diff --git a/src/common/slurm_jobacct.c b/src/common/slurm_jobacct.c
index 87d1cd5f789043485fc47497de64de175e12c5ca..055ea714e17e6bfea25bbfcfa44f6bf07de5a3a4 100644
--- a/src/common/slurm_jobacct.c
+++ b/src/common/slurm_jobacct.c
@@ -86,31 +86,6 @@ typedef struct slurm_jobacct_ops {
void (*jobacct_suspendpoll) ();
} slurm_jobacct_ops_t;
-/*
- * These strings must be in the same order as the fields declared
- * for slurm_jobacct_ops_t.
- */
-static const char *syms[] = {
- "jobacct_p_init_struct",
- "jobacct_p_alloc",
- "jobacct_p_free",
- "jobacct_p_setinfo",
- "jobacct_p_getinfo",
- "jobacct_p_aggregate",
- "jobacct_p_pack",
- "jobacct_p_unpack",
- "jobacct_p_init_slurmctld",
- "jobacct_p_fini_slurmctld",
- "jobacct_p_job_start_slurmctld",
- "jobacct_p_job_complete_slurmctld",
- "jobacct_p_step_start_slurmctld",
- "jobacct_p_step_complete_slurmctld",
- "jobacct_p_suspend_slurmctld",
- "jobacct_p_startpoll",
- "jobacct_p_endpoll",
- "jobacct_p_suspendpoll",
-};
-
/*
* A global job accounting context. "Global" in the sense that there's
* only one, with static bindings. We don't export it.
@@ -184,7 +159,31 @@ _slurm_jobacct_context_destroy( slurm_jobacct_context_t c )
static slurm_jobacct_ops_t *
_slurm_jobacct_get_ops( slurm_jobacct_context_t c )
{
- int n_syms = sizeof( syms ) / sizeof( char * );
+ /*
+ * These strings must be in the same order as the fields declared
+ * for slurm_jobacct_ops_t.
+ */
+ static const char *syms[] = {
+ "jobacct_p_init_struct",
+ "jobacct_p_alloc",
+ "jobacct_p_free",
+ "jobacct_p_setinfo",
+ "jobacct_p_getinfo",
+ "jobacct_p_aggregate",
+ "jobacct_p_pack",
+ "jobacct_p_unpack",
+ "jobacct_p_init_slurmctld",
+ "jobacct_p_fini_slurmctld",
+ "jobacct_p_job_start_slurmctld",
+ "jobacct_p_job_complete_slurmctld",
+ "jobacct_p_step_start_slurmctld",
+ "jobacct_p_step_complete_slurmctld",
+ "jobacct_p_suspend_slurmctld",
+ "jobacct_p_startpoll",
+ "jobacct_p_endpoll",
+ "jobacct_p_suspendpoll"
+ };
+ int n_syms = sizeof( syms ) / sizeof( char * );
int rc = 0;
/* Get the plugin list, if needed. */
if ( c->plugin_list == NULL ) {
@@ -303,6 +302,10 @@ extern jobacctinfo_t *jobacct_g_alloc()
extern int jobacct_g_free(jobacctinfo_t *jobacct)
{
int retval = SLURM_SUCCESS;
+
+ if (_slurm_jobacct_init() < 0)
+ return SLURM_ERROR;
+
slurm_mutex_lock( &g_jobacct_context_lock );
if ( g_jobacct_context )
retval = (*(g_jobacct_context->ops.jobacct_free))(jobacct);
@@ -316,6 +319,9 @@ extern int jobacct_g_setinfo(jobacctinfo_t *jobacct,
{
int retval = SLURM_SUCCESS;
+ if (_slurm_jobacct_init() < 0)
+ return SLURM_ERROR;
+
slurm_mutex_lock( &g_jobacct_context_lock );
if ( g_jobacct_context )
retval = (*(g_jobacct_context->ops.jobacct_setinfo))
@@ -329,6 +335,9 @@ extern int jobacct_g_getinfo(jobacctinfo_t *jobacct,
{
int retval = SLURM_SUCCESS;
+ if (_slurm_jobacct_init() < 0)
+ return SLURM_ERROR;
+
