unheld and aren't marked as
// currently being managed and aren't marked as not matched.
// If JobManaged is undefined, equate it with false.
// If Matched is undefined, equate it with true.
// NOTE: Schedds from Condor 6.6 and earlier don't include
// "(Universe==9)" in the constraint they give to the gridmanager,
// so this gridmanager will pull down non-globus-universe ads,
// although it won't use them. This is inefficient but not
// incorrect behavior.
if ( firstScheddContact ) {
// Grab all jobs for us to manage. This expression is a
// derivative of the expression below for new jobs. We add
// "|| Managed =?= TRUE" to also get jobs our previous
// incarnation was in the middle of managing when it died
// (if it died unexpectedly). With the new term, the
// "&& Managed =!= TRUE" from the new jobs expression becomes
// superfluous (by boolean logic), so we drop it.
sprintf( expr_buf,
"%s && %s && ((%s && %s) || %s)",
expr_schedd_job_constraint.c_str(),
expr_not_completely_done.c_str(),
expr_matched_or_undef.c_str(),
expr_not_held.c_str(),
expr_managed.c_str()
);
} else {
// Grab new jobs for us to manage
sprintf( expr_buf,
"%s && %s && %s && %s && %s",
expr_schedd_job_constraint.c_str(),
expr_not_completely_done.c_str(),
expr_matched_or_undef.c_str(),
expr_not_held.c_str(),
expr_not_managed.c_str()
);
}
dprintf( D_FULLDEBUG,"Using constraint %s\n",expr_buf);
next_ad = GetNextJobByConstraint( expr_buf, 1 );
while ( next_ad != NULL ) {
PROC_ID procID;
BaseJob *old_job;
int job_is_matched = 1; // default to true if not in ClassAd
next_ad->LookupInteger( ATTR_CLUSTER_ID, procID.cluster );
next_ad->LookupInteger( ATTR_PROC_ID, procID.proc );
bool job_is_managed = jobExternallyManaged(next_ad);
next_ad->LookupBool(ATTR_JOB_MATCHED,job_is_matched);
if ( BaseJob::JobsByProcId.lookup( procID, old_job ) != 0 ) {
JobType *job_type = NULL;
BaseJob *new_job = NULL;
// job had better be either managed or matched! (or both)
ASSERT( job_is_managed || job_is_matched );
if ( MustExpandJobAd( next_ad ) ) {
// Get the expanded ClassAd from the schedd, which
// has the GridResource filled in with info from
// the matched ad.
delete next_ad;
next_ad = NULL;
next_ad = GetJobAd(procID.cluster,procID.proc);
if ( next_ad == NULL && errno == ETIMEDOUT ) {
failure_line_num = __LINE__;
commit_transaction = false;
goto contact_schedd_disconnect;
}
if ( next_ad == NULL ) {
// We may get here if it was not possible to expand
// one of the $$() expressions. We don't want to
// roll back the transaction and blow away the
// hold that the schedd just put on the job, so
// simply skip over this ad.
dprintf(D_ALWAYS,"Failed to get expanded job ClassAd from Schedd for %d.%d. errno=%d\n",procID.cluster,procID.proc,errno);
goto contact_schedd_next_add_job;
}
}
// Search our job types for one that'll handle this job
jobTypes.Rewind();
while ( jobTypes.Next( job_type ) ) {
if ( job_type->AdMatchFunc( next_ad ) ) {
// Found one!
dprintf( D_FULLDEBUG, "Using job type %s for job %d.%d\n",
job_type->Name, procID.cluster, procID.proc );
break;
}
}
if ( job_type != NULL ) {
new_job = job_type->CreateFunc( next_ad );
} else {
dprintf( D_ALWAYS, "No handlers for job %d.%d\n",
procID.cluster, procID.proc );
new_job = new BaseJob( next_ad );
}
ASSERT(new_job);
new_job->SetEvaluateState();
dprintf(D_ALWAYS,"Found job %d.%d --- inserting\n",
new_job->procID.cluster,new_job->procID.proc);
num_ads++;
if ( !job_is_managed ) {
rc = tSetAttributeString( new_job->procID.cluster,
new_job->procID.proc,
ATTR_JOB_MANAGED,
MANAGED_EXTERNAL);
if ( rc < 0 ) {
failure_line_num = __LINE__;
commit_transaction = false;
goto contact_schedd_disconnect;
}
}
} else {
// We already know about this job, skip
// But also set Managed=true on the schedd so that it won't
// keep signalling us about it
delete next_ad;
rc = tSetAttributeString( procID.cluster, procID.proc,
ATTR_JOB_MANAGED, MANAGED_EXTERNAL );
if ( rc < 0 ) {
failure_line_num = __LINE__;
commit_transaction = false;
goto contact_schedd_disconnect;
}
}
contact_schedd_next_add_job:
next_ad = GetNextJobByConstraint( expr_buf, 0 );
} // end of while next_ad
if ( errno == ETIMEDOUT ) {
failure_line_num = __LINE__;
commit_transaction = false;
goto contact_schedd_disconnect;
}
dprintf(D_FULLDEBUG,"Fetched %d new job ads from schedd\n",num_ads);
} // end of handling add jobs
// RemoveJobs
/////////////////////////////////////////////////////
// We always want to perform this check. Otherwise, we may overwrite a
// REMOVED/HELD/COMPLETED status with something else below.
{
int num_ads = 0;
dprintf( D_FULLDEBUG, "querying for removed/held jobs\n" );
// Grab jobs marked as REMOVED/COMPLETED or marked as HELD that we
// haven't previously indicated that we're done with (by setting
// JobManaged to "Schedd".
sprintf( expr_buf, "(%s) && (%s) && (%s == %d || %s == %d || (%s == %d && %s =?= \"%s\"))",
ScheddJobConstraint, expr_not_completely_done.c_str(),
ATTR_JOB_STATUS, REMOVED,
ATTR_JOB_STATUS, COMPLETED, ATTR_JOB_STATUS, HELD,
ATTR_JOB_MANAGED, MANAGED_EXTERNAL );
dprintf( D_FULLDEBUG,"Using constraint %s\n",expr_buf);
next_ad = GetNextJobByConstraint( expr_buf, 1 );
while ( next_ad != NULL ) {
PROC_ID procID;
BaseJob *next_job;
int curr_status;
next_ad->LookupInteger( ATTR_CLUSTER_ID, procID.cluster );
next_ad->LookupInteger( ATTR_PROC_ID, procID.proc );
next_ad->LookupInteger( ATTR_JOB_STATUS, curr_status );
if ( BaseJob::JobsByProcId.lookup( procID, next_job ) == 0 ) {
// Should probably skip jobs we already have marked as
// held or removed
next_job->JobAdUpdateFromSchedd( next_ad, true );
num_ads++;
} else if ( curr_status == REMOVED ) {
// If we don't know about the job, act like we got an
// ADD_JOBS signal from the schedd the next time we
// connect, so that we'll create a Job object for it
// and decide how it needs to be handled.
// TODO The AddJobs and RemoveJobs queries shoule be
// combined into a single query.
dprintf( D_ALWAYS,
"Don't know about removed job %d.%d. "
"Will treat it as a new job to manage\n",
procID.cluster, procID.proc );
addJobsSignaled = true;
} else {
dprintf( D_ALWAYS, "Don't know about held/completed job %d.%d. "
"Ignoring it\n",
procID.cluster, procID.proc );
}
delete next_ad;
next_ad = GetNextJobByConstraint( expr_buf, 0 );
}
if ( errno == ETIMEDOUT ) {
failure_line_num = __LINE__;
commit_transaction = false;
goto contact_schedd_disconnect;
}
dprintf(D_FULLDEBUG,"Fetched %d job ads from schedd\n",num_ads);
}
if ( RemoteCommitTransaction() < 0 ) {
failure_line_num = __LINE__;
commit_transaction = false;
goto contact_schedd_disconnect;
}
add_remove_jobs_complete = true;
// Retrieve dirty attributes
/////////////////////////////////////////////////////
if ( updateJobsSignaled ) {
dprintf( D_FULLDEBUG, "querying for jobs with attribute updates\n" );
sprintf( expr_buf, "%s && %s && %s && %s",
expr_schedd_job_constraint.c_str(),
expr_not_completely_done.c_str(),
expr_not_held.c_str(),
expr_managed.c_str()
);
dprintf( D_FULLDEBUG,"Using constraint %s\n",expr_buf);
next_ad = GetNextDirtyJobByConstraint( expr_buf, 1 );
while ( next_ad != NULL ) {
ClassAd updates;
char str[PROC_ID_STR_BUFLEN];
next_ad->LookupInteger( ATTR_CLUSTER_ID, job_id.cluster );
next_ad->LookupInteger( ATTR_PROC_ID, job_id.proc );
if ( GetDirtyAttributes( job_id.cluster, job_id.proc, &updates ) < 0 ) {
dprintf( D_ALWAYS, "Failed to retrieve dirty attributes for job %d.%d\n", job_id.cluster, job_id.proc );
failure_line_num = __LINE__;
delete next_ad;
goto contact_schedd_disconnect;
}
else {
dprintf (D_FULLDEBUG, "Retrieved updated attributes for job %d.%d\n", job_id.cluster, job_id.proc);
dPrintAd(D_JOB, updates);
}
if ( BaseJob::JobsByProcId.lookup( job_id, curr_job ) == 0 ) {
curr_job->JobAdUpdateFromSchedd( &updates, false );
ProcIdToStr( job_id, str );
dirty_job_ids.append( str );
}
else {
dprintf( D_ALWAYS, "Don't know about updated job %d.%d. "
"Ignoring it\n",
job_id.cluster, job_id.proc );
}
delete next_ad;
next_ad = GetNextDirtyJobByConstraint( expr_buf, 0 );
}
}
update_jobs_complete = true;
// if ( BeginTransaction() < 0 ) {
errno = 0;
BeginTransaction();
if ( errno == ETIMEDOUT ) {
failure_line_num = __LINE__;
commit_transaction = false;
goto contact_schedd_disconnect;
}
// requestJobStatus
/////////////////////////////////////////////////////
if ( pendingJobStatus.getNumElements() != 0 ) {
JobStatusRequest curr_request;
pendingJobStatus.startIterations();
while ( pendingJobStatus.iterate( curr_request ) != 0 ) {
int status;
rc = GetAttributeInt( curr_request.job_id.cluster,
curr_request.job_id.proc,
ATTR_JOB_STATUS, &status );
if ( rc < 0 ) {
if ( errno == ETIMEDOUT ) {
failure_line_num = __LINE__;
commit_transaction = false;
goto contact_schedd_disconnect;
} else {
// The job is not in the schedd's job queue. This
// probably means that the user did a condor_rm -f,
// so return a job status of REMOVED.
