int
FinalStateQuery(char *input_string, int logs_to_read)
{
/*
tracejob -m -l -a <jobid>
In line:
04/23/2008 11:50:43 S Exit_status=0 resources_used.cput=00:00:01 resources_used.mem=11372kb resources_used.vmem=52804kb
resources_used.walltime=00:10:15
there are:
udate for the final state (04/23/2008 11:50:43):
exitcode Exit_status=
*/
/*
Filled entries:
batch_id (a list of jobid is given, one for each tracejob call)
status (always a final state 3 or 4)
exitcode
udate
Filled by submit script:
blah_id
Unfilled entries:
exitreason
*/
/*
[root@cream-12 server_logs]# tracejob -m -l -a 13
Job: 13.cream-12.pd.infn.it
04/23/2008 11:40:27 S enqueuing into cream_1, state 1 hop 1
04/23/2008 11:40:27 S Job Queued at request of [email protected], owner = [email protected], job name =
cream_365713239, queue = cream_1
04/23/2008 11:40:28 S Job Modified at request of [email protected]
04/23/2008 11:40:28 S Job Run at request of [email protected]
04/23/2008 11:50:43 S Exit_status=0 resources_used.cput=00:00:01 resources_used.mem=11372kb resources_used.vmem=52804kb
resources_used.walltime=00:10:15
04/23/2008 11:50:44 S dequeuing from cream_1, state COMPLETE
*/
FILE *fp;
char *line=NULL;
char **token;
char **jobid;
int maxtok_t=0,maxtok_j=0,k;
job_registry_entry en;
int ret;
char *timestamp;
time_t tmstampepoch;
char *exit_str=NULL;
int failed_count=0;
int time_to_add=0;
time_t now;
char *cp=NULL;
char *command_string=NULL;
char *pbs_spool=NULL;
char *string_now=NULL;
int tracejob_line_counter=0;
do_log(debuglogfile, debug, 3, "%s: input_string in FinalStateQuery is:%s\n",argv0,input_string);
maxtok_j = strtoken(input_string, ':', &jobid);
for(k=0;k<maxtok_j;k++){
if(jobid[k] && strlen(jobid[k])==0) continue;
pbs_spool=(pbs_spoolpath?make_message("-p %s ",pbs_spoolpath):make_message(""));
command_string=make_message("%s%s/tracejob %s-m -l -a -n %d %s",batch_command,pbs_binpath,pbs_spool,logs_to_read,jobid[k]);
free(pbs_spool);
fp = popen(command_string,"r");
do_log(debuglogfile, debug, 3, "%s: command_string in FinalStateQuery is:%s\n",argv0,command_string);
/* en.status is set =0 (UNDEFINED) here and it is tested if it is !=0 before the registry update: the update is done only if en.status is !=0*/
en.status=UNDEFINED;
JOB_REGISTRY_ASSIGN_ENTRY(en.batch_id,jobid[k]);
tracejob_line_counter=0;
if(fp!=NULL){
while(!feof(fp) && (line=get_line(fp))){
if(line && strlen(line)==0){
free(line);
continue;
}
if(tracejob_line_counter>tracejob_max_output){
do_log(debuglogfile, debug, 2, "%s: Tracejob output limit of %d lines reached. Skipping command.\n",argv0,tracejob_max_output);
free(line);
break;
}
if ((cp = strrchr (line, '\n')) != NULL){
*cp = '\0';
tracejob_line_counter++;
}
do_log(debuglogfile, debug, 3, "%s: line in FinalStateQuery is:%s\n",argv0,line);
now=time(0);
string_now=make_message("%d",now);
if(line && (strstr(line,"Job deleted") || (strstr(line,"dequeuing from") && strstr(line,"state RUNNING")))){
maxtok_t = strtoken(line, ' ', &token);
timestamp=make_message("%s %s",token[0],token[1]);
tmstampepoch=str2epoch(timestamp,"A");
free(timestamp);
freetoken(&token,maxtok_t);
en.udate=tmstampepoch;
en.status=REMOVED;
en.exitcode=-999;
JOB_REGISTRY_ASSIGN_ENTRY(en.updater_info,string_now);
JOB_REGISTRY_ASSIGN_ENTRY(en.exitreason,"\0");
}else if(line && strstr(line," Exit_status=") && en.status != REMOVED){
maxtok_t = strtoken(line, ' ', &token);
timestamp=make_message("%s %s",token[0],token[1]);
tmstampepoch=str2epoch(timestamp,"A");
exit_str=strdup(token[3]);
if(exit_str == NULL){
sysfatal("strdup failed for exit_str in FinalStateQuery: %r");
}
free(timestamp);
freetoken(&token,maxtok_t);
if(strstr(exit_str,"Exit_status=")){
maxtok_t = strtoken(exit_str, '=', &token);
if(maxtok_t == 2){
en.exitcode=atoi(token[1]);
freetoken(&token,maxtok_t);
}else{
en.exitcode=-1;
}
}else{
en.exitcode=-1;
}
free(exit_str);
en.udate=tmstampepoch;
en.status=COMPLETED;
JOB_REGISTRY_ASSIGN_ENTRY(en.updater_info,string_now);
JOB_REGISTRY_ASSIGN_ENTRY(en.exitreason,"\0");
}
