int AssignFinalState(char *batchid){
job_registry_entry en;
int ret,i;
time_t now;
now=time(0);
JOB_REGISTRY_ASSIGN_ENTRY(en.batch_id,batchid);
en.status=COMPLETED;
en.exitcode=999;
en.udate=now;
JOB_REGISTRY_ASSIGN_ENTRY(en.wn_addr,"\0");
JOB_REGISTRY_ASSIGN_ENTRY(en.exitreason,"\0");
if ((ret=job_registry_update(rha, &en)) < 0){
if(ret != JOB_REGISTRY_NOT_FOUND){
fprintf(stderr,"Update of record %d returns %d: ",i,ret);
perror("");
}
} else {
do_log(debuglogfile, debug, 2, "%s: registry update in AssignStateQuery for: jobid=%s creamjobid=%s status=%d\n",argv0,en.batch_id,en.user_prefix,en.status);
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 AssignFinalState\n",argv0);
}
}
}
return 0;
}
int AssignState (char *element, char *status, char *exit, char *reason, char *wn, char *udate){
char **id_element;
job_registry_entry en;
time_t now;
char *string_now=NULL;
int i=0;
int n=strtoken(element, '.', &id_element);
int iret;
if(id_element[0]){
JOB_REGISTRY_ASSIGN_ENTRY(en.batch_id,id_element[0]);
en.status=atoi(status);
en.exitcode=atoi(exit);
JOB_REGISTRY_ASSIGN_ENTRY(en.wn_addr,wn);
JOB_REGISTRY_ASSIGN_ENTRY(en.exitreason,reason);
now=time(0);
string_now=make_message("%d",now);
JOB_REGISTRY_ASSIGN_ENTRY(en.updater_info,string_now)
en.udate=now;
free(string_now);
}else{
if((element=calloc(STR_CHARS,1)) == 0){
sysfatal("can't malloc cmd in GetAndSend: %r");
}
}
if ((iret=job_registry_update(rha, &en)) < 0){
fprintf(stderr,"Update of record returns %d: \nJobId: %d", iret,en.batch_id);
perror("");
}else{
if (en.status == REMOVED || en.status == COMPLETED){
job_registry_unlink_proxy(rha, &en);
}
}
freetoken(&id_element,n);
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(char *query,char *queryStates, char *query_err){
char line[STR_CHARS],fail[6],qExit[10],qFailed[10],qHostname[100],qStatus[2],command_string[100];
char **saveptr1,**saveptr2,**list_query,**list_queryStates;
FILE *file_output;
int numQuery=0,numQueryStates=0,j=0,l=0,cont=0,cont2=0, nq=0;
time_t now;
char string_now[11];
job_registry_entry en;
int iret;
numQuery=strtoken(query,' ',&list_query);
nq=numQuery;
numQueryStates=strtoken(queryStates,' ',&list_queryStates);
if (numQuery!=numQueryStates) return 1;
sprintf(command_string,"%s/qstat -u '*'",sge_binpath);
if (debug) do_log(debuglogfile, debug, 1, "+-+line 433, command_string:%s\n",command_string);
//load in qstatJob list of jobids from qstat command exec
file_output = popen(command_string,"r");
if (file_output == NULL) return 0;
while (fgets(line,sizeof(line), file_output) != NULL){
cont=strtoken(line, ' ', &saveptr1);
if ((strcmp(saveptr1[0],"job-ID")!=0)&&(strncmp(saveptr1[0],"-",1)!=0)){
for (l=0;l<nq;l++){
if (strcmp(list_query[l],saveptr1[0])==0){
if (strcmp(list_queryStates[l],saveptr1[4])!=0){
now=time(0);
sprintf(string_now,"%d",now);
if (strcmp(saveptr1[4],"u")==0){
JOB_REGISTRY_ASSIGN_ENTRY(en.batch_id,list_query[l]);
en.status=0;
en.exitcode=0;
JOB_REGISTRY_ASSIGN_ENTRY(en.wn_addr,"");
JOB_REGISTRY_ASSIGN_ENTRY(en.exitreason,"0");
JOB_REGISTRY_ASSIGN_ENTRY(en.updater_info,string_now)
en.udate=now;
if ((iret=job_registry_update(rha, &en)) < 0){
fprintf(stderr,"Update of record returns %d: \nJobId: %d", iret,en.batch_id);
perror("");
}
}
if (strcmp(saveptr1[4],"q")==0){
JOB_REGISTRY_ASSIGN_ENTRY(en.batch_id,list_query[l]);