status = REMOVED;
}
}
// return status
dprintf( D_FULLDEBUG, "%d.%d job status: %d\n",
curr_request.job_id.cluster,
curr_request.job_id.proc, status );
pendingJobStatus.remove( curr_request.job_id );
curr_request.job_status = status;
daemonCore->Reset_Timer( curr_request.tid, 0 );
completedJobStatus.insert( curr_request.job_id,
curr_request );
}
}
// Update existing jobs
/////////////////////////////////////////////////////
ScheddUpdateRequest *curr_request;
pendingScheddUpdates.startIterations();
while ( pendingScheddUpdates.iterate( curr_request ) != 0 ) {
curr_job = curr_request->m_job;
dprintf(D_FULLDEBUG,"Updating classad values for %d.%d:\n",
curr_job->procID.cluster, curr_job->procID.proc);
const char *attr_name;
const char *attr_value;
ExprTree *expr;
bool fake_job_in_queue = false;
curr_job->jobAd->ResetExpr();
while ( curr_job->jobAd->NextDirtyExpr(attr_name, expr) == true &&
fake_job_in_queue == false ) {
attr_value = ExprTreeToString( expr );
dprintf(D_FULLDEBUG," %s = %s\n",attr_name,attr_value);
rc = SetAttribute( curr_job->procID.cluster,
curr_job->procID.proc,
attr_name,
attr_value);
if ( rc < 0 ) {
if ( errno == ETIMEDOUT ) {
failure_line_num = __LINE__;
commit_transaction = false;
goto contact_schedd_disconnect;
} else {
// The job is not in the schedd's job queue. This
// probably means that the user did a condor_rm -f,
// so pretend that all updates for the job succeed.
// Otherwise, we'll never make forward progress on
// the job.
// TODO We should also fake a job status of REMOVED
// to the job, so it can do what cleanup it can.
fake_job_in_queue = true;
break;
}
}
}
}
if ( RemoteCommitTransaction() < 0 ) {
failure_line_num = __LINE__;
commit_transaction = false;
goto contact_schedd_disconnect;
}
schedd_updates_complete = true;
// Delete existing jobs
/////////////////////////////////////////////////////
errno = 0;
BeginTransaction();
if ( errno == ETIMEDOUT ) {
failure_line_num = __LINE__;
commit_transaction = false;
goto contact_schedd_disconnect;
}
pendingScheddUpdates.startIterations();
while ( pendingScheddUpdates.iterate( curr_request ) != 0 ) {
curr_job = curr_request->m_job;
if ( curr_job->deleteFromSchedd ) {
dprintf(D_FULLDEBUG,"Deleting job %d.%d from schedd\n",
curr_job->procID.cluster, curr_job->procID.proc);
rc = DestroyProc(curr_job->procID.cluster,
curr_job->procID.proc);
// NOENT means the job doesn't exist. Good enough for us.
if ( rc < 0 && rc != DESTROYPROC_ENOENT) {
failure_line_num = __LINE__;
commit_transaction = false;
goto contact_schedd_disconnect;
}
}
}
if ( RemoteCommitTransaction() < 0 ) {
failure_line_num = __LINE__;
commit_transaction = false;
goto contact_schedd_disconnect;
}
schedd_deletes_complete = true;
contact_schedd_disconnect:
DisconnectQ( schedd, commit_transaction );
if ( add_remove_jobs_complete == true ) {
firstScheddContact = false;
addJobsSignaled = false;
} else {
formatstr( error_str, "Schedd connection error during Add/RemoveJobs at line %d!", failure_line_num );
goto contact_schedd_failure;
}
if ( update_jobs_complete == true ) {
updateJobsSignaled = false;
} else {
formatstr( error_str, "Schedd connection error during dirty attribute update at line %d!", failure_line_num );
goto contact_schedd_failure;
}
if ( schedd_updates_complete == false ) {
formatstr( error_str, "Schedd connection error during updates at line %d!", failure_line_num );
goto contact_schedd_failure;
}
// Clear dirty bits for all jobs updated
if ( !dirty_job_ids.isEmpty() ) {
ClassAd *rval;
dprintf( D_FULLDEBUG, "Calling clearDirtyAttrs on %d jobs\n",
dirty_job_ids.number() );
dirty_job_ids.rewind();
rval = ScheddObj->clearDirtyAttrs( &dirty_job_ids, &errstack );
if ( rval == NULL ) {
dprintf(D_ALWAYS, "Failed to notify schedd to clear dirty attributes. CondorError: %s\n", errstack.getFullText().c_str() );
}
delete rval;
}
// Wake up jobs that had schedd updates pending and delete job
// objects that wanted to be deleted
pendingScheddUpdates.startIterations();
while ( pendingScheddUpdates.iterate( curr_request ) != 0 ) {
curr_job = curr_request->m_job;
curr_job->jobAd->ClearAllDirtyFlags();
if ( curr_job->deleteFromGridmanager ) {
// If the Job object wants to delete the job from the
// schedd but we failed to do so, don't delete the job
// object yet; wait until we successfully delete the job
// from the schedd.
if ( curr_job->deleteFromSchedd == true &&
schedd_deletes_complete == false ) {
continue;
}
// If wantRematch is set, send a reschedule now
if ( curr_job->wantRematch ) {
send_reschedule = true;
}
pendingScheddUpdates.remove( curr_job->procID );
pendingScheddVacates.remove( curr_job->procID );
pendingJobStatus.remove( curr_job->procID );
completedJobStatus.remove( curr_job->procID );
completedScheddVacates.remove( curr_job->procID );
delete curr_job;
} else {
pendingScheddUpdates.remove( curr_job->procID );
if ( curr_request->m_notify ) {
curr_job->SetEvaluateState();
}
}
delete curr_request;
}
// Poke objects that wanted to be notified when a schedd update completed
// successfully (possibly minus deletes)
int timer_id;
scheddUpdateNotifications.Rewind();
while ( scheddUpdateNotifications.Next( timer_id ) ) {
daemonCore->Reset_Timer( timer_id, 0 );
}
scheddUpdateNotifications.Clear();
if ( send_reschedule == true ) {
ScheddObj->reschedule();
}
// Check if we have any jobs left to manage. If not, exit.
if ( BaseJob::JobsByProcId.getNumElements() == 0 ) {
dprintf( D_ALWAYS, "No jobs left, shutting down\n" );
daemonCore->Send_Signal( daemonCore->getpid(), SIGTERM );
}
lastContactSchedd = time(NULL);
if ( schedd_deletes_complete == false ) {
error_str = "Problem using DestroyProc to delete jobs!";
goto contact_schedd_failure;
}
scheddFailureCount = 0;
// For each job that had dirty attributes, re-evaluate the policy
dirty_job_ids.rewind();
while ( (job_id_str = dirty_job_ids.next()) != NULL ) {
StrToProcIdFixMe(job_id_str, job_id);
if ( BaseJob::JobsByProcId.lookup( job_id, curr_job ) == 0 ) {
curr_job->EvalPeriodicJobExpr();
}
}
dprintf(D_FULLDEBUG,"leaving doContactSchedd()\n");
return;
contact_schedd_failure:
scheddFailureCount++;
if ( error_str == "" ) {
error_str = "Failure in doContactSchedd";
}
if ( scheddFailureCount >= maxScheddFailures ) {
dprintf( D_ALWAYS, "%s\n", error_str.c_str() );
EXCEPT( "Too many failures connecting to schedd!" );
}
dprintf( D_ALWAYS, "%s Will retry\n", error_str.c_str() );
lastContactSchedd = time(NULL);
RequestContactSchedd();
return;
}
void
update_job_status( struct rusage *localp, struct rusage *remotep )
{
int status = -1;
double utime = 0.0;
double stime = 0.0;
int tot_sus=0, cum_sus=0, last_sus=0;
char buf[1024*50];
// If the job completed, and there is no HISTORY file specified,
// the don't bother to update the job ClassAd since it is about to be
// flushed into the bit bucket by the schedd anyway.