slurm_mutex_lock( &g_jobacct_context_lock );
if ( g_jobacct_context )
retval = (*(g_jobacct_context->ops.jobacct_getinfo))
@@ -339,6 +348,9 @@ extern int jobacct_g_getinfo(jobacctinfo_t *jobacct,
extern void jobacct_g_aggregate(jobacctinfo_t *dest, jobacctinfo_t *from)
{
+ if (_slurm_jobacct_init() < 0)
+ return;
+
slurm_mutex_lock( &g_jobacct_context_lock );
if ( g_jobacct_context )
(*(g_jobacct_context->ops.jobacct_aggregate))(dest, from);
@@ -348,6 +360,9 @@ extern void jobacct_g_aggregate(jobacctinfo_t *dest, jobacctinfo_t *from)
extern void jobacct_g_pack(jobacctinfo_t *jobacct, Buf buffer)
{
+ if (_slurm_jobacct_init() < 0)
+ return;
+
slurm_mutex_lock( &g_jobacct_context_lock );
if ( g_jobacct_context )
(*(g_jobacct_context->ops.jobacct_pack))(jobacct, buffer);
@@ -359,6 +374,9 @@ extern int jobacct_g_unpack(jobacctinfo_t **jobacct, Buf buffer)
{
int retval = SLURM_SUCCESS;
+ if (_slurm_jobacct_init() < 0)
+ return SLURM_ERROR;
+
slurm_mutex_lock( &g_jobacct_context_lock );
if ( g_jobacct_context )
retval = (*(g_jobacct_context->ops.jobacct_unpack))
@@ -384,6 +402,9 @@ extern int jobacct_g_init_slurmctld(char *job_acct_log)
extern int jobacct_g_fini_slurmctld()
{
int retval = SLURM_SUCCESS;
+ if (_slurm_jobacct_init() < 0)
+ return SLURM_ERROR;
+
slurm_mutex_lock( &g_jobacct_context_lock );
if ( g_jobacct_context )
retval = (*(g_jobacct_context->ops.jobacct_fini))();
@@ -397,7 +418,9 @@ extern int jobacct_g_fini_slurmctld()
extern int jobacct_g_job_start_slurmctld(struct job_record *job_ptr)
{
int retval = SLURM_SUCCESS;
-
+ if (_slurm_jobacct_init() < 0)
+ return SLURM_ERROR;
+
slurm_mutex_lock( &g_jobacct_context_lock );
if ( g_jobacct_context )
retval = (*(g_jobacct_context->ops.jobacct_job_start))
@@ -409,7 +432,9 @@ extern int jobacct_g_job_start_slurmctld(struct job_record *job_ptr)
extern int jobacct_g_job_complete_slurmctld(struct job_record *job_ptr)
{
int retval = SLURM_SUCCESS;
-
+ if (_slurm_jobacct_init() < 0)
+ return SLURM_ERROR;
+
slurm_mutex_lock( &g_jobacct_context_lock );
if ( g_jobacct_context )
retval = (*(g_jobacct_context->ops.jobacct_job_complete))
@@ -421,7 +446,9 @@ extern int jobacct_g_job_complete_slurmctld(struct job_record *job_ptr)
extern int jobacct_g_step_start_slurmctld(struct step_record *step_ptr)
{
int retval = SLURM_SUCCESS;
-
+ if (_slurm_jobacct_init() < 0)
+ return SLURM_ERROR;
+
slurm_mutex_lock( &g_jobacct_context_lock );
if ( g_jobacct_context )
retval = (*(g_jobacct_context->ops.jobacct_step_start))
@@ -433,7 +460,9 @@ extern int jobacct_g_step_start_slurmctld(struct step_record *step_ptr)
extern int jobacct_g_step_complete_slurmctld(struct step_record *step_ptr)
{
int retval = SLURM_SUCCESS;
-
+ if (_slurm_jobacct_init() < 0)
+ return SLURM_ERROR;
+
slurm_mutex_lock( &g_jobacct_context_lock );
if ( g_jobacct_context )
retval = (*(g_jobacct_context->ops.jobacct_step_complete))