en.status=1;
en.exitcode=0;
JOB_REGISTRY_ASSIGN_ENTRY(en.wn_addr,"");
JOB_REGISTRY_ASSIGN_ENTRY(en.exitreason,"0");
JOB_REGISTRY_ASSIGN_ENTRY(en.updater_info,string_now)
en.udate=now;
if ((iret=job_registry_update(rha, &en)) < 0){
fprintf(stderr,"Update of record returns %d: \nJobId: %d", iret,en.batch_id);
perror("");
}
}
if (strcmp(saveptr1[4],"r")==0){
cont2=strtoken(saveptr1[7], '@', &saveptr2);
JOB_REGISTRY_ASSIGN_ENTRY(en.batch_id,list_query[l]);
en.status=2;
en.exitcode=0;
JOB_REGISTRY_ASSIGN_ENTRY(en.wn_addr,saveptr2[1]);
JOB_REGISTRY_ASSIGN_ENTRY(en.exitreason,"0");
JOB_REGISTRY_ASSIGN_ENTRY(en.updater_info,string_now)
en.udate=now;
if ((iret=job_registry_update(rha, &en)) < 0){
fprintf(stderr,"Update of record returns %d: \nJobId: %d", iret,en.batch_id);
perror("");
}
freetoken(&saveptr2,cont2);
}
if ((strcmp(saveptr1[4],"hr")==0)||strcmp(saveptr1[4],"hqw")==0){
JOB_REGISTRY_ASSIGN_ENTRY(en.batch_id,list_query[l]);
en.status=5;
en.exitcode=0;
JOB_REGISTRY_ASSIGN_ENTRY(en.wn_addr,"");
JOB_REGISTRY_ASSIGN_ENTRY(en.exitreason,"0");
JOB_REGISTRY_ASSIGN_ENTRY(en.updater_info,string_now)
en.udate=now;
if ((iret=job_registry_update(rha, &en)) < 0){
fprintf(stderr,"Update of record returns %d: \nJobId: %d", iret,en.batch_id);
perror("");
}
}
}
//i must put out element from query
for (j=l;j<nq;j++)
if (list_query[j+1]!=NULL) strcpy(list_query[j],list_query[j+1]);
for (j=l;j<nq;j++)
if (list_queryStates[j+1]!=NULL) strcpy(list_queryStates[j],list_queryStates[j+1]);
nq--;
break;
}
}
}
line[0]='\0';
freetoken(&saveptr1,cont);
}
pclose( file_output );
sprintf(query_err,"\0");
//now we have check in list_query only states that not change status
//because they're not in qstat result
for (l=0; l<nq; l++){
sprintf(command_string,"%s/qacct -j '%s'",sge_binpath,list_query[l]);
if (debug) do_log(debuglogfile, debug, 1, "+-+line 520,command_string:%s\n",command_string);
file_output = popen(command_string,"r");
if (file_output == NULL) return 1;
//if a job number is here means that job was in query previously and
//if now it's not in query and not finished (NULL qstat) it was deleted
//or it's on transition time
if (fgets( line,sizeof(line), file_output )==NULL){
strcat(query_err,list_query[l]);
strcat(query_err," ");
pclose( file_output );
continue;
}
//there is no problem to lost first line with previous fgets, because
//it's only a line of =============================================
while (fgets( line,sizeof(line), file_output )!=NULL){
cont=strtoken(line, ' ', &saveptr1);
if (strcmp(saveptr1[0],"hostname")==0) strcpy(qHostname,saveptr1[1]);;
if (strcmp(saveptr1[0],"failed")==0) strcpy(qFailed,saveptr1[1]);
if (strcmp(saveptr1[0],"exit_status")==0) strcpy(qExit,saveptr1[1]);
freetoken(&saveptr1,cont);
}
pclose( file_output );
now=time(0);
sprintf(string_now,"%d",now);
if ((strcmp(qExit,"137")==0)||(strcmp(qExit,"143")==0)){
JOB_REGISTRY_ASSIGN_ENTRY(en.batch_id,list_query[l]);
en.status=3;
en.exitcode=atoi(qExit);
JOB_REGISTRY_ASSIGN_ENTRY(en.wn_addr,qHostname);
JOB_REGISTRY_ASSIGN_ENTRY(en.exitreason,"");
JOB_REGISTRY_ASSIGN_ENTRY(en.updater_info,string_now)
en.udate=now;
if ((iret=job_registry_update(rha, &en)) < 0){
fprintf(stderr,"Update of record returns %d: \nJobId: %d", iret,en.batch_id);
perror("");
}else job_registry_unlink_proxy(rha, &en);