char *myHistoryFile = param("HISTORY");
if ((Proc->status == COMPLETED) && (myHistoryFile==NULL)) {
return;
}
if (myHistoryFile) {
free(myHistoryFile);
}
if (!JobAd)
{
EXCEPT( "update_job_status(): No job ad");
}
JobAd->LookupInteger(ATTR_TOTAL_SUSPENSIONS, tot_sus);
JobAd->LookupInteger(ATTR_CUMULATIVE_SUSPENSION_TIME, cum_sus);
JobAd->LookupInteger(ATTR_LAST_SUSPENSION_TIME, last_sus);
//new syntax, can use filesystem to authenticate
if (!ConnectQ(schedd, SHADOW_QMGMT_TIMEOUT) ||
GetAttributeInt(Proc->id.cluster, Proc->id.proc, ATTR_JOB_STATUS,
&status) < 0) {
EXCEPT("Failed to connect to schedd!");
}
job_report_update_queue( Proc );
if( status == REMOVED ) {
dprintf( D_ALWAYS, "update_job_status(): Job %d.%d has been removed "
"by condor_rm\n", Proc->id.cluster, Proc->id.proc );
} else {
SetAttributeInt(Proc->id.cluster, Proc->id.proc,
ATTR_TOTAL_SUSPENSIONS, tot_sus);
SetAttributeInt(Proc->id.cluster, Proc->id.proc,
ATTR_CUMULATIVE_SUSPENSION_TIME, cum_sus);
SetAttributeInt(Proc->id.cluster, Proc->id.proc,
ATTR_LAST_SUSPENSION_TIME, last_sus);
update_job_rusage( localp, remotep );
Proc->image_size = ImageSize;
SetAttributeInt(Proc->id.cluster, Proc->id.proc, ATTR_IMAGE_SIZE,
ImageSize);
// For standard universe. MemoryUsed==ImageSize, no need to param this one.
// because imagesize is already the best measure of memory usage.
SetAttribute(Proc->id.cluster, Proc->id.proc, ATTR_MEMORY_USAGE, "((ImageSize+1023)/1024)");
SetAttributeInt(Proc->id.cluster, Proc->id.proc, ATTR_JOB_EXIT_STATUS,
JobExitStatus);
rusage_to_float( Proc->local_usage, &utime, &stime );
SetAttributeFloat(Proc->id.cluster, Proc->id.proc,
ATTR_JOB_LOCAL_USER_CPU, utime);
SetAttributeFloat(Proc->id.cluster, Proc->id.proc,
ATTR_JOB_LOCAL_SYS_CPU, stime);
rusage_to_float( Proc->remote_usage[0], &utime, &stime );
SetAttributeFloat(Proc->id.cluster, Proc->id.proc,
ATTR_JOB_REMOTE_USER_CPU, utime);
SetAttributeFloat(Proc->id.cluster, Proc->id.proc,
ATTR_JOB_REMOTE_SYS_CPU, stime);
dprintf(D_FULLDEBUG,"TIME DEBUG 3 USR remotep=%lu Proc=%lu utime=%f\n",
remotep->ru_utime.tv_sec,
Proc->remote_usage[0].ru_utime.tv_sec, utime);
dprintf(D_FULLDEBUG,"TIME DEBUG 4 SYS remotep=%lu Proc=%lu utime=%f\n",
remotep->ru_stime.tv_sec,
Proc->remote_usage[0].ru_stime.tv_sec, stime);
if( sock_RSC1 ) {
float TotalBytesSentUpdate =
TotalBytesSent + sock_RSC1->get_bytes_sent() + BytesSent;
float TotalBytesRecvdUpdate =
TotalBytesRecvd + sock_RSC1->get_bytes_recvd() + BytesRecvd;
SetAttributeFloat( Proc->id.cluster, Proc->id.proc,
ATTR_BYTES_SENT, TotalBytesSentUpdate );
SetAttributeFloat( Proc->id.cluster, Proc->id.proc,
ATTR_BYTES_RECVD, TotalBytesRecvdUpdate );
float RSCBytesSentUpdate =
sock_RSC1->get_bytes_sent() + RSCBytesSent;
float RSCBytesRecvdUpdate =
sock_RSC1->get_bytes_recvd() + RSCBytesRecvd;
SetAttributeFloat( Proc->id.cluster, Proc->id.proc,
ATTR_RSC_BYTES_SENT, RSCBytesSentUpdate );
SetAttributeFloat( Proc->id.cluster, Proc->id.proc,
ATTR_RSC_BYTES_RECVD, RSCBytesRecvdUpdate );
}
if( ExitReason == JOB_CKPTED || ExitReason == JOB_NOT_CKPTED ) {
SetAttributeInt( Proc->id.cluster, Proc->id.proc,
ATTR_LAST_VACATE_TIME, time(0) );
}
if( ExitReason == JOB_CKPTED || LastCkptTime > LastRestartTime ) {
int uncommitted_suspension_time = 0;
JobAd->LookupInteger(ATTR_UNCOMMITTED_SUSPENSION_TIME, uncommitted_suspension_time);
if( uncommitted_suspension_time > 0 ) {
int committed_suspension_time = 0;
GetAttributeInt(Proc->id.cluster, Proc->id.proc,
ATTR_COMMITTED_SUSPENSION_TIME, &committed_suspension_time);
committed_suspension_time += uncommitted_suspension_time;
SetAttributeInt(Proc->id.cluster, Proc->id.proc,
ATTR_COMMITTED_SUSPENSION_TIME, committed_suspension_time);
}
}
// if we had checkpointed, then save all of these attributes as well.
if (LastCkptTime > LastRestartTime) {
SetAttributeInt(Proc->id.cluster, Proc->id.proc,
ATTR_LAST_CKPT_TIME, LastCkptTime);
CommittedTime=0;
GetAttributeInt(Proc->id.cluster, Proc->id.proc,
ATTR_JOB_COMMITTED_TIME, &CommittedTime);
CommittedTime += LastCkptTime - LastRestartTime;
SetAttributeInt(Proc->id.cluster, Proc->id.proc,
ATTR_JOB_COMMITTED_TIME, CommittedTime);
LastRestartTime = LastCkptTime;
SetAttributeInt(Proc->id.cluster, Proc->id.proc,
ATTR_NUM_CKPTS, NumCkpts);
SetAttributeInt(Proc->id.cluster, Proc->id.proc,
ATTR_NUM_RESTARTS, NumRestarts);
if (Executing_Arch) {
SetAttributeString(Proc->id.cluster, Proc->id.proc,
ATTR_CKPT_ARCH, Executing_Arch);
}
if (Executing_OpSys) {
SetAttributeString(Proc->id.cluster, Proc->id.proc,
ATTR_CKPT_OPSYS, Executing_OpSys);
}
// If we wrote a checkpoint, store the location in the
// LastCkptServer attribute. If we didn't use a checkpoint
// server (i.e., we stored it locally), then make sure
// no LastCkptServer attribute is set.
if (LastCkptServer) {
SetAttributeString(Proc->id.cluster, Proc->id.proc,
ATTR_LAST_CKPT_SERVER, LastCkptServer);
} else {
DeleteAttribute(Proc->id.cluster, Proc->id.proc,
ATTR_LAST_CKPT_SERVER);
}
if (LastCkptPlatform) {
SetAttributeString(Proc->id.cluster, Proc->id.proc,
ATTR_LAST_CHECKPOINT_PLATFORM,
LastCkptPlatform);
}
}
// if the job completed, we should include the run-time in
// committed time, since it contributed to the completion of
// the job. Also, commit the exit code/signal stuff, plus any
// core filenames.
if (Proc->status == COMPLETED) {
int exit_code, exit_signal, exit_by_signal;
int pending;
// update the time.
