int
IntStateQuery()
{
/*
qstat -f
Job Id: 11.cream-12.pd.infn.it
Job_Name = cream_579184706
job_state = R
ctime = Wed Apr 23 11:39:55 2008
exec_host = cream-wn-029.pn.pd.infn.it/0
*/
/*
Filled entries:
batch_id
wn_addr
status
udate
Filled by submit script:
blah_id
Unfilled entries:
exitreason
*/
FILE *fp;
char *line=NULL;
char **token;
int maxtok_t=0;
job_registry_entry en;
int ret;
char *timestamp;
time_t tmstampepoch;
char *batch_str=NULL;
char *wn_str=NULL;
char *twn_str=NULL;
char *status_str=NULL;
char *ex_str=NULL;
int ex_code=0;
char *cp=NULL;
char *command_string=NULL;
job_registry_entry *ren=NULL;
int first=TRUE;
time_t now;
char *string_now=NULL;
command_string=make_message("%s%s/qstat -f",batch_command,pbs_binpath);
fp = popen(command_string,"r");
en.status=UNDEFINED;
JOB_REGISTRY_ASSIGN_ENTRY(en.wn_addr,"\0");
JOB_REGISTRY_ASSIGN_ENTRY(en.exitreason,"\0");
JOB_REGISTRY_ASSIGN_ENTRY(en.updater_info,"\0");
en.exitcode=-1;
bupdater_free_active_jobs(&bact);
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';
}
do_log(debuglogfile, debug, 3, "%s: line in IntStateQuery is:%s\n",argv0,line);
now=time(0);
string_now=make_message("%d",now);
if(line && strstr(line,"Job Id: ")){
if(!first && en.status!=UNDEFINED && ren && ren->status!=REMOVED && ren->status!=COMPLETED){
if ((ret=job_registry_update_recn_select(rha, &en, ren->recnum,
JOB_REGISTRY_UPDATE_WN_ADDR|
JOB_REGISTRY_UPDATE_STATUS|
JOB_REGISTRY_UPDATE_UDATE|
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 {
if(ret==JOB_REGISTRY_SUCCESS){
if (en.status == REMOVED || en.status == COMPLETED) {
do_log(debuglogfile, debug, 2, "%s: registry update in IntStateQuery for: jobid=%s wn=%s status=%d exitcode=%d\n",argv0,en.batch_id,en.wn_addr,en.status,en.exitcode);
job_registry_unlink_proxy(rha, &en);
}else{
do_log(debuglogfile, debug, 2, "%s: registry update in IntStateQuery for: jobid=%s wn=%s status=%d\n",argv0,en.batch_id,en.wn_addr,en.status);
}
}
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 IntStateQuery\n",argv0);
}
}
}
en.status = UNDEFINED;
JOB_REGISTRY_ASSIGN_ENTRY(en.wn_addr,"\0");
JOB_REGISTRY_ASSIGN_ENTRY(en.exitreason,"\0");
JOB_REGISTRY_ASSIGN_ENTRY(en.updater_info,"\0");
en.exitcode=-1;
}
maxtok_t = strtoken(line, ':', &token);
batch_str=strdel(token[1]," ");
JOB_REGISTRY_ASSIGN_ENTRY(en.batch_id,batch_str);
JOB_REGISTRY_ASSIGN_ENTRY(en.updater_info,string_now);
en.exitcode=-1;
bupdater_push_active_job(&bact, en.batch_id);
free(batch_str);
freetoken(&token,maxtok_t);
if(!first) free(ren);
if ((ren=job_registry_get(rha, en.batch_id)) == NULL){
fprintf(stderr,"Get of record returns error for %s ",en.batch_id);
perror("");
}
first=FALSE;
}else if(line && strstr(line,"job_state = ")){
maxtok_t = strtoken(line, '=', &token);
status_str=strdel(token[1]," ");
if(status_str && strcmp(status_str,"Q")==0){
en.status=IDLE;
en.exitcode=-1;
JOB_REGISTRY_ASSIGN_ENTRY(en.wn_addr,"\0");
}else if(status_str && strcmp(status_str,"W")==0){
en.status=IDLE;
en.exitcode=-1;
}else if(status_str && strcmp(status_str,"R")==0){
en.status=RUNNING;
en.exitcode=-1;
}else if(status_str && strcmp(status_str,"C")==0){
en.status=COMPLETED;
JOB_REGISTRY_ASSIGN_ENTRY(en.exitreason,"\0");
}else if(status_str && strcmp(status_str,"H")==0){
