From d2911e0a64d068830c8cf94d51d1057c34fb8c1e Mon Sep 17 00:00:00 2001
From: Moe Jette <jette1@llnl.gov>
Date: Mon, 24 Jul 2006 23:22:09 +0000
Subject: [PATCH] svn merge -r8670:8680
https://eris.llnl.gov/svn/slurm/branches/slurm-1.1
---
NEWS | 2 +-
src/slurmctld/agent.c | 35 +++++++++++++++++++++++++++++++--
src/srun/pmi.c | 18 +++++++++++++----
testsuite/expect/test7.2.prog.c | 8 ++++----
4 files changed, 52 insertions(+), 11 deletions(-)
diff --git a/NEWS b/NEWS
index 08e42e6e23a..e41c089bebc 100644
--- a/NEWS
+++ b/NEWS
@@ -1783,7 +1783,7 @@ documents those changes that are of interest to users and admins.
-- "fatal: _shm_unlock: Numerical result out of range" bug fixed in slurmd.
-- Config file parsing is now case insensitive.
-- SLURM_NODELIST environment variable now set in allocate mode.
-
+
* Changes in SLURM 0.2.0-pre2
=============================
diff --git a/src/slurmctld/agent.c b/src/slurmctld/agent.c
index 78c36d2eadd..ef25a028b57 100644
--- a/src/slurmctld/agent.c
+++ b/src/slurmctld/agent.c
@@ -1208,15 +1208,30 @@ extern int agent_retry (int min_wait)
int list_size = 0;
time_t now = time(NULL);
queued_request_t *queued_req_ptr = NULL;
+ agent_arg_t *agent_arg_ptr;
+ ListIterator retry_iter;
if (retry_list) {
static time_t last_msg_time = (time_t) 0;
+ uint32_t msg_type[5], i = 0;
list_size = list_count(retry_list);
if ((list_size > MAX_AGENT_CNT)
&& (difftime(now, last_msg_time) > 300)) {
/* Note sizable backlog of work */
info("WARNING: agent retry_list size is %d",
list_size);
+ retry_iter = list_iterator_create(retry_list);
+ while ((queued_req_ptr = (queued_request_t *)
+ list_next(retry_iter))) {
+ agent_arg_ptr = queued_req_ptr->agent_arg_ptr;
+ msg_type[i++] = agent_arg_ptr->msg_type;
+ if (i == 5)
+ break;
+ }
+ list_iterator_destroy(retry_iter);
+ info(" retry_list msg_type=%u,%u,%u,%u,%u",
+ msg_type[0], msg_type[1], msg_type[2],
+ msg_type[3], msg_type[4]);
last_msg_time = now;
}
}
@@ -1225,10 +1240,26 @@ extern int agent_retry (int min_wait)
slurm_mutex_lock(&retry_mutex);
if (retry_list) {
- ListIterator retry_iter;
+ /* first try to find a new (never tried) record */
+
+ retry_iter = list_iterator_create(retry_list);
+ while ((queued_req_ptr = (queued_request_t *)
+ list_next(retry_iter))) {
+ if (queued_req_ptr->last_attempt == 0)
+ list_remove(retry_iter);
+ list_size--;
+ break;
+ }
+ list_iterator_destroy(retry_iter);
+ }
+
+ if (retry_list && (queued_req_ptr == NULL)) {
+ /* now try to find a requeue request that is
+ * relatively old */
double age = 0;
retry_iter = list_iterator_create(retry_list);
+ /* next try to find an older record to retry */
while ((queued_req_ptr = (queued_request_t *)
list_next(retry_iter))) {
age = difftime(now, queued_req_ptr->last_attempt);
@@ -1243,7 +1274,7 @@ extern int agent_retry (int min_wait)
slurm_mutex_unlock(&retry_mutex);
if (queued_req_ptr) {
- agent_arg_t *agent_arg_ptr = queued_req_ptr->agent_arg_ptr;
+ agent_arg_ptr = queued_req_ptr->agent_arg_ptr;
xfree(queued_req_ptr);
if (agent_arg_ptr)
_spawn_retry_agent(agent_arg_ptr);
diff --git a/src/srun/pmi.c b/src/srun/pmi.c
index 289275d3cf4..bf40d48b71b 100644
--- a/src/srun/pmi.c
+++ b/src/srun/pmi.c
@@ -47,6 +47,7 @@
/* Global variables */
pthread_mutex_t kvs_mutex = PTHREAD_MUTEX_INITIALIZER;
int kvs_comm_cnt = 0;
+int kvs_updated = 0;
struct kvs_comm **kvs_comm_ptr = NULL;
struct barrier_resp {
@@ -95,15 +96,23 @@ static void _kvs_xmit_tasks(void)
#if _DEBUG
info("All tasks at barrier, transmit KVS keypairs now");
#endif
- /* copy the data */
+ /* reset barrier info */
args = xmalloc(sizeof(struct agent_arg));
args->barrier_xmit_ptr = barrier_ptr;
args->barrier_xmit_cnt = barrier_cnt;
barrier_ptr = NULL;
barrier_resp_cnt = 0;
barrier_cnt = 0;
- args->kvs_xmit_ptr = _kvs_comm_dup();
- args->kvs_xmit_cnt = kvs_comm_cnt;
+
+ /* copy the new kvs data */
+ if (kvs_updated) {
+ args->kvs_xmit_ptr = _kvs_comm_dup();
+ args->kvs_xmit_cnt = kvs_comm_cnt;
+ kvs_updated = 0;
+ } else { /* No new data to transmit */
+ args->kvs_xmit_ptr = xmalloc(0);
+ args->kvs_xmit_cnt = 0;
+ }
/* Spawn a pthread to transmit it */
slurm_attr_init(&attr);
@@ -316,12 +325,13 @@ extern int pmi_kvs_put(struct kvs_comm_set *kvs_set_ptr)
_merge_named_kvs(kvs_ptr,
kvs_set_ptr->kvs_comm_ptr[i]);
} else {
- _move_kvs(kvs_set_ptr-> kvs_comm_ptr[i]);
+ _move_kvs(kvs_set_ptr->kvs_comm_ptr[i]);
kvs_set_ptr-> kvs_comm_ptr[i] = NULL;
}
}
slurm_free_kvs_comm_set(kvs_set_ptr);
_print_kvs();
+ kvs_updated = 1;
pthread_mutex_unlock(&kvs_mutex);
return SLURM_SUCCESS;
}
diff --git a/testsuite/expect/test7.2.prog.c b/testsuite/expect/test7.2.prog.c
index ad46a923498..a73c636b3fa 100644
--- a/testsuite/expect/test7.2.prog.c
+++ b/testsuite/expect/test7.2.prog.c
@@ -34,11 +34,11 @@
#else
/* Typical MVAPICH2 use
*
- * Typically takes very long time for large task count
- * adjust job time limit and timeout in test7.2 as needed
+ * Adjust job time limit and timeout in test7.2 as needed
+ * for large values.
*/
-# define BARRIER_CNT 7
-# define PUTS_PER_BARRIER 32
+# define BARRIER_CNT 4
+# define PUTS_PER_BARRIER 0
#endif
#define OFFSET_1 1234
--
GitLab