CommittedTime = 0;
GetAttributeInt(Proc->id.cluster, Proc->id.proc,
ATTR_JOB_COMMITTED_TIME, &CommittedTime);
CommittedTime += Proc->completion_date - LastRestartTime;
SetAttributeInt(Proc->id.cluster, Proc->id.proc,
ATTR_JOB_COMMITTED_TIME, CommittedTime);
// if there is a core file, update that too.
if (JobAd->LookupString(ATTR_JOB_CORE_FILENAME, buf, sizeof(buf))) {
SetAttributeString(Proc->id.cluster, Proc->id.proc,
ATTR_JOB_CORE_FILENAME, buf);
}
// only new style ads have ATTR_ON_EXIT_BY_SIGNAL, so only
// SetAttribute for those types of ads
if (JobAd->LookupInteger(ATTR_ON_EXIT_BY_SIGNAL, exit_by_signal)==1)
{
SetAttributeInt(Proc->id.cluster, Proc->id.proc,
ATTR_ON_EXIT_BY_SIGNAL, exit_by_signal);
if (exit_by_signal == 1) /* exited via signal */
{
JobAd->LookupInteger(ATTR_ON_EXIT_SIGNAL, exit_signal);
SetAttributeInt(Proc->id.cluster, Proc->id.proc,
ATTR_ON_EXIT_SIGNAL, exit_signal);
}
else
{
JobAd->LookupInteger(ATTR_ON_EXIT_CODE, exit_code);
SetAttributeInt(Proc->id.cluster, Proc->id.proc,
ATTR_ON_EXIT_CODE, exit_code);
}
}
// and now, let's try and mark this job as a terminate pending
// job. If the job already is, then fine. We'll mark it again.
if (JobAd->LookupBool(ATTR_TERMINATION_PENDING, pending)) {
SetAttribute(Proc->id.cluster, Proc->id.proc,
ATTR_TERMINATION_PENDING, pending?"TRUE":"FALSE");
} else {
// if it isn't in the job ad, then add it to the saved ad in the
// schedd.
SetAttribute(Proc->id.cluster, Proc->id.proc,
ATTR_TERMINATION_PENDING, "TRUE");
}
// store the reason why the job is marked completed.
if (JobAd->LookupString(ATTR_TERMINATION_REASON, buf, sizeof(buf))) {
SetAttributeString(Proc->id.cluster, Proc->id.proc,
ATTR_TERMINATION_REASON, buf);
}
// Set up the exit code the shadow was about to exit with to
// help support the terminate pending "state".
SetAttributeInt(Proc->id.cluster, Proc->id.proc,
ATTR_TERMINATION_EXITREASON, ExitReason);
// Put the job status as created by waitpid() into the job ad
// itself. This is to implement the terminate_pending feature. It
// is done like this because EVERYWHERE in this codebase we do
// stuff like WIFEXITED(JobStatus) and it turns out there are no
// user level macros to will one of those status values as returned
// by waitpid() into existance. So, we'll put it directly into the
// job ad to prevent me having to reimplement a few large functions
// which deal with JobStatus directly--as it is sadly a global
// variable.
SetAttributeInt(Proc->id.cluster, Proc->id.proc,
ATTR_WAITPID_STATUS, JobStatus);
}
}
if (!DisconnectQ(0)) {
EXCEPT("Failed to commit updated job queue status!");
}
}
/* Mess up the in memory job ad with interesting statistics about suspensions */
void record_suspension_hack(unsigned int action)
{
char tmp[256];
int total_suspensions;
int last_suspension_time;
int cumulative_suspension_time;
extern char *schedd;
if (!JobAd)
{
EXCEPT("Suspension code: Non-existant JobAd");
}
switch(action)
{
case ULOG_JOB_SUSPENDED:
/* Add to ad number of suspensions */
JobAd->LookupInteger(ATTR_TOTAL_SUSPENSIONS, total_suspensions);
total_suspensions++;
sprintf(tmp, "%s = %d", ATTR_TOTAL_SUSPENSIONS, total_suspensions);
JobAd->Insert(tmp);
/* Add to ad the current suspension time */
last_suspension_time = time(NULL);
sprintf(tmp, "%s = %d", ATTR_LAST_SUSPENSION_TIME,
last_suspension_time);
JobAd->Insert(tmp);
break;
case ULOG_JOB_UNSUSPENDED: {
/* add in the time I spent suspended to a running total */
JobAd->LookupInteger(ATTR_CUMULATIVE_SUSPENSION_TIME,
cumulative_suspension_time);
JobAd->LookupInteger(ATTR_LAST_SUSPENSION_TIME,
last_suspension_time);
int delta = time(NULL) - last_suspension_time;
cumulative_suspension_time += delta;
sprintf(tmp, "%s = %d", ATTR_CUMULATIVE_SUSPENSION_TIME,
cumulative_suspension_time);
JobAd->Insert(tmp);
int uncommitted_suspension_time = 0;
JobAd->LookupInteger(ATTR_UNCOMMITTED_SUSPENSION_TIME,
uncommitted_suspension_time);
uncommitted_suspension_time += delta;
JobAd->Assign(ATTR_UNCOMMITTED_SUSPENSION_TIME,uncommitted_suspension_time);
/* set the current suspension time to zero, meaning not suspended */
last_suspension_time = 0;
sprintf(tmp, "%s = %d", ATTR_LAST_SUSPENSION_TIME,
last_suspension_time);
JobAd->Insert(tmp);
break;
}
default:
EXCEPT("record_suspension_hack(): Action event not recognized.");
break;
}
/* Sanity output */
JobAd->LookupInteger(ATTR_TOTAL_SUSPENSIONS, total_suspensions);
dprintf(D_FULLDEBUG,"%s = %d\n", ATTR_TOTAL_SUSPENSIONS, total_suspensions);
JobAd->LookupInteger(ATTR_LAST_SUSPENSION_TIME, last_suspension_time);
dprintf(D_FULLDEBUG, "%s = %d\n", ATTR_LAST_SUSPENSION_TIME,
last_suspension_time);
JobAd->LookupInteger(ATTR_CUMULATIVE_SUSPENSION_TIME,
cumulative_suspension_time);
dprintf(D_FULLDEBUG, "%s = %d\n", ATTR_CUMULATIVE_SUSPENSION_TIME,
cumulative_suspension_time);
/* If we've been asked to perform real time updates of the suspension
information, then connect to the queue and do it here. */
if (param_boolean("REAL_TIME_JOB_SUSPEND_UPDATES", false))
{
dprintf( D_ALWAYS, "Updating suspension info to schedd.\n" );
if (!ConnectQ(schedd, SHADOW_QMGMT_TIMEOUT)) {
/* Since these attributes aren't updated periodically, if
the schedd is busy and a resume event update is lost,
the the job will be marked suspended when it really isn't.
The new shadow eventually corrects this via a periodic
update of various calssad attributes, but I
suspect it won't be corrected in the event of a
bad connect here for this shadow. */
dprintf( D_ALWAYS,
"Timeout connecting to schedd. Suspension update lost.\n");
return;
}
SetAttributeInt(Proc->id.cluster, Proc->id.proc,
ATTR_TOTAL_SUSPENSIONS, total_suspensions);
SetAttributeInt(Proc->id.cluster, Proc->id.proc,
ATTR_CUMULATIVE_SUSPENSION_TIME, cumulative_suspension_time);
SetAttributeInt(Proc->id.cluster, Proc->id.proc,
ATTR_LAST_SUSPENSION_TIME, last_suspension_time);
DisconnectQ(NULL);
}
}
int
main(int argc, char *argv[])
{
has_proc = false;
MyString constraint;
Qmgr_connection *q;
int nextarg = 1, cluster=0, proc=0;
bool UseConstraint = false;
MyString schedd_name;
MyString pool_name;
ExprTree* value_expr;
myDistro->Init( argc, argv );
config();
#if !defined(WIN32)
install_sig_handler(SIGPIPE, SIG_IGN );
#endif
if (argc < 2) {
usage(argv[0]);
}
// if -debug is present, it must be first. sigh.
if (argv[nextarg][0] == '-' && argv[nextarg][1] == 'd') {
// output dprintf messages to stderror at TOOL_DEBUG level
dprintf_set_tool_debug("TOOL", 0);
nextarg++;
}
// if it is present, it must be first after debug.
