From f0527ca74ba0bdb72b30c66fbd8fa78b77632bc3 Mon Sep 17 00:00:00 2001
From: Danny Auble <da@llnl.gov>
Date: Tue, 23 May 2006 23:07:29 +0000
Subject: [PATCH] mods to sfree to make it more robust
---
src/plugins/select/bluegene/plugin/bluegene.c | 9 ++--
src/plugins/select/bluegene/plugin/opts.c | 20 ++++---
src/plugins/select/bluegene/plugin/sfree.c | 53 +++++++++++--------
src/plugins/select/bluegene/plugin/sfree.h | 1 +
4 files changed, 51 insertions(+), 32 deletions(-)
diff --git a/src/plugins/select/bluegene/plugin/bluegene.c b/src/plugins/select/bluegene/plugin/bluegene.c
index c38351512e5..35d9131b5ce 100644
--- a/src/plugins/select/bluegene/plugin/bluegene.c
+++ b/src/plugins/select/bluegene/plugin/bluegene.c
@@ -842,11 +842,12 @@ extern int create_defined_blocks(bg_layout_t overlapped)
#ifdef HAVE_BG_FILES
bg_record_t *found_record = NULL;
- char *name = NULL;
int geo[BA_SYSTEM_DIMENSIONS];
int i;
ListIterator itr_found;
init_wires();
+#else
+ char *name = NULL;
#endif
slurm_mutex_lock(&block_state_mutex);
reset_ba_system();
@@ -1016,12 +1017,14 @@ extern int create_dynamic_block(ba_request_t *request, List my_block_list)
bg_record_t *bg_record = NULL;
List results = NULL;
uint16_t num_quarter=0, num_nodecard=0;
- char *name = NULL;
bitstr_t *my_bitmap = NULL;
int geo[BA_SYSTEM_DIMENSIONS];
int i;
blockreq_t blockreq;
-
+#ifndef HAVE_BG_FILES
+ char *name = NULL;
+#endif
+
slurm_mutex_lock(&block_state_mutex);
reset_ba_system();
diff --git a/src/plugins/select/bluegene/plugin/opts.c b/src/plugins/select/bluegene/plugin/opts.c
index 75c7f7ed50b..300e27578f6 100644
--- a/src/plugins/select/bluegene/plugin/opts.c
+++ b/src/plugins/select/bluegene/plugin/opts.c
@@ -44,6 +44,7 @@ void parse_command_line(int argc, char *argv[])
{"all", no_argument, 0, 'a'},
{"bgblock", required_argument, 0, 'b'},
{"partition", required_argument, 0, 'p'},
+ {"wait", no_argument, 0, 'w'},
{"version", no_argument, 0, 'V'},
{"help", no_argument, 0, 'h'},
{"usage", no_argument, 0, 'u'},
@@ -51,7 +52,7 @@ void parse_command_line(int argc, char *argv[])
};
while ((opt_char =
- getopt_long(argc, argv, "ab:hup:V",
+ getopt_long(argc, argv, "ab:hup:wV",
long_options, &option_index)) != -1) {
switch (opt_char) {
case (int) '?':
@@ -69,6 +70,9 @@ void parse_command_line(int argc, char *argv[])
case (int) 'p':
bg_block_id = optarg;
break;
+ case (int) 'w':
+ wait_full = true;
+ break;
case (int) 'h':
case (int) OPT_LONG_HELP:
_help();
@@ -114,7 +118,7 @@ static void _print_version(void)
static void _usage(void)
{
- printf("Usage: sfree [-huVa] [-b]\n");
+ printf("Usage: sfree [-huwVa] [-b]\n");
}
static void _help(void)
@@ -125,9 +129,13 @@ static void _help(void)
printf("\
Usage: sfree [OPTIONS]\n\
-b, --bgblock free specific bgblock named\n\
- -a, --all free all bgblocks\n\
- -V, --version output version information and exit\n\
+ -a, --all free all bgblocks\n\
+ -w, --wait wait to make sure all blocks have been freed\n\
+ (Otherwise sfree will start the free and once\n\
+ sure the block(s) have started to free will\n\
+ exit)\n\
+ -V, --version output version information and exit\n\
\nHelp options:\n\
- --help show this help message\n\
- --usage display brief usage message\n");
+ --help show this help message\n\
+ --usage display brief usage message\n");
}
diff --git a/src/plugins/select/bluegene/plugin/sfree.c b/src/plugins/select/bluegene/plugin/sfree.c
index 310e24be4d5..6052dece6c1 100644
--- a/src/plugins/select/bluegene/plugin/sfree.c
+++ b/src/plugins/select/bluegene/plugin/sfree.c
@@ -36,6 +36,7 @@
int all_blocks = 0;
char *bg_block_id = NULL;
+bool wait_full = false;
#ifdef HAVE_BG_FILES
