From ccacd390233e075dd6953869ff25366b4198a2fe Mon Sep 17 00:00:00 2001
From: Danny Auble <da@llnl.gov>
Date: Thu, 15 Oct 2009 17:40:33 +0000
Subject: [PATCH] svn merge -r18877:18906
https://eris.llnl.gov/svn/slurm/branches/slurm-2.0
---
NEWS | 2 ++
.../bluegene/block_allocator/bridge_linker.c | 18 +++++++++++++++---
.../select/bluegene/plugin/bg_block_info.c | 19 +++++++++++++++++++
src/plugins/select/bluegene/plugin/bluegene.c | 17 ++++++++++-------
4 files changed, 46 insertions(+), 10 deletions(-)
diff --git a/NEWS b/NEWS
index c9a7f83f929..d2b86f9ad95 100644
--- a/NEWS
+++ b/NEWS
@@ -273,6 +273,8 @@ documents those changes that are of interest to users and admins.
-- Add range check for SuspendTime configuration parameter.
-- Moved unzipped python-hostname tarball out and the tarball in.
-- BLUEGENE - Patched memory leak when running state test.
+ -- BLUEGENE - fixed slow down generated by slow call rm_get_BG
+ and polling thread.
* Changes in SLURM 2.0.6
========================
diff --git a/src/plugins/select/bluegene/block_allocator/bridge_linker.c b/src/plugins/select/bluegene/block_allocator/bridge_linker.c
index f16d3ca052a..5b386bc44a4 100644
--- a/src/plugins/select/bluegene/block_allocator/bridge_linker.c
+++ b/src/plugins/select/bluegene/block_allocator/bridge_linker.c
@@ -251,8 +251,9 @@ extern int bridge_fini()
{
if(handle)
dlclose(handle);
-
- return SLURM_ERROR;
+ initialized = false;
+
+ return SLURM_SUCCESS;
}
extern status_t bridge_get_bg(my_bluegene_t **bg)
@@ -300,8 +301,19 @@ extern status_t bridge_get_block_info(pm_partition_id_t pid,
int rc = CONNECTION_ERROR;
if(!bridge_init())
return rc;
+
+ /* this is here to make sure we don't lock up things with
+ polling and the long running get_BG call */
+ rc = pthread_mutex_trylock(&api_file_mutex);
+ if (rc == EBUSY)
+ return rc;
+ else if(rc) {
+ errno = rc;
+ error("%s:%d %s: pthread_mutex_trylock(): %m",
+ __FILE__, __LINE__, __CURRENT_FUNC__);
+ }
- slurm_mutex_lock(&api_file_mutex);
+ //slurm_mutex_lock(&api_file_mutex);
rc = (*(bridge_api.get_partition_info))(pid, partition);
slurm_mutex_unlock(&api_file_mutex);
return rc;
diff --git a/src/plugins/select/bluegene/plugin/bg_block_info.c b/src/plugins/select/bluegene/plugin/bg_block_info.c
index 76e8809f1e5..d6c4f4e75fb 100644
--- a/src/plugins/select/bluegene/plugin/bg_block_info.c
+++ b/src/plugins/select/bluegene/plugin/bg_block_info.c
@@ -283,6 +283,16 @@ extern int update_block_list()
break;
}
}
+
+ /* If the call was busy, just skip this
+ iteration. It usually means something like
+ rm_get_BG was called which can be a very
+ long call */
+ if(rc == EBUSY) {
+ debug5("lock was busy, aborting");
+ break;
+ }
+
error("bridge_get_block_info(%s): %s",
name,
bg_err_str(rc));
@@ -621,6 +631,15 @@ extern int update_freeing_block_list()
break;
}
}
+ /* If the call was busy, just skip this
+ iteration. It usually means something like
+ rm_get_BG was called which can be a very
+ long call */
+ if(rc == EBUSY) {
+ debug5("lock was busy, aborting");
+ break;
+ }
+
error("bridge_get_block_info(%s): %s",
name,
bg_err_str(rc));
diff --git a/src/plugins/select/bluegene/plugin/bluegene.c b/src/plugins/select/bluegene/plugin/bluegene.c
index 78b8cccadaa..0e766499046 100644
--- a/src/plugins/select/bluegene/plugin/bluegene.c
+++ b/src/plugins/select/bluegene/plugin/bluegene.c
@@ -41,9 +41,10 @@
#include "defined_block.h"
#include <stdio.h>
-#define MMCS_POLL_TIME 30 /* poll MMCS for down switches and nodes
- * every 120 secs */
-#define BG_POLL_TIME 0 /* poll bg blocks every 3 secs */
+#define MMCS_POLL_TIME 30 /* seconds between poll of MMCS for
+ * down switches and nodes */
+#define BG_POLL_TIME 1 /* seconds between poll of state
+ * change in bg blocks */
#define _DEBUG 0
@@ -313,7 +314,6 @@ extern void *block_agent(void *args)
last_bg_test = now - BG_POLL_TIME;
while (!agent_fini) {
-
if (difftime(now, last_bg_test) >= BG_POLL_TIME) {
if (agent_fini) /* don't bother */
break; /* quit now */
@@ -332,10 +332,10 @@ extern void *block_agent(void *args)
"update_block_list 2");
}
}
- now = time(NULL);
}
sleep(1);
+ now = time(NULL);
}
return NULL;
}
@@ -356,9 +356,12 @@ extern void *state_agent(void *args)
if (agent_fini) /* don't bother */
break; /* quit now */
if(blocks_are_created) {
- last_mmcs_test = now;
- /* can run for a while */
+ /* can run for a while so set the
+ * time after the call so there is
+ * always MMCS_POLL_TIME between
+ * calls */
test_mmcs_failures();
+ last_mmcs_test = time(NULL);
}
}
--
GitLab