@@ -445,7 +474,9 @@ extern int jobacct_g_step_complete_slurmctld(struct step_record *step_ptr)
extern int jobacct_g_suspend_slurmctld(struct job_record *job_ptr)
{
int retval = SLURM_SUCCESS;
-
+ if (_slurm_jobacct_init() < 0)
+ return SLURM_ERROR;
+
slurm_mutex_lock( &g_jobacct_context_lock );
if ( g_jobacct_context )
retval = (*(g_jobacct_context->ops.jobacct_suspend))
@@ -457,6 +488,8 @@ extern int jobacct_g_suspend_slurmctld(struct job_record *job_ptr)
extern int jobacct_g_startpoll(int frequency)
{
int retval = SLURM_SUCCESS;
+ if (_slurm_jobacct_init() < 0)
+ return SLURM_ERROR;
if (_slurm_jobacct_init() < 0)
return SLURM_ERROR;
@@ -473,7 +506,9 @@ extern int jobacct_g_startpoll(int frequency)
extern int jobacct_g_endpoll(slurmd_job_t *job)
{
int retval = SLURM_SUCCESS;
-
+ if (_slurm_jobacct_init() < 0)
+ return SLURM_ERROR;
+
slurm_mutex_lock( &g_jobacct_context_lock );
if ( g_jobacct_context )
retval = (*(g_jobacct_context->ops.jobacct_endpoll))(job);
@@ -483,6 +518,9 @@ extern int jobacct_g_endpoll(slurmd_job_t *job)
extern void jobacct_g_suspendpoll()
{
+ if (_slurm_jobacct_init() < 0)
+ return;
+
slurm_mutex_lock( &g_jobacct_context_lock );
if ( g_jobacct_context )
(*(g_jobacct_context->ops.jobacct_suspendpoll))();
diff --git a/src/common/slurm_protocol_api.c b/src/common/slurm_protocol_api.c
index 98ed476a67cf1812c69726db6d0b5821d24402e3..6cf3bc0a255925007bcb8afbee33001507be6e09 100644
--- a/src/common/slurm_protocol_api.c
+++ b/src/common/slurm_protocol_api.c
@@ -765,7 +765,7 @@ List slurm_receive_msg(slurm_fd fd, slurm_msg_t *msg, int timeout)
msg->forward_struct->timeout = timeout-header.forward.timeout;
msg->forward_struct->fwd_cnt = header.forward.cnt;
- debug("forwarding messages to %d nodes!!!!",
+ debug3("forwarding messages to %d nodes!!!!",
msg->forward_struct->fwd_cnt);
if(forward_msg(msg->forward_struct, &header) == SLURM_ERROR) {
@@ -978,7 +978,7 @@ int slurm_send_node_msg(slurm_fd fd, slurm_msg_t * msg)
/* wait for all the other messages on the tree under us */
if(msg->forward_struct_init == FORWARD_INIT && msg->forward_struct) {
- debug("looking for %d", msg->forward_struct->fwd_cnt);
+ debug3("looking for %d", msg->forward_struct->fwd_cnt);
slurm_mutex_lock(&msg->forward_struct->forward_mutex);
count = 0;
itr = list_iterator_create(msg->ret_list);
@@ -987,7 +987,7 @@ int slurm_send_node_msg(slurm_fd fd, slurm_msg_t * msg)
count += list_count(ret_type->ret_data_list);
}
list_iterator_destroy(itr);
- debug("Got back %d", count);
+ debug3("Got back %d", count);
while((count < msg->forward_struct->fwd_cnt)) {
pthread_cond_wait(&msg->forward_struct->notify,
&msg->forward_struct->forward_mutex);
@@ -998,10 +998,10 @@ int slurm_send_node_msg(slurm_fd fd, slurm_msg_t * msg)
count += list_count(ret_type->ret_data_list);
}
list_iterator_destroy(itr);
- debug("Got back %d", count);
+ debug3("Got back %d", count);
}