free(string_now);
free(line);
}
pclose(fp);
}
if(en.status !=UNDEFINED && en.status!=IDLE){
if ((ret=job_registry_update_select(rha, &en,
JOB_REGISTRY_UPDATE_UDATE |
JOB_REGISTRY_UPDATE_STATUS |
JOB_REGISTRY_UPDATE_UPDATER_INFO |
JOB_REGISTRY_UPDATE_EXITCODE |
JOB_REGISTRY_UPDATE_EXITREASON )) < 0){
if(ret != JOB_REGISTRY_NOT_FOUND){
fprintf(stderr,"Update of record returns %d: ",ret);
perror("");
}
} else {
do_log(debuglogfile, debug, 2, "%s: registry update in FinalStateQuery for: jobid=%s exitcode=%d status=%d\n",argv0,en.batch_id,en.exitcode,en.status);
if (en.status == REMOVED || en.status == COMPLETED){
job_registry_unlink_proxy(rha, &en);
}
if (remupd_conf != NULL){
if (ret=job_registry_send_update(remupd_head_send,&en,NULL,NULL)<=0){
do_log(debuglogfile, debug, 2, "%s: Error creating endpoint in FinalStateQuery\n",argv0);
}
}
}
}else{
failed_count++;
}
free(command_string);
}
now=time(0);
if(failed_count>10){
failed_count=10;
}
time_to_add=pow(failed_count,1.5);
next_finalstatequery=now+time_to_add;
do_log(debuglogfile, debug, 3, "%s: next FinalStatequery will be in %d seconds\n",argv0,time_to_add);
freetoken(&jobid,maxtok_j);
return failed_count;
}
int
FinalStateQuery(time_t start_date, int logs_to_read)
{
FILE *fp;
char *line=NULL;
char **token;
char **token_l;
int maxtok_t=0;
int maxtok_l=0;
job_registry_entry en;
int ret;
time_t tmstampepoch;
char *cp=NULL;
char *command_string=NULL;
time_t now;
char *string_now=NULL;
job_registry_entry *ren=NULL;
command_string=make_message("%s/sacct -nap -o JobID,JobName,State,ExitCode,submit,start,end 2>/dev/null",slurm_binpath);
fp = popen(command_string,"r");
do_log(debuglogfile, debug, 3, "%s: command_string in FinalStateQuery is:%s\n",argv0,command_string);
en.status=UNDEFINED;
JOB_REGISTRY_ASSIGN_ENTRY(en.exitreason,"\0");
if(fp!=NULL){
while(!feof(fp) && (line=get_line(fp))){
if(line && strlen(line)==0){
free(line);
continue;
}
if ((cp = strrchr (line, '\n')) != NULL){
*cp = '\0';
}
en.status=UNDEFINED;
do_log(debuglogfile, debug, 3, "%s: line in FinalStateQuery is:%s\n",argv0,line);
now=time(0);
string_now=make_message("%d",now);
maxtok_t = strtoken(line, '|', &token);
JOB_REGISTRY_ASSIGN_ENTRY(en.batch_id,token[0]);
JOB_REGISTRY_ASSIGN_ENTRY(en.updater_info,string_now);
if(token[2] && strstr(token[2],"COMPLETED")){
en.status=COMPLETED;
JOB_REGISTRY_ASSIGN_ENTRY(en.updater_info,string_now);
}else if(token[2] && strstr(token[2],"CANCELLED")){
en.status=REMOVED;
en.exitcode=-999;
JOB_REGISTRY_ASSIGN_ENTRY(en.updater_info,string_now);
}else if(token[2] && strstr(token[2],"FAILED")){
en.status=COMPLETED;
JOB_REGISTRY_ASSIGN_ENTRY(en.updater_info,string_now);
}
if(!(token[6] && strstr(token[6],"Unknown"))){
tmstampepoch=str2epoch(token[6],"N");
en.udate=tmstampepoch;
}
if(en.status==COMPLETED){
maxtok_l = strtoken(token[3], ':', &token_l);
en.exitcode=atoi(token_l[0]);
freetoken(&token_l,maxtok_l);
}
if ((ren=job_registry_get(rha, en.batch_id)) == NULL){
fprintf(stderr,"Get of record returns error ");
perror("");
}
if(en.status!=UNDEFINED && en.status!=IDLE && ren && ren->status!=REMOVED && ren->status!=COMPLETED){
if ((ret=job_registry_update_select(rha, &en,
JOB_REGISTRY_UPDATE_UDATE |
JOB_REGISTRY_UPDATE_STATUS |
JOB_REGISTRY_UPDATE_UPDATER_INFO |
JOB_REGISTRY_UPDATE_EXITCODE |
JOB_REGISTRY_UPDATE_EXITREASON )) < 0){
if(ret != JOB_REGISTRY_NOT_FOUND){
fprintf(stderr,"Update of record returns %d: ",ret);
perror("");
}
} else {
do_log(debuglogfile, debug, 2, "%s: f registry update in FinalStateQuery for: jobid=%s creamjobid=%s status=%d\n",argv0,en.batch_id,en.user_prefix,en.status);
if (en.status == REMOVED || en.status == COMPLETED){
job_registry_unlink_proxy(rha, &en);
}
if (remupd_conf != NULL){
if ((ret=job_registry_send_update(remupd_head_send,&en,NULL,NULL))<=0){
do_log(debuglogfile, debug, 2, "%s: Error creating endpoint in FinalStateQuery\n",argv0);
}
}
}
}
free(string_now);
free(line);
freetoken(&token,maxtok_t);
free(ren);
}
pclose(fp);
}
free(command_string);
return 0;
}