if (argv[nextarg][0] == '-' && argv[nextarg][1] == 'n') {
nextarg++;
// use the given name as the schedd name to connect to
if (argc <= nextarg) {
fprintf(stderr, "%s: -n requires another argument\n",
argv[0]);
exit(1);
}
schedd_name = argv[nextarg];
nextarg++;
}
if (argc <= nextarg) {
usage(argv[0]);
}
// if it is present, it must be just after -n flag
if (argv[nextarg][0] == '-' && argv[nextarg][1] == 'p') {
nextarg++;
if (argc <= nextarg) {
fprintf(stderr, "%s: -pool requires another argument\n",
argv[0]);
exit(1);
}
pool_name = argv[nextarg];
nextarg++;
}
DCSchedd schedd((schedd_name.Length() == 0) ? NULL : schedd_name.Value(),
(pool_name.Length() == 0) ? NULL : pool_name.Value());
if ( schedd.locate() == false ) {
if (schedd_name == "") {
fprintf( stderr, "%s: ERROR: Can't find address of local schedd\n",
argv[0] );
exit(1);
}
if (pool_name == "") {
fprintf( stderr, "%s: No such schedd named %s in local pool\n",
argv[0], schedd_name.Value() );
} else {
fprintf( stderr, "%s: No such schedd named %s in "
"pool %s\n",
argv[0], schedd_name.Value(), pool_name.Value() );
}
exit(1);
}
// Open job queue
q = ConnectQ( schedd.addr(), 0, false, NULL, NULL, schedd.version() );
if( !q ) {
fprintf( stderr, "Failed to connect to queue manager %s\n",
schedd.addr() );
exit(1);
}
if (argc <= nextarg) {
usage(argv[0]);
}
if (isdigit(argv[nextarg][0])) {
char *tmp;
cluster = strtol(argv[nextarg], &tmp, 10);
if (cluster <= 0) {
fprintf( stderr, "Invalid cluster # from %s.\n", argv[nextarg]);
exit(1);
}
if (*tmp == '.') {
proc = strtol(tmp + 1, &tmp, 10);
if (cluster <= 0) {
fprintf( stderr, "Invalid proc # from %s.\n", argv[nextarg]);
exit(1);
}
UseConstraint = false;
has_proc = true;
} else {
constraint.formatstr("(%s == %d)", ATTR_CLUSTER_ID, cluster);
UseConstraint = true;
}
nextarg++;
} else if (!match_prefix(argv[nextarg], "-constraint")) {
constraint.formatstr("(%s == \"%s\")", ATTR_OWNER, argv[nextarg]);
nextarg++;
UseConstraint = true;
}
if (argc <= nextarg) {
usage(argv[0]);
}
while (match_prefix(argv[nextarg], "-constraint")) {
if ( has_proc ){
fprintf(stderr, "condor_qedit: proc_id specified. Ignoring constraint option\n");
nextarg+=2;
continue;
}
nextarg++;
if (argc <= nextarg) {
usage(argv[0]);
}
if ( !UseConstraint ){
constraint = argv[nextarg];
}
else{
constraint = "( " + constraint + " ) && " + argv[nextarg];
}
nextarg++;
UseConstraint = true;
}
if (argc <= nextarg) {
usage(argv[0]);
}
for (; nextarg < argc; nextarg += 2) {
if (argc <= nextarg+1) {
usage(argv[0]);
}
if (ProtectedAttribute(argv[nextarg])) {
fprintf(stderr, "Update of attribute \"%s\" is not allowed.\n",
argv[nextarg]);
fprintf(stderr,
"Transaction failed. No attributes were set.\n");
exit(1);
}
// Check validity of attribute-name
if ( blankline(argv[nextarg]) ||
!IsValidAttrName(argv[nextarg]) )
{
fprintf(stderr,
"Update aborted, illegal attribute-name specified for attribute \"%s\".\n",
argv[nextarg]);
fprintf(stderr,
"Transaction failed. No attributes were set.\n");
exit(1);
}
// Check validity of attribute-value
value_expr = NULL;
if ( blankline(argv[nextarg+1]) ||
!IsValidAttrValue(argv[nextarg+1]) ||
ParseClassAdRvalExpr(argv[nextarg+1], value_expr) )
{
fprintf(stderr,
"Update aborted, illegal attribute-value specified for attribute \"%s\".\n",
argv[nextarg]);
fprintf(stderr,
"Transaction failed. No attributes were set.\n");
exit(1);
}
if (value_expr) delete value_expr;
if (UseConstraint) {
// Try to communicate with the newer protocol first
if (SetAttributeByConstraint(constraint.Value(),
argv[nextarg],
argv[nextarg+1],
SETDIRTY) < 0) {
if (SetAttributeByConstraint(constraint.Value(),
argv[nextarg],
argv[nextarg+1]) < 0) {
fprintf(stderr,
"Failed to set attribute \"%s\" by constraint: %s\n",
argv[nextarg], constraint.Value());
fprintf(stderr,
"Transaction failed. No attributes were set.\n");
exit(1);
}
}
} else {
if (SetAttribute(cluster, proc, argv[nextarg],
argv[nextarg+1], SETDIRTY) < 0) {
fprintf(stderr,
"Failed to set attribute \"%s\" for job %d.%d.\n",
argv[nextarg], cluster, proc);
fprintf(stderr,
"Transaction failed. No attributes were set.\n");
exit(1);
}
}
printf("Set attribute \"%s\".\n", argv[nextarg]);
}
if (!DisconnectQ(q)) {
fprintf(stderr,
"Queue transaction failed. No attributes were set.\n");
exit(1);
}
return 0;
}
void
doContactSchedd()
{
if (command_queue.IsEmpty()) {
daemonCore->Reset_Timer( contactScheddTid, contact_schedd_interval ); // Come back in a min
return;
}
dprintf(D_FULLDEBUG,"in doContactSchedd\n");
SchedDRequest * current_command = NULL;
int error=FALSE;
std::string error_msg;
CondorError errstack;
bool do_reschedule = false;
int failure_line_num = 0;
int failure_errno = 0;
// Try connecting to schedd
DCSchedd dc_schedd ( ScheddAddr, ScheddPool );
if (dc_schedd.error() || !dc_schedd.locate()) {
sprintf( error_msg, "Error locating schedd %s", ScheddAddr );
dprintf( D_ALWAYS, "%s\n", error_msg.c_str() );
// If you can't connect return "Failure" on every job request
command_queue.Rewind();
while (command_queue.Next(current_command)) {
if (current_command->status != SchedDRequest::SDCS_NEW)
continue;
if (current_command->command == SchedDRequest::SDC_STATUS_CONSTRAINED) {
const char * result[] = {
GAHP_RESULT_FAILURE,
error_msg.c_str(),
"0"};
enqueue_result (current_command->request_id, result, 3);
} else if (current_command->command == SchedDRequest::SDC_SUBMIT_JOB) {
const char * result[] = {
GAHP_RESULT_FAILURE,
NULL,
error_msg.c_str() };
enqueue_result (current_command->request_id, result, 3);
} else if (current_command->command == SchedDRequest::SDC_UPDATE_LEASE) {
const char * result[] = {
GAHP_RESULT_FAILURE,
error_msg.c_str(),
NULL };
enqueue_result (current_command->request_id, result, 3);
} else {
const char * result[] = {
GAHP_RESULT_FAILURE,
error_msg.c_str() };
enqueue_result (current_command->request_id, result, 2);
}
current_command->status = SchedDRequest::SDCS_COMPLETED;
}
}
SchedDRequest::schedd_command_type commands [] = {
SchedDRequest::SDC_REMOVE_JOB,
SchedDRequest::SDC_HOLD_JOB,
SchedDRequest::SDC_RELEASE_JOB };
const char * command_titles [] = {
"REMOVE_JOB", "HOLD_JOB", "RELEASE_JOB" };
// REMOVE
// HOLD
// RELEASE
int i=0;
while (i<3) {
StringList id_list;
SimpleList <SchedDRequest*> this_batch;
SchedDRequest::schedd_command_type this_command = commands[i];
const char * this_action = command_titles[i];
const char * this_reason = NULL;
dprintf (D_FULLDEBUG, "Processing %s requests\n", this_action);
error = FALSE;
// Create a batch of commands with the same command type AND the same reason
command_queue.Rewind();
while (command_queue.Next(current_command)) {
if (current_command->status != SchedDRequest::SDCS_NEW)
continue;
if (current_command->command != this_command)
continue;
if ((this_reason != NULL) && (strcmp (current_command->reason, this_reason) != 0))
continue;
if (this_reason == NULL)
this_reason = current_command->reason;
char job_id_buff[30];
sprintf (job_id_buff, "%d.%d",
current_command->cluster_id,
current_command->proc_id);
id_list.append (job_id_buff);
this_batch.Append (current_command);
}
// If we haven't found any....