en.status=HELD;
en.exitcode=-1;
JOB_REGISTRY_ASSIGN_ENTRY(en.wn_addr,"\0");
}
free(status_str);
freetoken(&token,maxtok_t);
}else if(line && strstr(line,"unable to run job")){
en.status=IDLE;
en.exitcode=-1;
}else if(line && strstr(line,"exit_status = ")){
maxtok_t = strtoken(line, '=', &token);
ex_str=strdel(token[1]," ");
ex_code=atoi(ex_str);
if(ex_code==0){
en.exitcode=0;
}else if(ex_code==271){
en.status=REMOVED;
en.exitcode=-999;
}else{
en.exitcode=ex_code;
}
free(ex_str);
freetoken(&token,maxtok_t);
}else if(line && strstr(line,"exec_host = ")){
maxtok_t = strtoken(line, '=', &token);
twn_str=strdup(token[1]);
if(twn_str == NULL){
sysfatal("strdup failed for twn_str in IntStateQuery: %r");
}
freetoken(&token,maxtok_t);
maxtok_t = strtoken(twn_str, '/', &token);
wn_str=strdel(token[0]," ");
JOB_REGISTRY_ASSIGN_ENTRY(en.wn_addr,wn_str);
free(twn_str);
free(wn_str);
freetoken(&token,maxtok_t);
}else if(line && strstr(line,"mtime = ")){
maxtok_t = strtoken(line, ' ', &token);
timestamp=make_message("%s %s %s %s %s",token[2],token[3],token[4],token[5],token[6]);
tmstampepoch=str2epoch(timestamp,"L");
free(timestamp);
en.udate=tmstampepoch;
freetoken(&token,maxtok_t);
}
free(line);
free(string_now);
}
pclose(fp);
}
if(en.status!=UNDEFINED && ren && ren->status!=REMOVED && ren->status!=COMPLETED){
if ((ret=job_registry_update_recn_select(rha, &en, ren->recnum,
JOB_REGISTRY_UPDATE_WN_ADDR|
JOB_REGISTRY_UPDATE_STATUS|
JOB_REGISTRY_UPDATE_UDATE|
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 {
if(ret==JOB_REGISTRY_SUCCESS){
if (en.status == REMOVED || en.status == COMPLETED) {
do_log(debuglogfile, debug, 2, "%s: registry update in IntStateQuery for: jobid=%s wn=%s status=%d exitcode=%d\n",argv0,en.batch_id,en.wn_addr,en.status,en.exitcode);
job_registry_unlink_proxy(rha, &en);
}else{
do_log(debuglogfile, debug, 2, "%s: registry update in IntStateQuery for: jobid=%s wn=%s status=%d\n",argv0,en.batch_id,en.wn_addr,en.status);
}
}
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 IntStateQuery\n",argv0);
}
}
}
}
free(ren);
free(command_string);
return 0;
}
int
ReceiveUpdateFromNetwork()
{
char *proxy_path, *proxy_subject;
int timeout_ms = 0;
int ent, ret, prret, rhret;
job_registry_entry *nen;
job_registry_entry *ren;
proxy_path = NULL;
proxy_subject = NULL;
while (nen = job_registry_receive_update(remupd_pollset, remupd_nfds,timeout_ms, &proxy_subject, &proxy_path)){
JOB_REGISTRY_ASSIGN_ENTRY(nen->subject_hash,"\0");
JOB_REGISTRY_ASSIGN_ENTRY(nen->proxy_link,"\0");
if ((ren=job_registry_get(rha, nen->batch_id)) == NULL){
if ((ret=job_registry_append(rha, nen)) < 0){
fprintf(stderr,"%s: Warning: job_registry_append returns %d: ",argv0,ret);
perror("");
}
}else{
if(ren->subject_hash!=NULL && strlen(ren->subject_hash) && ren->proxy_link!=NULL && strlen(ren->proxy_link)){
JOB_REGISTRY_ASSIGN_ENTRY(nen->subject_hash,ren->subject_hash);
JOB_REGISTRY_ASSIGN_ENTRY(nen->proxy_link,ren->proxy_link);
}else{
if (proxy_path != NULL && strlen(proxy_path) > 0){
prret = job_registry_set_proxy(rha, nen, proxy_path);
if (prret < 0){
do_log(debuglogfile, debug, 1, "%s: warning: setting proxy to %s\n",argv0,proxy_path);
fprintf(stderr,"%s: warning: setting proxy to %s: ",argv0,proxy_path);
perror("");
