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 );
}
void
publishNotifyEmail( FILE* mailer, char* buf, PROC* proc )
{
double rutime, rstime, lutime, lstime; /* remote/local user/sys times */
double trtime, tltime; /* Total remote/local time */
double real_time;
float run_time = 0.0;
time_t arch_time=0; /* time_t is 8 bytes some archs and 4 bytes on other
archs, and this means that doing a (time_t*)
cast on & of a 4 byte int makes my life hell.
So we fix it by assigning the time we want to
a real time_t variable, then using ctime()
to convert it to a string */
fprintf(mailer, "Your condor job " );
#if defined(NEW_PROC)
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] );
#else
ArgList args;
MyString args_string;
args.AppendArgsV1or2(proc->args_v1or2,NULL);
args.GetArgsStringForDisplay(&args_string);
fprintf(mailer, "%s %s ", proc->cmd, args_string.Value() );
#endif
fprintf(mailer, "%s\n\n", buf );
job_report_display_errors( mailer );
char* user = NULL;
char* full_user = NULL;
if( ! JobAd->LookupString(ATTR_USER, &user) ) {
JobAd->LookupString(ATTR_OWNER, &user);
}
if( user ) {
full_user = email_check_domain( user, JobAd );
fprintf( mailer, "\nSubmitted by: %s\n", full_user );
free( full_user );
full_user = NULL;
free( user );
user = NULL;
}
arch_time = proc->q_date;
fprintf(mailer, "\nTime:\n");
fprintf(mailer, "\tSubmitted at: %s", ctime(&arch_time));
if( proc->completion_date ) {
real_time = proc->completion_date - proc->q_date;
arch_time = proc->completion_date;
fprintf(mailer, "\tCompleted at: %s\n",
ctime(&arch_time));
fprintf(mailer, "\tReal Time: %s\n",
d_format_time(real_time));
if (JobAd) {
JobAd->LookupFloat(ATTR_JOB_REMOTE_WALL_CLOCK, run_time);
}
run_time += proc->completion_date - ShadowBDate;
fprintf(mailer, "\tRun Time: %s\n",
d_format_time(run_time));
if (CommittedTime > 0) {
fprintf(mailer, "\tCommitted Time: %s\n",
d_format_time(CommittedTime));
}
}
fprintf( mailer, "\n" );
#if defined(NEW_PROC)
rutime = proc->remote_usage[0].ru_utime.tv_sec;
rstime = proc->remote_usage[0].ru_stime.tv_sec;
trtime = rutime + rstime;
#else
rutime = proc->remote_usage.ru_utime.tv_sec;
rstime = proc->remote_usage.ru_stime.tv_sec;
trtime = rutime + rstime;
#endif
lutime = proc->local_usage.ru_utime.tv_sec;
lstime = proc->local_usage.ru_stime.tv_sec;
tltime = lutime + lstime;
fprintf(mailer, "\tRemote User Time: %s\n", d_format_time(rutime) );
fprintf(mailer, "\tRemote System Time: %s\n", d_format_time(rstime) );
fprintf(mailer, "\tTotal Remote Time: %s\n\n", d_format_time(trtime));
fprintf(mailer, "\tLocal User Time: %s\n", d_format_time(lutime) );
fprintf(mailer, "\tLocal System Time: %s\n", d_format_time(lstime) );
fprintf(mailer, "\tTotal Local Time: %s\n\n", d_format_time(tltime));
if( tltime >= 1.0 ) {
fprintf(mailer, "\tLeveraging Factor: %2.1f\n", trtime / tltime);
}
fprintf(mailer, "\tVirtual Image Size: %d Kilobytes\n", proc->image_size);
if (NumCkpts > 0) {
fprintf(mailer, "\nCheckpoints written: %d\n", NumCkpts);
fprintf(mailer, "Checkpoint restarts: %d\n", NumRestarts);
}
// TotalBytesSent and TotalBytesRecvd are from the shadow's
// perspective, and we want to display the stats from the job's
// perspective. Note also that TotalBytesSent and TotalBytesRecvd
// don't include our current run, so we need to include the
// stats from our syscall_sock (if we have one) and the BytesSent
// and BytesRecvd variables. This is ugly and confusing, which
// explains why I keep getting it wrong. :-(
float network_bytes = TotalBytesSent + BytesSent;
if (syscall_sock) {
network_bytes += syscall_sock->get_bytes_sent();
}
if (network_bytes > 0.0) {
fprintf(mailer,"\nNetwork:\n");
fprintf(mailer,"\t%s read\n",metric_units(network_bytes));
network_bytes = TotalBytesRecvd + BytesRecvd;
if (syscall_sock) {
network_bytes += syscall_sock->get_bytes_recvd();
}
fprintf(mailer,"\t%s written\n",metric_units(network_bytes));
}
if (JobIsStandard()) {
job_report_display_file_info( mailer, (int) run_time );
job_report_display_calls( mailer );
}
email_custom_attributes( mailer, JobAd );
}