if (id_list.isEmpty()) {
i++;
continue; // ... then try the next command
}
// Perform the appropriate command on the current batch
ClassAd * result_ad= NULL;
if (this_command == SchedDRequest::SDC_REMOVE_JOB) {
errstack.clear();
result_ad=
dc_schedd.removeJobs (
&id_list,
this_reason,
&errstack);
} else if (this_command == SchedDRequest::SDC_HOLD_JOB) {
errstack.clear();
result_ad=
dc_schedd.holdJobs (
&id_list,
this_reason,
NULL,
&errstack);
} else if (this_command == SchedDRequest::SDC_RELEASE_JOB) {
errstack.clear();
result_ad=
dc_schedd.releaseJobs (
&id_list,
this_reason,
&errstack);
} else {
EXCEPT( "Unexpected command type %d in doContactSchedd",
this_command );
}
// Analyze the result ad
if (!result_ad) {
error = TRUE;
sprintf( error_msg, "Error connecting to schedd %s %s: %s",
ScheddAddr, dc_schedd.addr(), errstack.getFullText() );
}
else {
result_ad->dPrint (D_FULLDEBUG);
if ( this_command == SchedDRequest::SDC_RELEASE_JOB ) {
do_reschedule = true;
}
}
// Go through the batch again, and create responses for each request
this_batch.Rewind();
while (this_batch.Next(current_command)) {
// Check the result
char job_id_buff[30];
if (result_ad && (error == FALSE)) {
sprintf (job_id_buff, "job_%d_%d",
current_command->cluster_id,
current_command->proc_id);
int remove_result;
if (result_ad->LookupInteger (job_id_buff, remove_result)) {
switch (remove_result) {
case AR_ERROR:
error = TRUE;
error_msg = "General Error";
break;
case AR_SUCCESS:
error = FALSE;
break;
case AR_NOT_FOUND:
error = TRUE;
error_msg = "Job not found";
break;
case AR_BAD_STATUS:
error = TRUE;
error_msg = "Bad job status";
break;
case AR_ALREADY_DONE:
error = TRUE;
error_msg = "Already done";
break;
case AR_PERMISSION_DENIED:
error = TRUE;
error_msg = "Permission denied";
break;
default:
error = TRUE;
error_msg = "Unknown Result";
} // hctiws
} else {
error_msg = "Unable to get result";
} // fi lookup result for job
} // fi error == FALSE
if (error) {
dprintf (D_ALWAYS, "Error (operation: %s) %d.%d: %s\n",
this_action,
current_command->cluster_id,
current_command->proc_id,
error_msg.c_str());
const char * result[2];
result[0] = GAHP_RESULT_FAILURE;
result[1] = error_msg.c_str();
enqueue_result (current_command->request_id, result, 2);
} else {
dprintf (D_ALWAYS, "Succeess (operation: %s) %d.%d\n",
this_action,
current_command->cluster_id,
current_command->proc_id);
const char * result[2];
result[0] = GAHP_RESULT_SUCCESS;
result[1] = NULL;
enqueue_result (current_command->request_id, result, 2);
} // fi error
// Mark the status
current_command->status = SchedDRequest::SDCS_COMPLETED;
} // elihw this_batch
if ( result_ad ) {
delete result_ad;
}
}
dprintf (D_FULLDEBUG, "Processing JOB_STAGE_IN requests\n");
// JOB_STAGE_IN
int MAX_BATCH_SIZE=1; // This should be a config param
SimpleList <SchedDRequest*> stage_in_batch;
do {
stage_in_batch.Clear();
command_queue.Rewind();
while (command_queue.Next(current_command)) {
if (current_command->status != SchedDRequest::SDCS_NEW)
continue;
if (current_command->command != SchedDRequest::SDC_JOB_STAGE_IN)
continue;
dprintf (D_ALWAYS, "Adding %d.%d to STAGE_IN batch\n",
current_command->cluster_id,
current_command->proc_id);
stage_in_batch.Append (current_command);
if (stage_in_batch.Number() >= MAX_BATCH_SIZE)
break;
}
if (stage_in_batch.Number() > 0) {
ClassAd ** array = new ClassAd*[stage_in_batch.Number()];
i=0;
stage_in_batch.Rewind();
while (stage_in_batch.Next(current_command)) {
array[i++] = current_command->classad;
}
error = FALSE;
errstack.clear();
if (!dc_schedd.spoolJobFiles( stage_in_batch.Number(),
array,
&errstack )) {
error = TRUE;
sprintf( error_msg, "Error sending files to schedd %s: %s", ScheddAddr, errstack.getFullText() );
dprintf( D_ALWAYS, "%s\n", error_msg.c_str() );
}
delete [] array;
stage_in_batch.Rewind();
while (stage_in_batch.Next(current_command)) {
current_command->status = SchedDRequest::SDCS_COMPLETED;
if (error) {
const char * result[] = {
GAHP_RESULT_FAILURE,
error_msg.c_str() };
enqueue_result (current_command->request_id, result, 2);
} else {
const char * result[] = {
GAHP_RESULT_SUCCESS,
NULL };
enqueue_result (current_command->request_id, result, 2);
}
} // elihw (command_queue)
} // fi has STAGE_IN requests
} while (stage_in_batch.Number() > 0);
dprintf (D_FULLDEBUG, "Processing JOB_STAGE_OUT requests\n");
// JOB_STAGE_OUT
SimpleList <SchedDRequest*> stage_out_batch;
command_queue.Rewind();
while (command_queue.Next(current_command)) {
if (current_command->status != SchedDRequest::SDCS_NEW)
continue;
if (current_command->command != SchedDRequest::SDC_JOB_STAGE_OUT)
continue;
stage_out_batch.Append (current_command);
}
if (stage_out_batch.Number() > 0) {
std::string constraint = "";
stage_out_batch.Rewind();
int jobsexpected = stage_out_batch.Number();
while (stage_out_batch.Next(current_command)) {
sprintf_cat( constraint, "(ClusterId==%d&&ProcId==%d)||",
current_command->cluster_id,
current_command->proc_id );
}
constraint += "False";
error = FALSE;
errstack.clear();
int jobssent;
if (!dc_schedd.receiveJobSandbox( constraint.c_str(),
&errstack, &jobssent )) {
error = TRUE;
sprintf( error_msg, "Error receiving files from schedd %s: %s",
ScheddAddr, errstack.getFullText() );
dprintf( D_ALWAYS, "%s\n", error_msg.c_str() );
}
if(error == FALSE && jobssent != jobsexpected) {
error = TRUE;
sprintf( error_msg, "Schedd %s didn't send expected files",
ScheddAddr );
dprintf (D_ALWAYS, "Transfered files for %d jobs but got files for %d jobs. (Schedd %s with contraint %s\n", jobsexpected, jobssent, ScheddAddr, constraint.c_str());
}
stage_out_batch.Rewind();
while (stage_out_batch.Next(current_command)) {
current_command->status = SchedDRequest::SDCS_COMPLETED;
if (error) {
const char * result[] = {
GAHP_RESULT_FAILURE,
error_msg.c_str() };
enqueue_result (current_command->request_id, result, 2);
} else {
const char * result[] = {
GAHP_RESULT_SUCCESS,
NULL };
enqueue_result (current_command->request_id, result, 2);
}
} // elihw (command_queue)
} // fi has STAGE_OUT requests
dprintf (D_FULLDEBUG, "Processing JOB_REFRESH_PROXY requests\n");
CondorVersionInfo ver_info(dc_schedd.version());
bool delegate_credential;
if ( ver_info.built_since_version(6,7,19) &&
param_boolean( "DELEGATE_JOB_GSI_CREDENTIALS", true ) ) {
delegate_credential = true;
} else {
delegate_credential = false;
}
// JOB_REFRESH_PROXY
command_queue.Rewind();
while (command_queue.Next(current_command)) {
if (current_command->status != SchedDRequest::SDCS_NEW)
continue;
if (current_command->command != SchedDRequest::SDC_JOB_REFRESH_PROXY)
continue;
time_t expiration_time = GetDesiredDelegatedJobCredentialExpiration(current_command->classad);
time_t result_expiration_time = 0;
bool result;
errstack.clear();
if ( delegate_credential ) {
result = dc_schedd.delegateGSIcredential(
current_command->cluster_id,
current_command->proc_id,
current_command->proxy_file,
expiration_time,
&result_expiration_time,
&errstack );
// Currently, we do not propagate the actual resulting
// expiration time back to the gridmanager. We
// probably should.