/* Make sure we don't renew non-existing proxies */
nen->renew_proxy = 0;
}
free(proxy_path);
nen->subject_hash[0] = '\000';
if (proxy_subject != NULL && strlen(proxy_subject) > 0){
job_registry_compute_subject_hash(nen, proxy_subject);
rhret = job_registry_record_subject_hash(rha, nen->subject_hash, proxy_subject, TRUE);
if (rhret < 0){
do_log(debuglogfile, debug, 1, "%s: warning: recording proxy subject %s (hash %s)\n",argv0, proxy_subject, nen->subject_hash);
fprintf(stderr,"%s: warning: recording proxy subject %s (hash %s): ",argv0, proxy_subject, nen->subject_hash);
perror("");
}
}
free(proxy_subject);
}
}
if(job_registry_need_update(ren,nen,JOB_REGISTRY_UPDATE_ALL)){
if ((ret=job_registry_update(rha, nen)) < 0){
fprintf(stderr,"%s: Warning: job_registry_update returns %d: ",argv0,ret);
perror("");
}
}
}
free(nen);
}
return 0;
}
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;
}
int
PollDB()
{
FILE *fd;
job_registry_entry *en;
job_registry_handle *rha;
job_registry_handle *rhc;
char *buffer=NULL;
char *cdate=NULL;
time_t now;
int maxtok,i,maxtokl,j;
char **tbuf;
char **lbuf;
int len=0,flen=0;
struct stat sbuf;
int rc;
char *regfile;
char *cp=NULL;
int to_sleep=FALSE;
int skip_reg_open=FALSE;
int ret;
rha=job_registry_init(registry_file, BY_BATCH_ID);
if (rha == NULL){
do_log(debuglogfile, debug, 1, "%s: Error initialising job registry %s\n",argv0,registry_file);
fprintf(stderr,"%s: Error initialising job registry %s :",argv0,registry_file);
perror("");
}
for(;;){
now=time(NULL);
to_sleep=TRUE;
/* cycle over connections: sleep if startnotify, startnotifyjob and sentendonce are not set.
If startnotifyjob is set the conn is served.
*/
for(i=0; i<MAX_CONNECTIONS; i++){
if(!connections[i].startnotify && !connections[i].startnotifyjob && !(connections[i].firstnotify && connections[i].sentendonce)) continue;
if(connections[i].startnotify) to_sleep=FALSE;
if(connections[i].startnotifyjob){
to_sleep=FALSE;
rhc=job_registry_init(registry_file, BY_USER_PREFIX);
if (rhc == NULL){
do_log(debuglogfile, debug, 1, "%s: Error initialising job registry %s\n",argv0,registry_file);
fprintf(stderr,"%s: Error initialising job registry %s :",argv0,registry_file);
perror("");
}
do_log(debuglogfile, debug, 2, "%s:Job list for notification:%s\n",argv0,connections[i].joblist_string);
maxtok=strtoken(connections[i].joblist_string,',',&tbuf);
for(j=0;j<maxtok;j++){
if ((en=job_registry_get(rhc, tbuf[j])) != NULL){
buffer=ComposeClassad(en);
}else{
if(remupd_conf == NULL){
cdate=iepoch2str(now);
maxtokl=strtoken(tbuf[j],'_',&lbuf);
if(lbuf[1]){
if ((cp = strrchr (lbuf[1], '\n')) != NULL){
*cp = '\0';
}
if ((cp = strrchr (lbuf[1], '\r')) != NULL){
*cp = '\0';
}
buffer=make_message("[BlahJobName=\"%s\"; ClientJobId=\"%s\"; JobStatus=4; JwExitCode=999; ExitReason=\"BUpdater is not able to find the job anymore\"; Reason=\"BUpdater is not able to find the job anymore\"; ChangeTime=\"%s\"; ]\n",tbuf[j],lbuf[1],cdate);
}
freetoken(&lbuf,maxtokl);
free(cdate);
}else{
maxtokl=strtoken(tbuf[j],':',&lbuf);
JOB_REGISTRY_ASSIGN_ENTRY(en->batch_id,lbuf[0]);
JOB_REGISTRY_ASSIGN_ENTRY(en->blah_id,lbuf[1]);
freetoken(&lbuf,maxtokl);
en->status = 0;
if ((ret=job_registry_append(rhc, en))<0){
if(ret != JOB_REGISTRY_NOT_FOUND){
