int main(int argc, char *argv[]){
FILE *fd;
job_registry_entry *en;
time_t now;
time_t purge_time=0;
time_t last_consistency_check=0;
char *pidfile=NULL;
char *first_duplicate=NULL;
struct pollfd *remupd_pollset = NULL;
int remupd_nfds;
int version=0;
int first=TRUE;
int tmptim;
time_t finalquery_start_date;
int loop_interval=DEFAULT_LOOP_INTERVAL;
int rc;
int c;
pthread_t RecUpdNetThd;
int confirm_time=0;
static int help;
static int short_help;
bact.njobs = 0;
bact.jobs = NULL;
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'},
{"prefix", required_argument, 0, 'p'},
{0, 0, 0, 0}
};
int option_index = 0;
c = getopt_long (argc, argv, "vop:",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 'p':
break;
case '?':
break;
default:
abort ();
}
}
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("slurm_binpath",cha);
if (ret == NULL){
do_log(debuglogfile, debug, 1, "%s: key slurm_binpath not found\n",argv0);
} else {
slurm_binpath=strdup(ret->value);
if(slurm_binpath == NULL){
sysfatal("strdup failed for slurm_binpath 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 {
registry_file=strdup(ret->value);
if(registry_file == NULL){
sysfatal("strdup failed for registry_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_consistency_check_interval",cha);
if (ret == NULL){
do_log(debuglogfile, debug, 1, "%s: key bupdater_consistency_check_interval not found using the default:%d\n",argv0,bupdater_consistency_check_interval);
} else {
bupdater_consistency_check_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("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("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);
}
remupd_conf = config_get("job_registry_add_remote",cha);
if (remupd_conf == NULL){
do_log(debuglogfile, debug, 1, "%s: key job_registry_add_remote not found\n",argv0);
}else{
if (job_registry_updater_setup_receiver(remupd_conf->values,remupd_conf->n_values,&remupd_head) < 0){
do_log(debuglogfile, debug, 1, "%s: Cannot set network receiver(s) up for remote update\n",argv0);
fprintf(stderr,"%s: Cannot set network receiver(s) up for remote update \n",argv0);
}
if (remupd_head == NULL){
do_log(debuglogfile, debug, 1, "%s: Cannot find values for network endpoints in configuration file (attribute 'job_registry_add_remote').\n",argv0);
fprintf(stderr,"%s: Cannot find values for network endpoints in configuration file (attribute 'job_registry_add_remote').\n", argv0);
}
if ((remupd_nfds = job_registry_updater_get_pollfd(remupd_head, &remupd_pollset)) < 0){
do_log(debuglogfile, debug, 1, "%s: Cannot setup poll set for receiving data.\n",argv0);
fprintf(stderr,"%s: Cannot setup poll set for receiving data.\n", argv0);
}
if (remupd_pollset == NULL || remupd_nfds == 0){
do_log(debuglogfile, debug, 1, "%s: No poll set available for receiving data.\n",argv0);
fprintf(stderr,"%s: No poll set available for receiving data.\n",argv0);
}
}
if( !nodmn ) daemonize();
if( pidfile ){
writepid(pidfile);
free(pidfile);
}
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("");
}
if (remupd_conf != NULL){
pthread_create(&RecUpdNetThd, NULL, (void *(*)(void *))ReceiveUpdateFromNetwork, (void *)NULL);
if (job_registry_updater_setup_sender(remupd_conf->values,remupd_conf->n_values,0,&remupd_head_send) < 0){
do_log(debuglogfile, debug, 1, "%s: Cannot set network sender(s) up for remote update\n",argv0);
fprintf(stderr,"%s: Cannot set network sender(s) up for remote update \n",argv0);
}