} else {
result = dc_schedd.updateGSIcredential(
current_command->cluster_id,
current_command->proc_id,
current_command->proxy_file,
&errstack );
}
current_command->status = SchedDRequest::SDCS_COMPLETED;
if (result == false) {
sprintf( error_msg, "Error refreshing proxy to schedd %s: %s",
ScheddAddr, errstack.getFullText() );
dprintf( D_ALWAYS, "%s\n", error_msg.c_str() );
const char * result_to_queue[] = {
GAHP_RESULT_FAILURE,
error_msg.c_str() };
enqueue_result (current_command->request_id, result_to_queue, 2);
} else {
const char * result_to_queue[] = {
GAHP_RESULT_SUCCESS,
NULL };
enqueue_result (current_command->request_id, result_to_queue, 2);
}
}
// Now do all the QMGMT transactions
error = FALSE;
// Try connecting to the queue
Qmgr_connection * qmgr_connection;
if ((qmgr_connection = ConnectQ(dc_schedd.addr(), QMGMT_TIMEOUT, false, NULL, NULL, dc_schedd.version() )) == NULL) {
error = TRUE;
sprintf( error_msg, "Error connecting to schedd %s", ScheddAddr );
dprintf( D_ALWAYS, "%s\n", error_msg.c_str() );
} else {
errno = 0;
AbortTransaction(); // Just so we can call BeginTransaction() in the loop
if ( errno == ETIMEDOUT ) {
failure_line_num = __LINE__;
failure_errno = errno;
goto contact_schedd_disconnect;
}
}
dprintf (D_FULLDEBUG, "Processing UPDATE_CONSTRAINED/UDATE_JOB requests\n");
// UPDATE_CONSTRAINED
// UDATE_JOB
command_queue.Rewind();
while (command_queue.Next(current_command)) {
if (current_command->status != SchedDRequest::SDCS_NEW)
continue;
if ((current_command->command != SchedDRequest::SDC_UPDATE_CONSTRAINED) &&
(current_command->command != SchedDRequest::SDC_UPDATE_JOB))
continue;
if (qmgr_connection == NULL)
goto update_report_result;
error = FALSE;
errno = 0;
BeginTransaction();
if ( errno == ETIMEDOUT ) {
failure_line_num = __LINE__;
failure_errno = errno;
goto contact_schedd_disconnect;
}
current_command->classad->ResetExpr();
ExprTree *tree;
const char *lhstr, *rhstr;
while( current_command->classad->NextExpr(lhstr, tree) ) {
rhstr = ExprTreeToString( tree );
if( !lhstr || !rhstr) {
sprintf( error_msg, "ERROR: ClassAd problem in Updating by constraint %s",
current_command->constraint );
dprintf( D_ALWAYS, "%s\n", error_msg.c_str() );
error = TRUE;
} else {
if (current_command->command == SchedDRequest::SDC_UPDATE_CONSTRAINED) {
if( SetAttributeByConstraint(current_command->constraint,
lhstr,
rhstr) == -1 ) {
if ( errno == ETIMEDOUT ) {
failure_line_num = __LINE__;
failure_errno = errno;
goto contact_schedd_disconnect;
}
sprintf( error_msg, "ERROR: Failed (errno=%d) to SetAttributeByConstraint %s=%s for constraint %s",
errno, lhstr, rhstr, current_command->constraint );
dprintf( D_ALWAYS, "%s\n", error_msg.c_str() );
error = TRUE;
}
} else if (current_command->command == SchedDRequest::SDC_UPDATE_JOB) {
if( SetAttribute(current_command->cluster_id,
current_command->proc_id,
lhstr,
rhstr) == -1 ) {
if ( errno == ETIMEDOUT ) {
failure_line_num = __LINE__;
failure_errno = errno;
goto contact_schedd_disconnect;
}
sprintf( error_msg, "ERROR: Failed to SetAttribute() %s=%s for job %d.%d",
lhstr, rhstr, current_command->cluster_id, current_command->proc_id);
dprintf( D_ALWAYS, "%s\n", error_msg.c_str() );
error = TRUE;
}
}
}
if (error)
break;
} // elihw classad
update_report_result:
if (error) {
const char * result[] = {
GAHP_RESULT_FAILURE,
error_msg.c_str() };
//RemoteCommitTransaction();
enqueue_result (current_command->request_id, result, 2);
current_command->status = SchedDRequest::SDCS_COMPLETED;
if ( qmgr_connection != NULL ) {
errno = 0;
AbortTransaction();
if ( errno == ETIMEDOUT ) {
failure_line_num = __LINE__;
failure_errno = errno;
goto contact_schedd_disconnect;
}
}
} else {
if ( RemoteCommitTransaction() < 0 ) {
failure_line_num = __LINE__;
failure_errno = errno;
goto contact_schedd_disconnect;
}
const char * result[] = {
GAHP_RESULT_SUCCESS,
NULL };
enqueue_result (current_command->request_id, result, 2);
current_command->status = SchedDRequest::SDCS_COMPLETED;
} // fi
} // elihw
dprintf (D_FULLDEBUG, "Processing UPDATE_LEASE requests\n");
// UPDATE_LEASE
command_queue.Rewind();
while (command_queue.Next(current_command)) {
error = FALSE;
if (current_command->status != SchedDRequest::SDCS_NEW)
continue;
if (current_command->command != SchedDRequest::SDC_UPDATE_LEASE)
continue;
std::string success_job_ids="";
if (qmgr_connection == NULL) {
sprintf( error_msg, "Error connecting to schedd %s", ScheddAddr );
error = TRUE;
} else {
error = FALSE;
errno = 0;
BeginTransaction();
if ( errno == ETIMEDOUT ) {
failure_line_num = __LINE__;
failure_errno = errno;
goto contact_schedd_disconnect;
}
for (i=0; i<current_command->num_jobs; i++) {
time_t time_now = time(NULL);
int duration =
current_command->expirations[i].expiration - time_now;
dprintf (D_FULLDEBUG,
"Job %d.%d SetTimerAttribute=%d\n",
current_command->expirations[i].cluster,
current_command->expirations[i].proc,
duration);
if (SetTimerAttribute (current_command->expirations[i].cluster,
current_command->expirations[i].proc,
ATTR_TIMER_REMOVE_CHECK,
duration) < 0) {
if ( errno == ETIMEDOUT ) {
failure_line_num = __LINE__;
failure_errno = errno;
goto contact_schedd_disconnect;
}
dprintf (D_ALWAYS,
"Unable to SetTimerAttribute(%d, %d), errno=%d\n",
current_command->expirations[i].cluster,
current_command->expirations[i].proc,
errno);
} else {
// Append job id to the result line
if (success_job_ids.length() > 0)
success_job_ids += ",";
sprintf_cat( success_job_ids,
"%d.%d",
current_command->expirations[i].cluster,
current_command->expirations[i].proc);
}
} //rof jobs for request
} // fi error
if (error) {
const char * result[] = {
GAHP_RESULT_FAILURE,
error_msg.c_str(),
NULL
};
//RemoteCommitTransaction();
enqueue_result (current_command->request_id, result, 3);
current_command->status = SchedDRequest::SDCS_COMPLETED;
if ( qmgr_connection != NULL ) {
errno = 0;
AbortTransaction();
if ( errno == ETIMEDOUT ) {
failure_line_num = __LINE__;
failure_errno = errno;
goto contact_schedd_disconnect;
}
}
} else {
if ( RemoteCommitTransaction() < 0 ) {
failure_line_num = __LINE__;
failure_errno = errno;
goto contact_schedd_disconnect;
}
const char * result[] = {
GAHP_RESULT_SUCCESS,
NULL,
success_job_ids.length()?success_job_ids.c_str():NULL
};
enqueue_result (current_command->request_id, result, 3);
current_command->status = SchedDRequest::SDCS_COMPLETED;
} // fi
} // elihw UPDATE_LEASE requests
dprintf (D_FULLDEBUG, "Processing SUBMIT_JOB requests\n");
// SUBMIT_JOB
command_queue.Rewind();
while (command_queue.Next(current_command)) {
if (current_command->status != SchedDRequest::SDCS_NEW)
continue;
if (current_command->command != SchedDRequest::SDC_SUBMIT_JOB)
continue;
int ClusterId = -1;
int ProcId = -1;
if (qmgr_connection == NULL) {
error = TRUE;
goto submit_report_result;
}
errno = 0;
BeginTransaction();
if ( errno == ETIMEDOUT ) {
failure_line_num = __LINE__;
failure_errno = errno;
goto contact_schedd_disconnect;
}
error = FALSE;
if ((ClusterId = NewCluster()) >= 0) {
ProcId = NewProc (ClusterId);
}
if ( errno == ETIMEDOUT ) {
failure_line_num = __LINE__;
failure_errno = errno;
goto contact_schedd_disconnect;
}
if ( ClusterId < 0 ) {
error = TRUE;
error_msg = "Unable to create a new job cluster";
dprintf( D_ALWAYS, "%s\n", error_msg.c_str() );
} else if ( ProcId < 0 ) {
error = TRUE;
error_msg = "Unable to create a new job proc";
dprintf( D_ALWAYS, "%s\n", error_msg.c_str() );
}
if ( ClusterId == -2 || ProcId == -2 ) {
error = TRUE;
error_msg =
"Number of submitted jobs would exceed MAX_JOBS_SUBMITTED\n";
dprintf( D_ALWAYS, "%s\n", error_msg.c_str() );
}
// Adjust the argument/environment syntax based on the version
// of the schedd we are talking to.
if( error == FALSE) {
CondorVersionInfo version_info(dc_schedd.version());
ArgList arglist;
MyString arg_error_msg;
Env env_obj;
MyString env_error_msg;
if(!arglist.AppendArgsFromClassAd(current_command->classad,&arg_error_msg) ||
! arglist.InsertArgsIntoClassAd(current_command->classad,&version_info,&arg_error_msg))
{
sprintf( error_msg,
"ERROR: ClassAd problem in converting arguments to syntax "
"for schedd (version=%s): %s\n",
dc_schedd.version() ? dc_schedd.version() : "NULL",
arg_error_msg.Value());
dprintf( D_ALWAYS,"%s\n", error_msg.c_str() );
error = TRUE;
}
if(!env_obj.MergeFrom(current_command->classad,&env_error_msg) ||
!env_obj.InsertEnvIntoClassAd(current_command->classad,&env_error_msg,NULL,&version_info))
{
sprintf( error_msg,
"ERROR: Failed to convert environment to target syntax"
" for schedd (version %s): %s\n",
dc_schedd.version() ? dc_schedd.version() : "NULL",
env_error_msg.Value());
dprintf( D_ALWAYS, "%s\n", error_msg.c_str() );
error = TRUE;
}
}
if( error == FALSE ) {
// See the comment in the function body of ExpandInputFileList
// for an explanation of what is going on here.