}else{
JOB_REGISTRY_ASSIGN_ENTRY(en.batch_id,list_query[l]);
en.status=4;
en.exitcode=atoi(qExit);
JOB_REGISTRY_ASSIGN_ENTRY(en.wn_addr,qHostname);
JOB_REGISTRY_ASSIGN_ENTRY(en.exitreason,qFailed);
JOB_REGISTRY_ASSIGN_ENTRY(en.updater_info,string_now)
en.udate=now;
if ((iret=job_registry_update(rha, &en)) < 0){
fprintf(stderr,"Update of record returns %d: \nJobId: %d", iret,en.batch_id);
perror("");
}else job_registry_unlink_proxy(rha, &en);
}
}
freetoken(&list_query,numQuery);
freetoken(&list_queryStates,numQueryStates);
if (debug) do_log(debuglogfile, debug, 1, "+-+query_err:%s\n",query_err);
//now check acumulated error jobids to verify if they are an error or not
if (strcmp(query_err,"\0")!=0){
sleep(60);
cont=0;
int n=0;
char cmd[10]="\0";
cont=strtoken(query_err, ' ', &list_query);
while (n < cont){
if(list_query[n]) strcpy(cmd,list_query[n]);
else return 1;
sprintf(command_string,"%s/qacct -j '%s'",sge_binpath,cmd);
if (debug) do_log(debuglogfile, debug, 1, "+-+line 587 error, command_string:%s\n",command_string);
file_output = popen(command_string,"r");
if (file_output == NULL) return 1;
//if a job number is here means that job was in query previously and
//if now it's not in query and not finished (NULL qstat) it was deleted
if (fgets( line,sizeof(line), file_output )==NULL){
JOB_REGISTRY_ASSIGN_ENTRY(en.batch_id,cmd);
en.status=3;
en.exitcode=3;
JOB_REGISTRY_ASSIGN_ENTRY(en.wn_addr,"");
JOB_REGISTRY_ASSIGN_ENTRY(en.exitreason,"reason=3");
now=time(0);
sprintf(string_now,"%d",now);
JOB_REGISTRY_ASSIGN_ENTRY(en.updater_info,string_now)
en.udate=now;
if ((iret=job_registry_update(rha, &en)) < 0){
fprintf(stderr,"Update of record returns %d: \nJobId: %d", iret,en.batch_id);
perror("");
}else job_registry_unlink_proxy(rha, &en);
pclose( file_output );
n++;
continue;
}
//there is no problem to lost first line with previous fgets, because
//it's only a line of =============================================
while (fgets( line,sizeof(line), file_output )!=NULL){
cont=strtoken(line, ' ', &saveptr1);
if (strcmp(saveptr1[0],"hostname")==0) strcpy(qHostname,saveptr1[1]);
if (strcmp(saveptr1[0],"failed")==0) strcpy(qFailed,saveptr1[1]);
if (strcmp(saveptr1[0],"exit_status")==0) strcpy(qExit,saveptr1[1]);
freetoken(&saveptr1,cont);
}
pclose( file_output );
now=time(0);
sprintf(string_now,"%d",now);
if ((strcmp(qExit,"137")==0)||(strcmp(qExit,"143")==0)){
JOB_REGISTRY_ASSIGN_ENTRY(en.batch_id,cmd);
en.status=3;
en.exitcode=atoi(qExit);
JOB_REGISTRY_ASSIGN_ENTRY(en.wn_addr,qHostname);
JOB_REGISTRY_ASSIGN_ENTRY(en.exitreason,"");
JOB_REGISTRY_ASSIGN_ENTRY(en.updater_info,string_now)
en.udate=now;
if ((iret=job_registry_update(rha, &en)) < 0){
fprintf(stderr,"Update of record returns %d: \nJobId: %d", iret,en.batch_id);
perror("");
}else job_registry_unlink_proxy(rha, &en);
}else{
JOB_REGISTRY_ASSIGN_ENTRY(en.batch_id,cmd);
en.status=4;
en.exitcode=atoi(qExit);
JOB_REGISTRY_ASSIGN_ENTRY(en.wn_addr,qHostname);
JOB_REGISTRY_ASSIGN_ENTRY(en.exitreason,qFailed);
JOB_REGISTRY_ASSIGN_ENTRY(en.updater_info,string_now)
en.udate=now;
if ((iret=job_registry_update(rha, &en)) < 0){
fprintf(stderr,"Update of record returns %d: \nJobId: %d", iret,en.batch_id);
perror("");
}else job_registry_unlink_proxy(rha, &en);
}
n++;
}
freetoken(&list_query,cont);
}
return 0;
}