int
GetJobList(char *buffer, char **joblist_string)
{
int maxtok;
char **tbuf;
char *cp=NULL;
maxtok=strtoken(buffer,'/',&tbuf);
if(tbuf[1]){
*joblist_string = strdup(tbuf[1]);
if(*joblist_string == NULL){
sysfatal("strdup failed for joblist_string in GetJobList: %r");
}
if ((cp = strrchr(*joblist_string, '\n')) != NULL) {
*cp = '\0';
}
if ((cp = strrchr(*joblist_string, '\r')) != NULL){
*cp = '\0';
}
}
freetoken(&tbuf,maxtok);
return 0;
}
char *
ComposeClassad(job_registry_entry *en)
{
char *strudate=NULL;
char *buffer=NULL;
char *wn=NULL;
char *excode=NULL;
char *exreas=NULL;
char *blahid=NULL;
char *clientid=NULL;
int maxtok;
char **tbuf;
char *cp=NULL;
if((buffer=calloc(STR_CHARS,1)) == 0){
sysfatal("can't malloc buffer in PollDB: %r");
}
strudate=iepoch2str(en->udate);
sprintf(buffer,"[BatchJobId=\"%s\"; JobStatus=%d; ChangeTime=\"%s\";",en->batch_id, en->status, strudate);
free(strudate);
if (strlen(en->wn_addr) > 0){
wn=make_message(" WorkerNode=\"%s\";",en->wn_addr);
strcat(buffer,wn);
free(wn);
}
if (en->status == 3 || en->status == 4){
excode=make_message(" JwExitCode=%d; Reason=\"reason=%d\";", en->exitcode, en->exitcode);
strcat(buffer,excode);
free(excode);
}
if (strlen(en->exitreason) > 0){
exreas=make_message(" ExitReason=\"%s\";", en->exitreason);
strcat(buffer,exreas);
free(exreas);
}
if (strlen(en->user_prefix) > 0){
maxtok=strtoken(en->user_prefix,'_',&tbuf);
if(tbuf[1]){
if ((cp = strrchr (tbuf[1], '\n')) != NULL){
*cp = '\0';
}
if ((cp = strrchr (tbuf[1], '\r')) != NULL){
*cp = '\0';
}
clientid=make_message(" ClientJobId=\"%s\";",tbuf[1]);
}
blahid=make_message("%s BlahJobName=\"%s\";",clientid, en->user_prefix);
strcat(buffer,blahid);
free(blahid);
freetoken(&tbuf,maxtok);
free(clientid);
}
strcat(buffer,"]\n");
return buffer;
}
int
NotifyStart(char *buffer, time_t *lastnotiftime)
{
int maxtok;
char **tbuf;
char *cp=NULL;
char *notifdate=NULL;
maxtok = strtoken(buffer,'/',&tbuf);
if(tbuf[1]){
notifdate=strdup(tbuf[1]);
if(notifdate == NULL){
sysfatal("strdup failed for notifdate in NotifyStart: %r");
}
if ((cp = strrchr (notifdate, '\n')) != NULL){
*cp = '\0';
}
if ((cp = strrchr (notifdate, '\r')) != NULL){
*cp = '\0';
}
}
freetoken(&tbuf,maxtok);
*lastnotiftime = str2epoch(notifdate,"S");
free(notifdate);
return 0;
}
int
GetFilter(char *buffer, const int conn_c, char **creamfilter)
{
int maxtok;
char **tbuf;
char *cp=NULL;
char * out_buf;
maxtok = strtoken(buffer,'/',&tbuf);
if(tbuf[1]){
*creamfilter = make_message("%s",tbuf[1]);
if(*creamfilter == NULL){
sysfatal("strdup failed for creamfilter in GetFilter: %r");
}
if ((cp = strrchr (*creamfilter, '\n')) != NULL) {
*cp = '\0';
}
if ((cp = strrchr (*creamfilter, '\r')) != NULL) {
*cp = '\0';
}
out_buf = make_message("CREAMFILTER set to %s\n", *creamfilter);
} else {
out_buf = make_message("CREAMFILTER ERROR\n");
}
Writeline(conn_c, out_buf, strlen(out_buf));
do_log(debuglogfile, debug, 1, "Sent Reply for CREAMFILTER command:%s\n",out_buf);
freetoken(&tbuf,maxtok);
free(out_buf);
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
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
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
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 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;
}
int main(int argc, char *argv[]){
FILE *fd;
job_registry_entry *en;
time_t now;
time_t purge_time=0;
char constraint[JOBID_MAX_LEN+1];
char constraint2[5];
char *query=NULL;
char *queryStates=NULL;
char *query_err=NULL;
char *pidfile=NULL;
char string_now[11];
char *tpath;
int version=0;
int tmptim;
int finstr_len=0;
int loop_interval=DEFAULT_LOOP_INTERVAL;
int fsq_ret=0;
int c;
int confirm_time=0;
static int help;
static int short_help;
while (1) {
static struct option long_options[] =
{
{"help", no_argument, &help, 1},
{"usage", no_argument, &short_help, 1},
{"nodaemon", no_argument, 0, 'o'},
{"version", no_argument, 0, 'v'},
{0, 0, 0, 0}
};
int option_index = 0;
c = getopt_long (argc, argv, "vo",long_options, &option_index);
if (c == -1){
break;