@@ -222,7 +223,15 @@ int main(int argc, char *argv[])
free(bg_block_id);
- delete_record->state = NO_VAL;
+ if ((rc = rm_get_data(block_ptr,
+ RM_PartitionState,
+ &delete_record->state))
+ != STATUS_OK) {
+ error("rm_get_data"
+ "(RM_PartitionState): %s",
+ _bg_err_str(rc));
+ }
+
list_push(delete_record_list, delete_record);
slurm_attr_init(&attr_agent);
@@ -272,7 +281,7 @@ static int _free_block(delete_record_t *delete_record)
info("freeing bgblock %s", delete_record->bg_block_id);
_term_jobs_on_block(delete_record->bg_block_id);
while (1) {
- if (delete_record->state != NO_VAL
+ if (delete_record->state != (rm_partition_state_t)NO_VAL
&& delete_record->state != RM_PARTITION_FREE
&& delete_record->state != RM_PARTITION_DEALLOCATING) {
info("pm_destroy %s",delete_record->bg_block_id);
@@ -289,10 +298,13 @@ static int _free_block(delete_record_t *delete_record)
}
}
- if(i>5)
- delete_record->state = RM_PARTITION_FREE;
- i++;
-
+ if(!wait_full) {
+ if(i>5)
+ delete_record->state = RM_PARTITION_FREE;
+
+ i++;
+ }
+
if ((delete_record->state == RM_PARTITION_FREE)
|| (delete_record->state == RM_PARTITION_ERROR))
break;
@@ -312,7 +324,8 @@ static int _update_bg_record_state()
rm_partition_t *block_ptr = NULL;
delete_record_t *delete_record = NULL;
ListIterator itr;
-
+ int found = 0;
+
if ((rc = rm_get_partitions_info(block_state, &block_list))
!= STATUS_OK) {
error("rm_get_partitions_info(): %s", _bg_err_str(rc));
@@ -322,7 +335,7 @@ static int _update_bg_record_state()
if ((rc = rm_get_data(block_list, RM_PartListSize, &num_blocks))
!= STATUS_OK) {
error("rm_get_data(RM_PartListSize): %s", _bg_err_str(rc));
- state = NO_VAL;
+ state = -1;
num_blocks = 0;
}
@@ -334,7 +347,7 @@ static int _update_bg_record_state()
!= STATUS_OK) {
error("rm_get_data(RM_PartListNextPart): %s",
_bg_err_str(rc));
- state = NO_VAL;
+ state = -1;
break;
}
} else {
@@ -344,7 +357,7 @@ static int _update_bg_record_state()
!= STATUS_OK) {
error("rm_get_data(RM_PartListFirstPart: %s",
_bg_err_str(rc));
- state = NO_VAL;
+ state = -1;
break;
}
}
@@ -354,7 +367,7 @@ static int _update_bg_record_state()
!= STATUS_OK) {
error("rm_get_data(RM_PartitionID): %s",
_bg_err_str(rc));
- state = NO_VAL;
+ state = -1;
break;
}
@@ -362,7 +375,6 @@ static int _update_bg_record_state()
error("No Partition ID was returned from database");
continue;
}
-
itr = list_iterator_create(delete_record_list);
while ((delete_record =
(delete_record_t*) list_next(itr))) {
@@ -371,17 +383,8 @@ static int _update_bg_record_state()
if(strcmp(delete_record->bg_block_id, name)) {
continue;
}
+ state = 1;
- if(state == NO_VAL)
- goto clean_up;
- else if(j>=num_blocks) {
- error("This bgblock, %s, "
- "doesn't exist in MMCS",
- bg_block_id);
- state = NO_VAL;
- goto clean_up;
- }
-
if ((rc = rm_get_data(block_ptr,
RM_PartitionState,
&delete_record->state))
@@ -395,7 +398,11 @@ static int _update_bg_record_state()
list_iterator_destroy(itr);
free(name);
}
-clean_up:
+ if(state != 1) {
+ error("The requested block %s was not found in system.",
+ bg_block_id);
+ num_block_to_free = num_block_freed;
+ }
if ((rc = rm_free_partition_list(block_list)) != STATUS_OK) {
error("rm_free_partition_list(): %s", _bg_err_str(rc));
}
diff --git a/src/plugins/select/bluegene/plugin/sfree.h b/src/plugins/select/bluegene/plugin/sfree.h
index dd49a418547..46ce804a37e 100644
--- a/src/plugins/select/bluegene/plugin/sfree.h
+++ b/src/plugins/select/bluegene/plugin/sfree.h
@@ -45,6 +45,7 @@ typedef void (*sighandler_t) (int);
extern char *bg_block_id;
extern int all_blocks;
+extern bool wait_full;
extern void parse_command_line(int argc, char *argv[]);
--
GitLab