if (remupd_head_send == NULL){
do_log(debuglogfile, debug, 1, "%s: Cannot find values for network endpoints in configuration file (attribute 'job_registry_add_remote').\n",argv0);
fprintf(stderr,"%s: Cannot find values for network endpoints in configuration file (attribute 'job_registry_add_remote').\n", argv0);
}
}
config_free(cha);
for(;;){
/* Purge old entries from registry */
now=time(0);
if(now - purge_time > 86400){
if((rc=job_registry_purge(registry_file, now-purge_interval,0))<0){
do_log(debuglogfile, debug, 1, "%s: Error purging job registry %s:%d\n",argv0,registry_file,rc);
fprintf(stderr,"%s: Error purging job registry %s :",argv0,registry_file);
perror("");
}
purge_time=time(0);
}
now=time(0);
if(now - last_consistency_check > bupdater_consistency_check_interval){
if(job_registry_check_index_key_uniqueness(rha,&first_duplicate)==JOB_REGISTRY_FAIL){
do_log(debuglogfile, debug, 1, "%s: Found job registry duplicate entry. The first one is:%s.\nJobid should be removed or registry directory should be removed.\n",argv0,first_duplicate);
fprintf(stderr,"%s: Found job registry duplicate entry. The first one is:%s.\nJobid should be removed or registry directory should be removed.",argv0,first_duplicate);
}
last_consistency_check=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,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 job registry %s\n",argv0,registry_file);
fprintf(stderr,"%s: Error read locking job registry %s :",argv0,registry_file);
perror("");
sleep(loop_interval);
continue;
}
job_registry_firstrec(rha,fd);
fseek(fd,0L,SEEK_SET);
first=TRUE;
finalquery_start_date = time(0);
while ((en = job_registry_get_next(rha, fd)) != NULL){
if((bupdater_lookup_active_jobs(&bact,en->batch_id) != BUPDATER_ACTIVE_JOBS_SUCCESS) && en->status!=REMOVED && en->status!=COMPLETED){
do_log(debuglogfile, debug, 3, "%s: bupdater_lookup_active_jobs returned: %d for jobid: %s\n",argv0,bupdater_lookup_active_jobs(&bact,en->batch_id),en->batch_id);
confirm_time=atoi(en->updater_info);
if(confirm_time==0){
confirm_time=en->mdate;
}
/* Assign Status=4 and ExitStatus=999 to all entries that after alldone_interval are still not in a final state(3 or 4)*/
if(now-confirm_time>alldone_interval){
AssignFinalState(en->batch_id);
free(en);
continue;
}
if(en->status==IDLE && strlen(en->updater_info)>0){
if (en->mdate < finalquery_start_date){
finalquery_start_date=en->mdate;
}
do_log(debuglogfile, debug, 2, "%s: FinalStateQuery needed for jobid=%s with status=%d\n",argv0,en->batch_id,en->status);
runfinal=TRUE;
}else if((now-confirm_time>finalstate_query_interval) && (now > next_finalstatequery)){
if (en->mdate < finalquery_start_date){
finalquery_start_date=en->mdate;
}
do_log(debuglogfile, debug, 2, "%s: FinalStateQuery needed for jobid=%s with status=%d\n",argv0,en->batch_id,en->status);
runfinal=TRUE;
}
}
free(en);
}
if(runfinal_oldlogs){
FinalStateQuery(0,1);
runfinal_oldlogs=FALSE;
runfinal=FALSE;
}else if(runfinal){
FinalStateQuery(finalquery_start_date,1);
runfinal=FALSE;
}
fclose(fd);
sleep(loop_interval);
}
job_registry_destroy(rha);
return 0;
}
int main(int argc, char *argv[]){
FILE *fd;
job_registry_entry *en;
time_t now;
time_t purge_time=0;
time_t last_consistency_check=0;
char *pidfile=NULL;
char *final_string=NULL;
char *cp=NULL;
char *tpath;
char *tspooldir;
char *first_duplicate=NULL;
struct pollfd *remupd_pollset = NULL;
int remupd_nfds;
int version=0;
int first=TRUE;
int tmptim;
int finstr_len=0;