- debug("Got them all");
+ debug3("Got them all");
slurm_mutex_unlock(&msg->forward_struct->forward_mutex);
destroy_forward_struct(msg->forward_struct);
}
diff --git a/src/common/slurm_protocol_defs.h b/src/common/slurm_protocol_defs.h
index a8523d1fbc37f6a6936b396469e2ed51bba995b6..488c50674c5bdc0a27dded6ff8cabf2d3b603e3b 100644
--- a/src/common/slurm_protocol_defs.h
+++ b/src/common/slurm_protocol_defs.h
@@ -683,4 +683,53 @@ extern char *job_state_string_compact(enum job_states inx);
extern char *node_state_string(enum node_states inx);
extern char *node_state_string_compact(enum node_states inx);
+#define safe_read(fd, buf, size) do { \
+ int remaining = size; \
+ void *ptr = buf; \
+ int rc; \
+ while (remaining > 0) { \
+ rc = read(fd, ptr, remaining); \
+ if (rc == 0) { \
+ debug("%s:%d: %s: safe_read (%d of %d) EOF", \
+ __FILE__, __LINE__, __CURRENT_FUNC__, \
+ remaining, (int)size); \
+ goto rwfail; \
+ } else if (rc < 0) { \
+ debug("%s:%d: %s: safe_read (%d of %d) failed: %m", \
+ __FILE__, __LINE__, __CURRENT_FUNC__, \
+ remaining, (int)size); \
+ goto rwfail; \
+ } else { \
+ ptr += rc; \
+ remaining -= rc; \
+ if (remaining > 0) \
+ debug3("%s:%d: %s: safe_read (%d of %d) partial read", \
+ __FILE__, __LINE__, __CURRENT_FUNC__, \
+ remaining, (int)size); \
+ } \
+ } \
+ } while (0)
+
+#define safe_write(fd, buf, size) do { \
+ int remaining = size; \
+ void *ptr = buf; \
+ int rc; \
+ while(remaining > 0) { \
+ rc = write(fd, ptr, remaining); \
+ if (rc < 0) { \
+ debug("%s:%d: %s: safe_write (%d of %d) failed: %m", \
+ __FILE__, __LINE__, __CURRENT_FUNC__, \
+ remaining, (int)size); \
+ goto rwfail; \
+ } else { \
+ ptr += rc; \
+ remaining -= rc; \
+ if (remaining > 0) \
+ debug3("%s:%d: %s: safe_write (%d of %d) partial write", \
+ __FILE__, __LINE__, __CURRENT_FUNC__, \
+ remaining, (int)size); \
+ } \
+ } \
+ } while (0)
+
#endif
diff --git a/src/plugins/jobacct/common/jobacct_common.c b/src/plugins/jobacct/common/jobacct_common.c
index de1ed72311dd2dc50f33f4ec86a3ec60976172f6..462868bed5521c5a67b89614e471eb2439664db3 100644
--- a/src/plugins/jobacct/common/jobacct_common.c
+++ b/src/plugins/jobacct/common/jobacct_common.c
@@ -71,7 +71,7 @@ extern int common_setinfo(struct jobacctinfo *jobacct,
enum jobacct_data_type type, void *data)
{
int rc = SLURM_SUCCESS;
- int *temp = (int *)data;
+ int *fd = (int *)data;
uint32_t *uint32 = (uint32_t *) data;
struct rusage *rusage = (struct rusage *) data;
struct jobacctinfo *send = (struct jobacctinfo *) data;
@@ -81,7 +81,7 @@ extern int common_setinfo(struct jobacctinfo *jobacct,
memcpy(jobacct, send, sizeof(struct jobacctinfo));
break;
case JOBACCT_DATA_PIPE:
- safe_write((int)*temp, jobacct, sizeof(struct jobacctinfo));
+ safe_write(*fd, jobacct, sizeof(struct jobacctinfo));
break;
case JOBACCT_DATA_RUSAGE:
memcpy(&jobacct->rusage, rusage, sizeof(struct rusage));