fprintf(stderr,"Update of record returns %d: ",ret);
perror("");
}
}else{
if(ret==JOB_REGISTRY_SUCCESS){
do_log(debuglogfile, debug, 2, "%s: registry append in PollDB for: jobid=%s blahjobid=%s\n",argv0,en->batch_id,en->blah_id);
}
}
}
}
free(en);
len=strlen(buffer);
if(connections[i].finalbuffer != NULL){
flen=strlen(connections[i].finalbuffer);
}else{
flen=0;
}
connections[i].finalbuffer = realloc(connections[i].finalbuffer,flen+len+2);
if (connections[i].finalbuffer == NULL){
sysfatal("can't realloc finalbuffer in PollDB: %r");
}
if(flen==0){
connections[i].finalbuffer[0]='\000';
}
strcat(connections[i].finalbuffer,buffer);
free(buffer);
}
freetoken(&tbuf,maxtok);
if(connections[i].finalbuffer != NULL){
if(NotifyCream(connections[i].finalbuffer,&connections[i])!=-1){
/* change last notification time */
connections[i].lastnotiftime=now;
connections[i].startnotifyjob=FALSE;
}
free(connections[i].finalbuffer);
connections[i].finalbuffer=NULL;
}
job_registry_destroy(rhc);
}
if(connections[i].firstnotify && connections[i].sentendonce){
to_sleep=FALSE;
if(NotifyCream("NTFDATE/END\n",&connections[i])!=-1){
connections[i].startnotify=TRUE;
connections[i].sentendonce=FALSE;
connections[i].firstnotify=FALSE;
connections[i].startnotifyjob=FALSE;
}
}
}
if(to_sleep){
sleep(loop_interval);
continue;
}
regfile=make_message("%s/registry",registry_file);
rc=stat(regfile,&sbuf);
free(regfile);
skip_reg_open=TRUE;
for(i=0; i<MAX_CONNECTIONS; i++){
if(sbuf.st_mtime>=connections[i].lastnotiftime){
skip_reg_open=FALSE;
break;
}
}
if(skip_reg_open){
do_log(debuglogfile, debug, 3, "Skip registry opening: mtime:%d lastn:%d\n",sbuf.st_mtime,connections[i].lastnotiftime);
sleep(loop_interval);
continue;
}
do_log(debuglogfile, debug, 3, "Normal registry opening\n");
fd = job_registry_open(rha, "r");
if (fd == NULL)
{
do_log(debuglogfile, debug, 1, "%s: Error opening job registry %s\n",argv0,registry_file);
fprintf(stderr,"%s: Error opening job registry %s :",argv0,registry_file);
perror("");
sleep(loop_interval);
continue;
}
if (job_registry_rdlock(rha, fd) < 0)
{
do_log(debuglogfile, debug, 1, "%s: Error read locking registry %s\n",argv0,registry_file);
fprintf(stderr,"%s: Error read locking registry %s :",argv0,registry_file);
perror("");
sleep(loop_interval);
continue;
}
while ((en = job_registry_get_next(rha, fd)) != NULL)
{
for(i=0; i<MAX_CONNECTIONS; i++){
if(connections[i].creamfilter==NULL) continue;
if(en->mdate >= connections[i].lastnotiftime && en->mdate < now && en->user_prefix && strstr(en->user_prefix,connections[i].creamfilter)!=NULL && strlen(en->updater_info)>0)
{
buffer=ComposeClassad(en);
len=strlen(buffer);
if(connections[i].finalbuffer != NULL){
flen=strlen(connections[i].finalbuffer);
}else{
flen=0;
}
connections[i].finalbuffer = realloc(connections[i].finalbuffer,flen+len+2);
if (connections[i].finalbuffer == NULL){
sysfatal("can't realloc finalbuffer in PollDB: %r");
}
if(flen==0){
connections[i].finalbuffer[0]='\000';
}
strcat(connections[i].finalbuffer,buffer);
free(buffer);
}
}
free(en);
}
for(i=0; i<MAX_CONNECTIONS; i++){
if(connections[i].finalbuffer != NULL){
if(NotifyCream(connections[i].finalbuffer,&connections[i])!=-1){
/* change last notification time */
connections[i].lastnotiftime=now;