}
switch (c)
{
case 0:
if (long_options[option_index].flag != 0){
break;
}
case 'v':
version=1;
break;
case 'o':
nodmn=1;
break;
case '?':
break;
default:
abort ();
}
}
//check if another instance is running
char **ptr;
char out[3];
fgets(out, sizeof(out),popen("ps -d | grep -c BUpdaterSGE","r"));
strtoken(out,'\n',&ptr);
if (strcmp(ptr[0],"1")!=0){
fprintf(stderr,"There is another instance of BUpdaterSGE running.\nExiting ...\n");
return -1;
}
freetoken(&ptr,1);
if(help){
usage();
}
if(short_help){
short_usage();
}
argv0 = argv[0];
signal(SIGHUP,sighup);
if(version) {
printf("%s Version: %s\n",progname,VERSION);
exit(EXIT_SUCCESS);
}
/* Checking configuration */
check_config_file("UPDATER");
cha = config_read(NULL);
if (cha == NULL)
{
fprintf(stderr,"Error reading config: ");
perror("");
return -1;
}
config_setenv(NULL);
ret = config_get("bupdater_child_poll_timeout",cha);
if (ret != NULL){
tmptim=atoi(ret->value);
if (tmptim > 0) bfunctions_poll_timeout = tmptim*1000;
}
ret = config_get("bupdater_debug_level",cha);
if (ret != NULL){
debug=atoi(ret->value);
}
ret = config_get("bupdater_debug_logfile",cha);
if (ret != NULL){
debuglogname=strdup(ret->value);
if(debuglogname == NULL){
sysfatal("strdup failed for debuglogname in main: %r");
}
}
if(debug <=0){
debug=0;
}
if(debuglogname){
if((debuglogfile = fopen(debuglogname, "a+"))==0){
debug = 0;
}
}else{
debug = 0;
}
ret = config_get("debug_level",cha);
if (ret != NULL){
debug=atoi(ret->value);
}
ret = config_get("debug_logfile",cha);
if (ret != NULL){
debuglogname=strdup(ret->value);
if(debuglogname == NULL){
sysfatal("strdup failed for debuglogname in main: %r");
}
}
if(debug <=0){
debug=0;
}
if(debuglogname){
if((debuglogfile = fopen(debuglogname, "a+"))==0){
debug = 0;
}
}else{
debug = 0;
}
ret = config_get("sge_binpath",cha);
if (ret == NULL){
do_log(debuglogfile, debug, 1, "%s: key sge_binpath not found\n",argv0);
} else {
sge_binpath=strdup(ret->value);
if(sge_binpath == NULL){
sysfatal("strdup failed for sge_binpath in main: %r");
}
}
ret = config_get("sge_rootpath",cha);
if (ret == NULL){
do_log(debuglogfile, debug, 1, "%s: key sge_rootpath not found\n",argv0);
} else {
sge_rootpath=strdup(ret->value);
if(sge_rootpath == NULL){
sysfatal("strdup failed for sge_rootpath in main: %r");
}
tpath=make_message("%s",sge_rootpath);
if (opendir(tpath)==NULL){
do_log(debuglogfile, debug, 1, "%s: dir %s does not exist or is not readable\n",argv0,tpath);
sysfatal("dir %s does not exist or is not readable: %r",tpath);
}
free(tpath);
}
ret = config_get("sge_cellname",cha);
if (ret == NULL){
do_log(debuglogfile, debug, 1, "%s: key sge_cellname not found\n",argv0);
} else {
sge_cellname=strdup(ret->value);
if(sge_cellname == NULL){
sysfatal("strdup failed for sge_cellname in main: %r");
}
}
ret = config_get("sge_rootpath",cha);
if (ret == NULL){
if(debug){
fprintf(debuglogfile, "%s: key sge_rootpath not found\n",argv0);
fflush(debuglogfile);
}
} else {
sge_rootpath=strdup(ret->value);
if(sge_rootpath == NULL){
sysfatal("strdup failed for sge_rootpath in main: %r");
}
}
ret = config_get("sge_cellname",cha);
if (ret == NULL){
if(debug){
fprintf(debuglogfile, "%s: key sge_cellname not found\n",argv0);
fflush(debuglogfile);
}
} else {
sge_cellname=strdup(ret->value);
if(sge_cellname == NULL){
sysfatal("strdup failed for sge_cellname in main: %r");
}
}
ret = config_get("job_registry",cha);
if (ret == NULL){
do_log(debuglogfile, debug, 1, "%s: key job_registry not found\n",argv0);
sysfatal("job_registry not defined. Exiting");
} else {
reg_file=strdup(ret->value);
if(reg_file == NULL){
sysfatal("strdup failed for reg_file in main: %r");
}
}
ret = config_get("purge_interval",cha);
if (ret == NULL){
do_log(debuglogfile, debug, 1, "%s: key purge_interval not found using the default:%d\n",argv0,purge_interval);
} else {
purge_interval=atoi(ret->value);
}
ret = config_get("finalstate_query_interval",cha);
if (ret == NULL){
do_log(debuglogfile, debug, 1, "%s: key finalstate_query_interval not found using the default:%d\n",argv0,finalstate_query_interval);