@@ -107,7 +107,7 @@ extern int common_getinfo(struct jobacctinfo *jobacct,
enum jobacct_data_type type, void *data)
{
int rc = SLURM_SUCCESS;
- int *temp = (int *)data;
+ int *fd = (int *)data;
uint32_t *uint32 = (uint32_t *) data;
struct rusage *rusage = (struct rusage *) data;
struct jobacctinfo *send = (struct jobacctinfo *) data;
@@ -117,7 +117,7 @@ extern int common_getinfo(struct jobacctinfo *jobacct,
memcpy(send, jobacct, sizeof(struct jobacctinfo));
break;
case JOBACCT_DATA_PIPE:
- safe_read((int)*temp, jobacct, sizeof(struct jobacctinfo));
+ safe_read(*fd, jobacct, sizeof(struct jobacctinfo));
break;
case JOBACCT_DATA_RUSAGE:
memcpy(rusage, &jobacct->rusage, sizeof(struct rusage));
diff --git a/src/slurmctld/controller.c b/src/slurmctld/controller.c
index f3eca51a5cd3e3236b4f024622ff40aae044d3bb..4c40197d85e01769b9db8c1f92383d691fef67f6 100644
--- a/src/slurmctld/controller.c
+++ b/src/slurmctld/controller.c
@@ -346,6 +346,7 @@ int main(int argc, char *argv[])
/* Plugins are needed to purge job/node data structures,
* unplug after other data structures are purged */
g_slurm_jobcomp_fini();
+ jobacct_g_fini_slurmctld();
slurm_sched_fini();
slurm_select_fini();
checkpoint_fini();
@@ -361,8 +362,7 @@ int main(int argc, char *argv[])
info("Slurmctld shutdown completing");
log_fini();
- jobacct_g_fini_slurmctld();
-
+
if (dump_core)
abort();
else
diff --git a/src/slurmd/common/stepd_api.c b/src/slurmd/common/stepd_api.c
index 9cd5e9b15bc66a6535618e8455ba6f6d8c79d997..b351c859c0384f1580518a810934a7512ac8908b 100644
--- a/src/slurmd/common/stepd_api.c
+++ b/src/slurmd/common/stepd_api.c
@@ -605,7 +605,6 @@ stepd_completion(int fd, step_complete_msg_t *sent)
safe_write(fd, &sent->range_last, sizeof(int));
safe_write(fd, &sent->step_rc, sizeof(int));
jobacct_g_setinfo(sent->jobacct, JOBACCT_DATA_PIPE, &fd);
-
/* Receive the return code and errno */
safe_read(fd, &rc, sizeof(int));
safe_read(fd, &errnum, sizeof(int));
diff --git a/src/slurmd/common/stepd_api.h b/src/slurmd/common/stepd_api.h
index 1a7d56c04c94ba20f3f3363969e1fa5004cf8df9..7523e3564008e87741ceb3da3ca49aaaac82913f 100644
--- a/src/slurmd/common/stepd_api.h
+++ b/src/slurmd/common/stepd_api.h
@@ -177,53 +177,4 @@ int stepd_resume(int fd);
*/
int stepd_completion(int fd, step_complete_msg_t *sent);
-#define safe_read(fd, buf, size) do { \
- int remaining = size; \
- void *ptr = buf; \
- int rc; \
- while (remaining > 0) { \
- rc = read(fd, ptr, remaining); \
- if (rc == 0) { \
- debug("%s:%d: %s: safe_read (%d of %d) EOF", \
- __FILE__, __LINE__, __CURRENT_FUNC__, \
- remaining, (int)size); \
- goto rwfail; \
- } else if (rc < 0) { \
- debug("%s:%d: %s: safe_read (%d of %d) failed: %m", \
- __FILE__, __LINE__, __CURRENT_FUNC__, \
- remaining, (int)size); \
- goto rwfail; \
- } else { \
- ptr += rc; \
- remaining -= rc; \
- if (remaining > 0) \