int loop_interval=DEFAULT_LOOP_INTERVAL;
int fsq_ret=0;
int c;
pthread_t RecUpdNetThd;
int confirm_time=0;
static int help;
static int short_help;
bact.njobs = 0;
bact.jobs = NULL;
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'},
{"prefix", required_argument, 0, 'p'},
{0, 0, 0, 0}
};
int option_index = 0;
c = getopt_long (argc, argv, "vop:",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 'p':
break;
case '?':
break;
default:
abort ();
}
}
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;
}
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("pbs_binpath",cha);
if (ret == NULL){
do_log(debuglogfile, debug, 1, "%s: key pbs_binpath not found\n",argv0);
} else {
pbs_binpath=strdup(ret->value);
if(pbs_binpath == NULL){
sysfatal("strdup failed for pbs_binpath in main: %r");
}
}
ret = config_get("pbs_spoolpath",cha);
if (ret == NULL){
do_log(debuglogfile, debug, 1, "%s: key pbs_spoolpath not found\n",argv0);
} else {
pbs_spoolpath=strdup(ret->value);
if(pbs_spoolpath == NULL){
sysfatal("strdup failed for pbs_spoolpath in main: %r");
}
tpath=make_message("%s/server_logs",pbs_spoolpath);
if (opendir(tpath)==NULL){
do_log(debuglogfile, debug, 1, "%s: dir %s does not exist or is not readable. Trying now pbs commands\n",argv0,tpath);
tspooldir=GetPBSSpoolPath(pbs_binpath);
free(tpath);
tpath=make_message("%s/server_logs",tspooldir);
free(tspooldir);
if (opendir(tpath)==NULL){
do_log(debuglogfile, debug, 1, "dir %s does not exist or is not readable (using pbs commands)",tpath);
sysfatal("dir %s does not exist or is not readable (using pbs commands): %r",tpath);
}
}
free(tpath);
}
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 {
registry_file=strdup(ret->value);
if(registry_file == NULL){
sysfatal("strdup failed for registry_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("tracejob_logs_to_read",cha);
if (ret == NULL){
do_log(debuglogfile, debug, 1, "%s: key tracejob_logs_to_read not found using the default:%d\n",argv0,tracejob_logs_to_read);
} else {
tracejob_logs_to_read=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_consistency_check_interval",cha);
if (ret == NULL){
do_log(debuglogfile, debug, 1, "%s: key bupdater_consistency_check_interval not found using the default:%d\n",argv0,bupdater_consistency_check_interval);
} else {
bupdater_consistency_check_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("pbs_batch_caching_enabled",cha);
if (ret == NULL){
do_log(debuglogfile, debug, 1, "%s: key pbs_batch_caching_enabled not found using default\n",argv0,pbs_batch_caching_enabled);
} else {
pbs_batch_caching_enabled=strdup(ret->value);
if(pbs_batch_caching_enabled == NULL){
sysfatal("strdup failed for pbs_batch_caching_enabled in main: %r");
}
}
ret = config_get("batch_command_caching_filter",cha);
if (ret == NULL){
do_log(debuglogfile, debug, 1, "%s: key batch_command_caching_filter not found using default\n",argv0,batch_command_caching_filter);
} else {
batch_command_caching_filter=strdup(ret->value);
if(batch_command_caching_filter == NULL){
sysfatal("strdup failed for batch_command_caching_filter in main: %r");
}
}
batch_command=(strcmp(pbs_batch_caching_enabled,"yes")==0?make_message("%s ",batch_command_caching_filter):make_message(""));
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);
}
ret = config_get("tracejob_max_output",cha);
if (ret == NULL){
do_log(debuglogfile, debug, 1, "%s: key tracejob_max_output not found using default\n",argv0,tracejob_max_output);