}
free(connections[i].finalbuffer);
connections[i].finalbuffer=NULL;
}
}
fclose(fd);
sleep(loop_interval);
}
job_registry_destroy(rha);
return 0;
}
int
IntStateQuery()
{
FILE *fp;
char *line=NULL;
char **token;
char **token_l;
char **token_e;
int maxtok_t=0;
int maxtok_l=0;
int maxtok_e=0;
job_registry_entry en;
int ret;
time_t tmstampepoch;
char *cp=NULL;
char *batch_str=NULL;
char *command_string=NULL;
job_registry_entry *ren=NULL;
int isresumed=FALSE;
int first=TRUE;
time_t now;
char *string_now=NULL;
command_string=make_message("%s/scontrol -a show jobid",slurm_binpath);
fp = popen(command_string,"r");
en.status=UNDEFINED;
JOB_REGISTRY_ASSIGN_ENTRY(en.wn_addr,"\0");
JOB_REGISTRY_ASSIGN_ENTRY(en.exitreason,"\0");
JOB_REGISTRY_ASSIGN_ENTRY(en.updater_info,"\0");
en.exitcode=-1;
bupdater_free_active_jobs(&bact);
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';
}
do_log(debuglogfile, debug, 3, "%s: line in IntStateQuery is:%s\n",argv0,line);
now=time(0);
string_now=make_message("%d",now);
maxtok_t = strtoken(line, ' ', &token);
if(line && strstr(line,"JobId=")){
isresumed=FALSE;
if(!first && en.status!=UNDEFINED && ren && ren->status!=REMOVED && ren->status!=COMPLETED){
if ((ret=job_registry_update_recn_select(rha, &en, ren->recnum,
JOB_REGISTRY_UPDATE_WN_ADDR|
JOB_REGISTRY_UPDATE_STATUS|
JOB_REGISTRY_UPDATE_UDATE|
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 {
if(ret==JOB_REGISTRY_SUCCESS){
if (en.status == REMOVED || en.status == COMPLETED) {
do_log(debuglogfile, debug, 2, "%s: registry update in IntStateQuery for: jobid=%s creamjobid=%s wn=%s status=%d exitcode=%d\n",argv0,en.batch_id,en.user_prefix,en.wn_addr,en.status,en.exitcode);
job_registry_unlink_proxy(rha, &en);
}else{
do_log(debuglogfile, debug, 2, "%s: registry update in IntStateQuery for: jobid=%s creamjobid=%s wn=%s status=%d\n",argv0,en.batch_id,en.user_prefix,en.wn_addr,en.status);
}
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 IntStateQuery\n",argv0);
}
}
}
}
en.status = UNDEFINED;
JOB_REGISTRY_ASSIGN_ENTRY(en.exitreason,"\0");
JOB_REGISTRY_ASSIGN_ENTRY(en.updater_info,"\0");
JOB_REGISTRY_ASSIGN_ENTRY(en.wn_addr,"\0");
en.exitcode=-1;
}
en.status = UNDEFINED;
maxtok_l = strtoken(token[0], '=', &token_l);
batch_str=strdup(token_l[1]);
JOB_REGISTRY_ASSIGN_ENTRY(en.batch_id,batch_str);
JOB_REGISTRY_ASSIGN_ENTRY(en.updater_info,string_now);
en.exitcode=-1;
bupdater_push_active_job(&bact, en.batch_id);
do_log(debuglogfile, debug, 4, "%s: bupdater_push_active_job done for %s\n",argv0,en.batch_id);
free(batch_str);
freetoken(&token_l,maxtok_l);
if(!first) free(ren);
if ((ren=job_registry_get(rha, en.batch_id)) == NULL){
fprintf(stderr,"Get of record returns error ");
perror("");
}
if(ren){
if(strlen(ren->updater_info)>0){
en.udate=ren->udate;
}else{
en.udate=time(0);
}
}
first=FALSE;
}else if(line && strstr(line," JobState=")){
if(token[0] && strstr(line,"JobState=")){
maxtok_l = strtoken(token[0], '=', &token_l);
if(token_l[1] && strstr(token_l[1],"PENDING")){
en.status=IDLE;
en.exitcode=-1;
JOB_REGISTRY_ASSIGN_ENTRY(en.updater_info,string_now);
}else if(token_l[1] && strstr(token_l[1],"RUNNING")){
en.status=RUNNING;
en.exitcode=-1;
JOB_REGISTRY_ASSIGN_ENTRY(en.updater_info,string_now);