- debug3("%s:%d: %s: safe_read (%d of %d) partial read", \
- __FILE__, __LINE__, __CURRENT_FUNC__, \
- remaining, (int)size); \
- } \
- } \
- } while (0)
-
-#define safe_write(fd, buf, size) do { \
- int remaining = size; \
- void *ptr = buf; \
- int rc; \
- while(remaining > 0) { \
- rc = write(fd, ptr, remaining); \
- if (rc < 0) { \
- debug("%s:%d: %s: safe_write (%d of %d) failed: %m", \
- __FILE__, __LINE__, __CURRENT_FUNC__, \
- remaining, (int)size); \
- goto rwfail; \
- } else { \
- ptr += rc; \
- remaining -= rc; \
- if (remaining > 0) \
- debug3("%s:%d: %s: safe_write (%d of %d) partial write", \
- __FILE__, __LINE__, __CURRENT_FUNC__, \
- remaining, (int)size); \
- } \
- } \
- } while (0)
-
#endif /* _STEPD_API_H */
diff --git a/src/slurmd/slurmstepd/req.c b/src/slurmd/slurmstepd/req.c
index e2aeda866f53c97785887e9e94e7cb71704c1cfa..4de222b730c135dfae47e6f5039d0f240426d5b7 100644
--- a/src/slurmd/slurmstepd/req.c
+++ b/src/slurmd/slurmstepd/req.c
@@ -1009,7 +1009,6 @@ _handle_completion(int fd, slurmd_job_t *job, uid_t uid)
step_complete.step_rc = MAX(step_complete.step_rc, step_rc);
/************* acct stuff ********************/
- info("read it all now aggregate");
jobacct_g_aggregate(step_complete.jobacct, jobacct);
jobacct_g_free(jobacct);
/*********************************************/
diff --git a/src/slurmd/slurmstepd/slurmstepd.c b/src/slurmd/slurmstepd/slurmstepd.c
index c95bfc9a97d856f88e3a66caad1e48d2b6f97c60..63cfef0c001e9bb5dca633e6b41752bfe95ecde6 100644
--- a/src/slurmd/slurmstepd/slurmstepd.c
+++ b/src/slurmd/slurmstepd/slurmstepd.c
@@ -204,7 +204,7 @@ _init_from_slurmd(int sock, char **argv,
conf->log_opts.stderr_level = LOG_LEVEL_QUIET;
if (conf->logfile)
conf->log_opts.syslog_level = LOG_LEVEL_QUIET;
- } else
+ } else
conf->log_opts.syslog_level = LOG_LEVEL_QUIET;
log_init(argv[0],conf->log_opts, LOG_DAEMON, conf->logfile);
diff --git a/src/srun/msg.c b/src/srun/msg.c
index 3d4b289904e56cde0a8f8f20c97bf0e158ef32b6..fff81bf4443aa2d15db37ab7af7430ee7bc5ab3f 100644
--- a/src/srun/msg.c
+++ b/src/srun/msg.c
@@ -108,27 +108,6 @@ static void _node_fail_handler(char *nodelist, srun_job_t *job);
#define _poll_wr_isset(pfd) ((pfd).revents & POLLOUT)
#define _poll_err(pfd) ((pfd).revents & POLLERR)
-#undef safe_read
-#define safe_read(fd, ptr, size) do { \
- if (read(fd, ptr, size) != size) { \
- debug("%s:%d: %s: read (%d bytes) failed: %m", \
- __FILE__, __LINE__, __CURRENT_FUNC__, \
- (int)size); \
- goto rwfail; \
- } \
- } while (0)
-
-#undef safe_write
-#define safe_write(fd, ptr, size) do { \
- if (write(fd, ptr, size) != size) { \
- debug("%s:%d: %s: write (%d bytes) failed: %m", \
- __FILE__, __LINE__, __CURRENT_FUNC__, \
- (int)size); \
- goto rwfail; \
- } \
- } while (0)
-
-
/* fd is job->forked_msg->par_msg->msg_pipe[1] */
static void _update_mpir_proctable(int fd, srun_job_t *job,
int nodeid, int ntasks, uint32_t *pid,