} else {
tracejob_max_output==atoi(ret->value);
}
remupd_conf = config_get("job_registry_add_remote",cha);
if (remupd_conf == NULL){
do_log(debuglogfile, debug, 1, "%s: key job_registry_add_remote not found\n",argv0);
}else{
if (job_registry_updater_setup_receiver(remupd_conf->values,remupd_conf->n_values,&remupd_head) < 0){
do_log(debuglogfile, debug, 1, "%s: Cannot set network receiver(s) up for remote update\n",argv0);
fprintf(stderr,"%s: Cannot set network receiver(s) up for remote update \n",argv0);
}
if (remupd_head == NULL){
do_log(debuglogfile, debug, 1, "%s: Cannot find values for network endpoints in configuration file (attribute 'job_registry_add_remote').\n",argv0);
fprintf(stderr,"%s: Cannot find values for network endpoints in configuration file (attribute 'job_registry_add_remote').\n", argv0);
}
if ((remupd_nfds = job_registry_updater_get_pollfd(remupd_head, &remupd_pollset)) < 0){
do_log(debuglogfile, debug, 1, "%s: Cannot setup poll set for receiving data.\n",argv0);
fprintf(stderr,"%s: Cannot setup poll set for receiving data.\n", argv0);
}
if (remupd_pollset == NULL || remupd_nfds == 0){
do_log(debuglogfile, debug, 1, "%s: No poll set available for receiving data.\n",argv0);
fprintf(stderr,"%s: No poll set available for receiving data.\n",argv0);
}
}
if( !nodmn ) daemonize();
if( pidfile ){
writepid(pidfile);
free(pidfile);
}
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("");
}
if (remupd_conf != NULL){
pthread_create(&RecUpdNetThd, NULL, (void *(*)(void *))ReceiveUpdateFromNetwork, (void *)NULL);
if (job_registry_updater_setup_sender(remupd_conf->values,remupd_conf->n_values,0,&remupd_head_send) < 0){
do_log(debuglogfile, debug, 1, "%s: Cannot set network sender(s) up for remote update\n",argv0);
fprintf(stderr,"%s: Cannot set network sender(s) up for remote update \n",argv0);
}
if (remupd_head_send == NULL){
do_log(debuglogfile, debug, 1, "%s: Cannot find values for network endpoints in configuration file (attribute 'job_registry_add_remote').\n",argv0);
fprintf(stderr,"%s: Cannot find values for network endpoints in configuration file (attribute 'job_registry_add_remote').\n", argv0);
}
}
config_free(cha);
for(;;){
/* Purge old entries from registry */
now=time(0);
if(now - purge_time > 86400){
if(job_registry_purge(registry_file, now-purge_interval,0)<0){
do_log(debuglogfile, debug, 1, "%s: Error purging job registry %s\n",argv0,registry_file);
fprintf(stderr,"%s: Error purging job registry %s :",argv0,registry_file);
perror("");
}else{
purge_time=time(0);
}
}
now=time(0);
if(now - last_consistency_check > bupdater_consistency_check_interval){
if(job_registry_check_index_key_uniqueness(rha,&first_duplicate)==JOB_REGISTRY_FAIL){
do_log(debuglogfile, debug, 1, "%s: Found job registry duplicate entry. The first one is:%s\n",argv0,first_duplicate);
fprintf(stderr,"%s: Found job registry duplicate entry. The first one is:%s",argv0,first_duplicate);
}else{
last_consistency_check=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,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 job registry %s\n",argv0,registry_file);
fprintf(stderr,"%s: Error read locking job registry %s :",argv0,registry_file);
perror("");
sleep(loop_interval);
continue;
}
job_registry_firstrec(rha,fd);
fseek(fd,0L,SEEK_SET);
first=TRUE;
while ((en = job_registry_get_next(rha, fd)) != NULL){
if((bupdater_lookup_active_jobs(&bact, en->batch_id) != BUPDATER_ACTIVE_JOBS_SUCCESS) && en->status!=REMOVED && en->status!=COMPLETED){