MyString transfer_input_error_msg;
if( !FileTransfer::ExpandInputFileList( current_command->classad, transfer_input_error_msg ) ) {
dprintf( D_ALWAYS, "%s\n", transfer_input_error_msg.Value() );
error = TRUE;
}
}
if ( error == FALSE ) {
current_command->classad->Assign(ATTR_CLUSTER_ID, ClusterId);
current_command->classad->Assign(ATTR_PROC_ID, ProcId);
// Special case for the job lease
int expire_time;
if ( current_command->classad->LookupInteger( ATTR_TIMER_REMOVE_CHECK, expire_time ) ) {
if ( SetTimerAttribute( ClusterId, ProcId,
ATTR_TIMER_REMOVE_CHECK,
expire_time - time(NULL) ) == -1 ) {
if ( errno == ETIMEDOUT ) {
failure_line_num = __LINE__;
failure_errno = errno;
goto contact_schedd_disconnect;
}
sprintf( error_msg, "ERROR: Failed to SetTimerAttribute %s=%ld for job %d.%d",
ATTR_TIMER_REMOVE_CHECK, expire_time - time(NULL), ClusterId, ProcId );
dprintf( D_ALWAYS, "%s\n", error_msg.c_str() );
error = TRUE;
goto submit_report_result;
}
current_command->classad->Delete( ATTR_TIMER_REMOVE_CHECK );
}
// Set all the classad attribute on the remote classad
current_command->classad->ResetExpr();
ExprTree *tree;
const char *lhstr, *rhstr;
while( current_command->classad->NextExpr(lhstr, tree) ) {
rhstr = ExprTreeToString( tree );
if( !lhstr || !rhstr) {
sprintf( error_msg, "ERROR: ClassAd problem in Updating by constraint %s",
current_command->constraint );
dprintf( D_ALWAYS, "%s\n", error_msg.c_str() );
error = TRUE;
} else if( SetAttribute (ClusterId, ProcId,
lhstr,
rhstr) == -1 ) {
if ( errno == ETIMEDOUT ) {
failure_line_num = __LINE__;
failure_errno = errno;
goto contact_schedd_disconnect;
}
sprintf( error_msg, "ERROR: Failed to SetAttribute %s=%s for job %d.%d",
lhstr, rhstr, ClusterId, ProcId );
dprintf( D_ALWAYS, "%s\n", error_msg.c_str() );
error = TRUE;
}
if (error) break;
} // elihw classad
} // fi error==FALSE
submit_report_result:
char job_id_buff[30];
sprintf (job_id_buff, "%d.%d", ClusterId, ProcId);
if (error) {
const char * result[] = {
GAHP_RESULT_FAILURE,
job_id_buff,
error_msg.c_str() };
enqueue_result (current_command->request_id, result, 3);
if ( qmgr_connection != NULL ) {
errno = 0;
AbortTransaction();
if ( errno == ETIMEDOUT ) {
failure_line_num = __LINE__;
failure_errno = errno;
goto contact_schedd_disconnect;
}
}
current_command->status = SchedDRequest::SDCS_COMPLETED;
} else {
if ( RemoteCommitTransaction() < 0 ) {
failure_line_num = __LINE__;
failure_errno = errno;
goto contact_schedd_disconnect;
}
const char * result[] = {
GAHP_RESULT_SUCCESS,
job_id_buff,
NULL };
enqueue_result (current_command->request_id, result, 3);
current_command->status = SchedDRequest::SDCS_COMPLETED;
}
} // elihw
dprintf (D_FULLDEBUG, "Processing STATUS_CONSTRAINED requests\n");
// STATUS_CONSTRAINED
command_queue.Rewind();
while (command_queue.Next(current_command)) {
if (current_command->status != SchedDRequest::SDCS_NEW)
continue;
if (current_command->command != SchedDRequest::SDC_STATUS_CONSTRAINED)
continue;
if (qmgr_connection != NULL) {
SimpleList <MyString *> matching_ads;
error = FALSE;
ClassAd *next_ad;
ClassAdList adlist;
// Only use GetAllJobsByConstraint if remote schedd is
// 6.9.5 or newer. Previous versions either did not
// support this call, or they closed the Qmgmt connection
// as a side-effect of this call.
if( ver_info.built_since_version(6,9,5) ) {
dprintf( D_FULLDEBUG, "Calling GetAllJobsByConstraint(%s)\n",
current_command->constraint );
// NOTE: this could be made more efficient if we knew
// the list of attributes to query. For lack of that,
// we just get all attributes.
GetAllJobsByConstraint( current_command->constraint, "", adlist);
}
else {
// This is the old latency-prone method.
dprintf( D_FULLDEBUG, "Calling GetNextJobByConstraint(%s)\n",
current_command->constraint );
next_ad = GetNextJobByConstraint( current_command->constraint, 1 );
while( next_ad != NULL ) {
adlist.Insert( next_ad );
next_ad = GetNextJobByConstraint( current_command->constraint, 0 );
}
}
// NOTE: ClassAdList will deallocate the ClassAds in it
adlist.Rewind();
while( (next_ad=adlist.Next()) ) {
MyString * da_buffer = new MyString(); // Use a ptr to avoid excessive copying
if ( useXMLClassads ) {
ClassAdXMLUnparser unparser;
unparser.SetUseCompactSpacing(true);
unparser.Unparse (next_ad, *da_buffer);
} else {
NewClassAdUnparser unparser;
unparser.SetUseCompactSpacing(true);
unparser.Unparse (next_ad, *da_buffer);
}
matching_ads.Append (da_buffer);
}
if ( errno == ETIMEDOUT ) {
failure_line_num = __LINE__;
failure_errno = errno;
goto contact_schedd_disconnect;
}
// now output this list of classads into a result
const char ** result = new const char* [matching_ads.Length() + 3];
std::string _ad_count;
sprintf( _ad_count, "%d", matching_ads.Length() );
int count=0;
result[count++] = GAHP_RESULT_SUCCESS;
result[count++] = NULL;
result[count++] = _ad_count.c_str();
MyString *next_string;
matching_ads.Rewind();
while (matching_ads.Next(next_string)) {
result[count++] = next_string->Value();
}
enqueue_result (current_command->request_id, result, count);
current_command->status = SchedDRequest::SDCS_COMPLETED;
// Cleanup
matching_ads.Rewind();
while (matching_ads.Next(next_string)) {
delete next_string;
}
//CommitTransaction();
delete [] result;
}
else {
const char * result[] = {
GAHP_RESULT_FAILURE,
error_msg.c_str(),
"0" };
//RemoteCommitTransaction();
enqueue_result (current_command->request_id, result, 3);
current_command->status = SchedDRequest::SDCS_COMPLETED;
}
} //elihw
contact_schedd_disconnect:
if ( qmgr_connection != NULL ) {
DisconnectQ (qmgr_connection, FALSE);
}
if ( failure_line_num ) {
// We had an error talking to the schedd. Take all of our
// incomplete commands and mark them as failed.
// TODO Consider retrying these commands, rather than
// immediately marking them as failed.
if ( failure_errno == ETIMEDOUT ) {
dprintf( D_ALWAYS, "Timed out talking to schedd at line %d in "
"doContactSchedd()\n", failure_line_num );
sprintf( error_msg, "Timed out talking to schedd" );
} else {
dprintf( D_ALWAYS, "Error talking to schedd at line %d in "
"doContactSchedd(), errno=%d (%s)\n", failure_line_num,
failure_errno, strerror(failure_errno) );
sprintf( error_msg, "Error talking to schedd" );
}
command_queue.Rewind();
while (command_queue.Next(current_command)) {
if ( current_command->status != SchedDRequest::SDCS_NEW ) {
continue;
}
switch( current_command->command ) {
case SchedDRequest::SDC_UPDATE_JOB:
case SchedDRequest::SDC_UPDATE_CONSTRAINED:
{
const char *result[2] = { GAHP_RESULT_FAILURE, error_msg.c_str() };
enqueue_result (current_command->request_id, result, 2);
current_command->status = SchedDRequest::SDCS_COMPLETED;
}
break;
case SchedDRequest::SDC_UPDATE_LEASE:
{
const char *result[3] = { GAHP_RESULT_FAILURE, error_msg.c_str(), NULL };
enqueue_result (current_command->request_id, result, 3);
current_command->status = SchedDRequest::SDCS_COMPLETED;
}
break;
case SchedDRequest::SDC_SUBMIT_JOB:
{
const char *result[3] = { GAHP_RESULT_FAILURE, "-1.-1", error_msg.c_str() };
enqueue_result (current_command->request_id, result, 3);
current_command->status = SchedDRequest::SDCS_COMPLETED;
}
break;
case SchedDRequest::SDC_STATUS_CONSTRAINED:
{
const char *result[3] = { GAHP_RESULT_FAILURE, error_msg.c_str(), "0" };
enqueue_result (current_command->request_id, result, 3);
current_command->status = SchedDRequest::SDCS_COMPLETED;
}
break;
default:
// Do nothing
;
}
}
}
if ( do_reschedule ) {
dc_schedd.reschedule();
}
// Write all of our results to our parent.
flush_results();
dprintf (D_FULLDEBUG, "Finishing doContactSchedd()\n");
// Clean up the list
command_queue.Rewind();
while (command_queue.Next(current_command)) {
if (current_command->status == SchedDRequest::SDCS_COMPLETED) {
command_queue.DeleteCurrent();
delete current_command;
}
}
// Come back soon..
// QUESTION: Should this always be a fixed time period?
daemonCore->Reset_Timer( contactScheddTid, contact_schedd_interval );
}