} else {
finalstate_query_interval=atoi(ret->value);
}
ret = config_get("alldone_interval",cha);
if (ret == NULL){
do_log(debuglogfile, debug, 1, "%s: key alldone_interval not found using the default:%d\n",argv0,alldone_interval);
} else {
alldone_interval=atoi(ret->value);
}
ret = config_get("bupdater_loop_interval",cha);
if (ret == NULL){
do_log(debuglogfile, debug, 1, "%s: key bupdater_loop_interval not found using the default:%d\n",argv0,loop_interval);
} else {
loop_interval=atoi(ret->value);
}
ret = config_get("bupdater_pidfile",cha);
if (ret == NULL){
do_log(debuglogfile, debug, 1, "%s: key bupdater_pidfile not found\n",argv0);
} else {
pidfile=strdup(ret->value);
if(pidfile == NULL){
sysfatal("strdup failed for pidfile in main: %r");
}
}
ret = config_get("job_registry_use_mmap",cha);
if (ret == NULL){
do_log(debuglogfile, debug, 1, "%s: key job_registry_use_mmap not found. Default is NO\n",argv0);
} else {
do_log(debuglogfile, debug, 1, "%s: key job_registry_use_mmap is set to %s\n",argv0,ret->value);
}
if( !nodmn ) daemonize();
if( pidfile ){
writepid(pidfile);
free(pidfile);
}
config_free(cha);
rha=job_registry_init(reg_file, BY_BATCH_ID);
if (rha == NULL){
do_log(debuglogfile, debug, 1, "%s: Error initialising job registry %s\n",argv0,reg_file);
fprintf(stderr,"%s: Error initialising job registry %s :",argv0,reg_file);
perror("");
}
for(;;){
/* Purge old entries from registry */
now=time(0);
if(now - purge_time > 86400){
if(job_registry_purge(reg_file, now-purge_interval,0)<0){
do_log(debuglogfile, debug, 1, "%s: Error purging job registry %s\n",argv0,reg_file);
fprintf(stderr,"%s: Error purging job registry %s :",argv0,reg_file);
perror("");
}else{
purge_time=time(0);
}
}
//IntStateQuery();
fd = job_registry_open(rha, "r");
if (fd == NULL)
{
do_log(debuglogfile, debug, 1, "%s: Error opening job registry %s\n",argv0,reg_file);
fprintf(stderr,"%s: Error opening job registry %s :",argv0,reg_file);
perror("");
sleep(loop_interval);
}
if (job_registry_rdlock(rha, fd) < 0)
{
do_log(debuglogfile, debug, 1, "%s: Error read locking job registry %s\n",argv0,reg_file);
fprintf(stderr,"%s: Error read locking job registry %s :",argv0,reg_file);
perror("");
sleep(loop_interval);
}
job_registry_firstrec(rha,fd);
fseek(fd,0L,SEEK_SET);
if((query=calloc(STR_CHARS*2,1)) == 0){
sysfatal("can't malloc query %r");
}
if((queryStates=calloc(STR_CHARS*2,1)) == 0){
sysfatal("can't malloc query %r");
}
query[0]=' ';
queryStates[0]=' ';
while ((en = job_registry_get_next(rha, fd)) != NULL)
{
if(((now - en->mdate) > finalstate_query_interval) && en->status!=3 && en->status!=4)
{
/* create the constraint that will be used in condor_history command in FinalStateQuery*/
snprintf(constraint, sizeof(constraint), " %s",en->batch_id);
if (en->status==0) snprintf(constraint2, sizeof(constraint2), " u");
if (en->status==1) snprintf(constraint2, sizeof(constraint2), " q");
if (en->status==2) snprintf(constraint2, sizeof(constraint2), " r");
if (en->status==5) snprintf(constraint2, sizeof(constraint2), " h");
query=realloc(query,strlen(query)+strlen(constraint)+1);
queryStates=realloc(queryStates,strlen(queryStates)+strlen(constraint2)+1);
strcat(query,constraint);
strcat(queryStates,constraint2);
runfinal=TRUE;
}
/* Assign Status=4 and ExitStatus=-1 to all entries that after alldone_interval are still not in a final state(3 or 4) */
if((now - en->mdate > alldone_interval) && en->status!=3 && en->status!=4 && !runfinal)
{
time_t now;
now=time(0);
snprintf(string_now,sizeof(string_now),"%d",now);
AssignState(en->batch_id,"4" ,"-1","\0","\0",string_now);
}
free(en);
}
if(runfinal){
if((query_err=calloc((int)strlen(query),1)) == 0)
sysfatal("can't malloc query_err %r");
FinalStateQuery(query,queryStates,query_err);
free(query_err);
}
free(query);
free(queryStates);
fclose(fd);
if (runfinal){
runfinal=FALSE;
}
sleep (loop_interval);
} //for
job_registry_destroy(rha);
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
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;
}