confirm_time=atoi(en->updater_info);
if(confirm_time==0){
confirm_time=en->mdate;
}
/* Assign Status=4 and ExitStatus=999 to all entries that after alldone_interval are still not in a final state(3 or 4)*/
if(now-confirm_time>alldone_interval){
AssignFinalState(en->batch_id);
free(en);
continue;
}
if((now-confirm_time>finalstate_query_interval) && (now > next_finalstatequery)){
if((final_string=realloc(final_string,finstr_len + strlen(en->batch_id) + 2)) == 0){
sysfatal("can't malloc final_string: %r");
} else {
if (finstr_len == 0) final_string[0] = '\000';
}
strcat(final_string,en->batch_id);
strcat(final_string,":");
finstr_len=strlen(final_string);
runfinal=TRUE;
}
}
free(en);
}
if(runfinal){
if (final_string[finstr_len-1] == ':' && (cp = strrchr (final_string, ':')) != NULL){
*cp = '\0';
}
if(fsq_ret != 0){
fsq_ret=FinalStateQuery(final_string,tracejob_logs_to_read);
}else{
fsq_ret=FinalStateQuery(final_string,1);
}
runfinal=FALSE;
}
if (final_string != NULL){
free(final_string);
final_string = NULL;
finstr_len = 0;
}
fclose(fd);
sleep(loop_interval);
}
job_registry_destroy(rha);
return 0;
}
int
main(int argc, char *argv[])
{
char *registry_file = NULL, *registry_file_env = NULL;
int need_to_free_registry_file = FALSE;
const char *default_registry_file = "blah_job_registry.bjr";
char *my_home;
job_registry_entry en;
job_status_t status=IDLE;
int exitcode = -1;
char *exitreason = "";
char *user_prefix = "";
char *user_proxy = "";
char *proxy_subject = "";
int renew_proxy = 0;
char *wn_addr = "";
time_t udate=0;
char *blah_id, *batch_id;
int ent, ret, prret, rhret;
config_handle *cha;
config_entry *rge, *remupd_conf;
job_registry_handle *rha, *rhano;
job_registry_index_mode rgin_mode = NO_INDEX;
job_registry_updater_endpoint *remupd_head = NULL;
if (argc > 1 && (strcmp(argv[1], "-u") == 0))
{
/* Obtain an index to the registry so that we can */
/* check that the record is unique. */
rgin_mode = BY_BLAH_ID;
argv[1] = argv[0];
argc--;
argv++;
}
if (argc < 3)
{
fprintf(stderr,"Usage: %s [-u] <BLAH id> <batch id> [job status] [udate] [user prefix] [user proxy] [renew proxy] [proxy subject] [worker node] [exit code] [exit reason]\n",argv[0]);
return 1;
}
blah_id = argv[1];
batch_id = argv[2];
if (argc > 3) status = atoi(argv[3]);
if (argc > 4) udate = atol(argv[4]);
if (argc > 5) user_prefix = argv[5];
if (argc > 6) user_proxy = argv[6];
if (argc > 7) renew_proxy = atoi(argv[7]);
if (argc > 8) proxy_subject = argv[8];
if (argc > 9) wn_addr = argv[9];
if (argc > 10) exitcode = atoi(argv[10]);
if (argc > 11) exitreason = argv[11];
cha = config_read(NULL); /* Read config from default locations. */
if (cha != NULL)
{
rge = config_get("job_registry", cha);
remupd_conf = config_get("job_registry_add_remote", cha);
if (remupd_conf != NULL)
{
if (job_registry_updater_setup_sender(remupd_conf->values,
remupd_conf->n_values, 2,
&remupd_head) < 0)
{
fprintf(stderr,"%s: warning: cannot set network sender(s) up for remote update to:\n",argv[0]);
for (ent = 0; ent < remupd_conf->n_values; ent++)
{
fprintf(stderr," - %s\n", remupd_conf->values[ent]);
}
}
}
if (rge != NULL) registry_file = rge->value;
}
/* Env variable takes precedence */
registry_file_env = getenv("BLAH_JOB_REGISTRY_FILE");
if (registry_file_env != NULL) registry_file = registry_file_env;
if (registry_file == NULL)
{
my_home = getenv("HOME");
if (my_home == NULL) my_home = ".";
registry_file = (char *)malloc(strlen(default_registry_file)+strlen(my_home)+2);