}else if(token_l[1] && strstr(token_l[1],"COMPLETED")){
en.status=COMPLETED;
en.exitcode=0;
JOB_REGISTRY_ASSIGN_ENTRY(en.updater_info,string_now);
}else if(token_l[1] && strstr(token_l[1],"CANCELLED")){
en.status=REMOVED;
en.exitcode=-999;
JOB_REGISTRY_ASSIGN_ENTRY(en.updater_info,string_now);
}else if(token_l[1] && strstr(token_l[1],"FAILED")){
en.status=COMPLETED;
en.exitcode=0;
JOB_REGISTRY_ASSIGN_ENTRY(en.updater_info,string_now);
}else if(token_l[1] && strstr(token_l[1],"SUSPENDED")){
en.status=HELD;
en.exitcode=-1;
JOB_REGISTRY_ASSIGN_ENTRY(en.updater_info,string_now);
}else if(token_l[1] && strstr(token_l[1],"COMPLETING")){
bupdater_remove_active_job(&bact, en.batch_id);
}
freetoken(&token_l,maxtok_l);
}
}else if(line && strstr(line," BatchHost=")){
if(token[0] && strstr(line,"BatchHost=")){
maxtok_l = strtoken(token[0], '=', &token_l);
if(en.status!=IDLE){
JOB_REGISTRY_ASSIGN_ENTRY(en.wn_addr,token_l[1]);
}
freetoken(&token_l,maxtok_l);
}
}else if(line && strstr(line," ExitCode=")){
if(token[3] && strstr(line,"ExitCode=")){
maxtok_l = strtoken(token[3], '=', &token_l);
maxtok_e = strtoken(token_l[1], ':', &token_e);
if(en.status==COMPLETED){
en.exitcode=atoi(token_e[0]);
}
freetoken(&token_l,maxtok_l);
freetoken(&token_e,maxtok_e);
}
}else if(line && strstr(line," SubmitTime=")){
if(en.status==IDLE){
if(token[0] && strstr(line,"SubmitTime=")){
maxtok_l = strtoken(token[0], '=', &token_l);
tmstampepoch=str2epoch(token_l[1],"N");
en.udate=tmstampepoch;
freetoken(&token_l,maxtok_l);
}
}
}else if(line && strstr(line," StartTime=")){
if(en.status==RUNNING){
if(token[0] && strstr(line,"StartTime=")){
maxtok_l = strtoken(token[0], '=', &token_l);
tmstampepoch=str2epoch(token_l[1],"N");
en.udate=tmstampepoch;
freetoken(&token_l,maxtok_l);
}
}
if(en.status==COMPLETED || en.status==REMOVED){
if(token[1] && strstr(line,"EndTime=")){
maxtok_l = strtoken(token[1], '=', &token_l);
tmstampepoch=str2epoch(token_l[1],"N");
en.udate=tmstampepoch;
freetoken(&token_l,maxtok_l);
}
}
}else if(line && strstr(line," SuspendTime=")){
if(en.status==HELD){
if(token[1] && strstr(line,"SuspendTime=")){
maxtok_l = strtoken(token[1], '=', &token_l);
tmstampepoch=str2epoch(token_l[1],"N");
en.udate=tmstampepoch;
freetoken(&token_l,maxtok_l);
}
}
}
free(line);
free(string_now);
freetoken(&token,maxtok_t);
}
pclose(fp);
}
if(en.status!=UNDEFINED && ren && ren->status!=REMOVED && ren->status!=COMPLETED){
if ((ret=job_registry_update_recn_select(rha, &en, ren->recnum,
JOB_REGISTRY_UPDATE_WN_ADDR|
JOB_REGISTRY_UPDATE_STATUS|
JOB_REGISTRY_UPDATE_UDATE|
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 {
if(ret==JOB_REGISTRY_SUCCESS){
if (en.status == REMOVED || en.status == COMPLETED) {
do_log(debuglogfile, debug, 2, "%s: registry update in IntStateQuery for: jobid=%s creamjobid=%s wn=%s status=%d exitcode=%d\n",argv0,en.batch_id,en.user_prefix,en.wn_addr,en.status,en.exitcode);
job_registry_unlink_proxy(rha, &en);
}else{
do_log(debuglogfile, debug, 2, "%s: registry update in IntStateQuery for: jobid=%s creamjobid=%s wn=%s status=%d\n",argv0,en.batch_id,en.user_prefix,en.wn_addr,en.status);
}
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 IntStateQuery\n",argv0);
}
}
}
}
}
free(ren);
free(command_string);
return 0;
}