if (registry_file != NULL)
{
sprintf(registry_file,"%s/%s",my_home,default_registry_file);
need_to_free_registry_file = TRUE;
}
else
{
if (cha != NULL) config_free(cha);
if (remupd_head != NULL) job_registry_updater_free_endpoints(remupd_head);
return 1;
}
}
JOB_REGISTRY_ASSIGN_ENTRY(en.blah_id,blah_id);
JOB_REGISTRY_ASSIGN_ENTRY(en.batch_id,batch_id);
en.status = status;
en.exitcode = exitcode;
JOB_REGISTRY_ASSIGN_ENTRY(en.wn_addr,wn_addr);
en.udate = udate;
JOB_REGISTRY_ASSIGN_ENTRY(en.exitreason,exitreason);
en.submitter = geteuid();
JOB_REGISTRY_ASSIGN_ENTRY(en.user_prefix,user_prefix);
en.proxy_link[0] = '\000'; /* Start with a valid string */
en.updater_info[0] = '\000';
en.renew_proxy = 0; /* Enable renewal only if a proxy is found */
rha=job_registry_init(registry_file, rgin_mode);
if (rha == NULL)
{
if (errno == EACCES)
{
/* Try nonpriv update. It may work. */
rhano = job_registry_init(registry_file, NAMES_ONLY);
if (cha != NULL) config_free(cha);
if (need_to_free_registry_file) free(registry_file);
if (rhano != NULL)
{
if (strlen(user_proxy) > 0)
{
prret = job_registry_set_proxy(rhano, &en, user_proxy);
if (prret < 0)
{
fprintf(stderr,"%s: warning: setting proxy to %s: ",argv[0],user_proxy);
perror("");
}
else en.renew_proxy = renew_proxy;
}
ret=job_registry_append_nonpriv(rhano, &en);
job_registry_destroy(rhano);
if (ret < 0)
{
fprintf(stderr,"%s: job_registry_append_nonpriv returns %d: ",argv[0],ret);
perror("");
if (remupd_head != NULL) job_registry_updater_free_endpoints(remupd_head);
return 4;
}
else goto happy_ending;
}
}
else
{
fprintf(stderr,"%s: error initialising job registry: ",argv[0]);
perror("");
}
if (remupd_head != NULL) job_registry_updater_free_endpoints(remupd_head);
return 2;
}
/* Filename is stored in job registry handle. - Don't need these anymore */
if (cha != NULL) config_free(cha);
if (need_to_free_registry_file) free(registry_file);
if (strlen(user_proxy) > 0)
{
prret = job_registry_set_proxy(rha, &en, user_proxy);
if (prret < 0)
{
fprintf(stderr,"%s: warning: setting proxy to %s: ",argv[0],user_proxy);
perror("");
}
else en.renew_proxy = renew_proxy;
}
en.subject_hash[0] = '\000';
if (strlen(proxy_subject) > 0)
{
job_registry_compute_subject_hash(&en, proxy_subject);
rhret = job_registry_record_subject_hash(rha, en.subject_hash,
proxy_subject, TRUE);
if (rhret < 0)
{
fprintf(stderr,"%s: warning: recording proxy subject %s (hash %s): ", argv[0], proxy_subject, en.subject_hash);
perror("");
}
}
if ((ret=job_registry_append(rha, &en)) < 0)
{
if (errno == EACCES)
{
/* Try nonpriv update. It may work. */
ret=job_registry_append_nonpriv(rha, &en);
job_registry_destroy(rha);
if (ret < 0)
{
fprintf(stderr,"%s: job_registry_append_nonpriv returns %d: ",argv[0],ret);
perror("");
if (remupd_head != NULL) job_registry_updater_free_endpoints(remupd_head);
return 5;
}
else goto happy_ending;
}
fprintf(stderr,"%s: job_registry_append returns %d: ",argv[0],ret);
perror("");
job_registry_destroy(rha);
if (remupd_head != NULL) job_registry_updater_free_endpoints(remupd_head);
return 3;
}
job_registry_destroy(rha);
happy_ending:
if (remupd_head != NULL)
{
if (job_registry_send_update(remupd_head, &en,
(strlen(proxy_subject) > 0 ? proxy_subject : NULL),
(strlen(user_proxy) > 0 ? user_proxy : NULL)) < 0)
{
fprintf(stderr,"%s: warning: sending network update: ",argv[0]);
perror("");
}
job_registry_updater_free_endpoints(remupd_head);
}
return 0;
}
