From 4be7316339ef6a386b9d779366b4e68bea2e2f0a Mon Sep 17 00:00:00 2001
From: Danny Auble <da@schedmd.com>
Date: Tue, 31 Jul 2012 16:02:57 -0700
Subject: [PATCH] Fix for sacct --state=S
Move code to handle multiple clusters. Previous code could have issues
where job_db_inx space could overlap.
---
.../mysql/as_mysql_jobacct_process.c | 187 ++++++++----------
.../mysql/as_mysql_jobacct_process.h | 6 +-
2 files changed, 84 insertions(+), 109 deletions(-)
diff --git a/src/plugins/accounting_storage/mysql/as_mysql_jobacct_process.c b/src/plugins/accounting_storage/mysql/as_mysql_jobacct_process.c
index c2236161d4e..cc5261a55ae 100644
--- a/src/plugins/accounting_storage/mysql/as_mysql_jobacct_process.c
+++ b/src/plugins/accounting_storage/mysql/as_mysql_jobacct_process.c
@@ -205,18 +205,10 @@ enum {
STEP_REQ_COUNT
};
-static void _state_time_string(mysql_conn_t *mysql_conn,
- slurmdb_job_cond_t *job_cond,
- char **extra, uint32_t state,
+static void _state_time_string(char **extra, uint32_t state,
uint32_t start, uint32_t end)
{
- MYSQL_RES *result = NULL;
- MYSQL_ROW row;
- ListIterator itr = NULL;
- char *query = NULL, *cluster_name = NULL;
int base_state = state & JOB_STATE_BASE;
- int set = 0, added = 0;
- List use_cluster_list = as_mysql_cluster_list;
if (!start && !end) {
xstrfmtcat(*extra, "t1.state='%u'", state);
@@ -250,96 +242,7 @@ static void _state_time_string(mysql_conn_t *mysql_conn,
}
break;
case JOB_SUSPENDED:
- /* Since we could be looking in the past for suspended
- * jobs we need to first look for jobs in each of the
- * cluster's suspend_table to see if we can use that
- * first.
- */
- if (job_cond->cluster_list
- && list_count(job_cond->cluster_list))
- use_cluster_list = job_cond->cluster_list;
- else
- slurm_mutex_lock(&as_mysql_cluster_list_lock);
-
- itr = list_iterator_create(use_cluster_list);
- while ((cluster_name = list_next(itr))) {
- query = xstrdup_printf(
- "select job_db_inx from \"%s_%s\"",
- cluster_name, suspend_table);
- if (start) {
- if (!end) {
- xstrfmtcat(query,
- " where (%d between "
- "time_start and time_end)",
- start);
- } else {
- xstrfmtcat(query,
- " where (time_start && "
- "((%d between time_start "
- "and time_end) "
- "|| (time_start between "
- "%d and %d)))",
- start, start, end);
- }
- } else if (end) {
- xstrfmtcat(query, " where (time_start && "
- "time_start < %d)", end);
- }
-
- debug3("%d(%s:%d) query\n%s",
- mysql_conn->conn, THIS_FILE, __LINE__, query);
- result = mysql_db_query_ret(mysql_conn, query, 0);
- xfree(query);
- if (!result)
- continue;
- set = 0;
- while ((row = mysql_fetch_row(result))) {
- if (set)
- xstrfmtcat(*extra,
- " || job_db_inx=%s", row[0]);
- else {
- if (added)
- xstrcat(*extra, ") || (");
- else {
- added = 1;
- xstrcat(*extra, "(");
- }
- set = 1;
- xstrfmtcat(*extra,
- "(job_db_inx=%s", row[0]);
- }
- }
- mysql_free_result(result);
- }
- if (use_cluster_list == as_mysql_cluster_list)
- slurm_mutex_unlock(&as_mysql_cluster_list_lock);
-
- /* If we didn't find anything we need to put something
- here to avoid a blank () in the calling sql.
- */
- if (added)
- xstrcat(*extra, "))");
- else if (start) {
- if (!end)
- xstrfmtcat(*extra,
- "(t1.time_start && "
- "((!t1.time_end && t1.state=%d) || "
- "(%d between t1.time_start "
- "and t1.time_end)))",
- JOB_SUSPENDED, start);
- else
- xstrfmtcat(*extra,
- "(t1.time_start && "
- "((%d between t1.time_start "
- "and t1.time_end) "
- "|| (t1.time_start between "
- "%d and %d)))",
- start, start,
- end);
- } else if (end)
- xstrfmtcat(*extra, "(t1.time_start && "
- "t1.time_start < %d)", end);
- break;
+ /* Handle this the same way we handle RUNNING. */
case JOB_RUNNING:
if (start) {
if (!end) {
@@ -348,7 +251,7 @@ static void _state_time_string(mysql_conn_t *mysql_conn,
"((!t1.time_end && t1.state=%d) || "
"(%d between t1.time_start "
"and t1.time_end)))",
- JOB_RUNNING, start);
+ base_state, start);
} else {
xstrfmtcat(*extra,
"(t1.time_start && "
@@ -1074,16 +977,16 @@ extern int good_nodes_from_inx(List local_cluster_list,
return 1;
}
-extern char *setup_job_cluster_cond_limits(mysql_conn_t *mysql_conn,
- slurmdb_job_cond_t *job_cond,
- char *cluster_name, char **extra)
+extern int setup_job_cluster_cond_limits(mysql_conn_t *mysql_conn,
+ slurmdb_job_cond_t *job_cond,
+ char *cluster_name, char **extra)
{
int set = 0;
ListIterator itr = NULL;
char *object = NULL;
if (!job_cond)
- return NULL;
+ return SLURM_SUCCESS;
/* this must be done before resvid_list since we set
resvid_list up here */
@@ -1137,6 +1040,78 @@ no_resv:
xstrcat(*extra, ")");
}
+ if (job_cond->state_list && list_count(job_cond->state_list)) {
+ itr = list_iterator_create(job_cond->state_list);
+ while ((object = list_next(itr))) {
+ uint32_t state = (uint32_t)slurm_atoul(object);
+ state &= JOB_STATE_BASE;
+ if (state == JOB_SUSPENDED)
+ break;
+ }
+ list_iterator_destroy(itr);
+
+ if (object) {
+ MYSQL_RES *result = NULL;
+ MYSQL_ROW row;
+ char *query = xstrdup_printf(
+ "select job_db_inx from \"%s_%s\"",
+ cluster_name, suspend_table);
+ if (job_cond->usage_start) {
+ if (!job_cond->usage_end) {
+ xstrfmtcat(query,
+ " where (!time_end "
+ "|| (%d between "
+ "time_start and time_end))",
+ (int)job_cond->usage_start);
+ } else {
+ xstrfmtcat(query,
+ " where (!time_end "
+ "|| (time_start && "
+ "((%d between time_start "
+ "and time_end) "
+ "|| (time_start between "
+ "%d and %d))))",
+ (int)job_cond->usage_start,
+ (int)job_cond->usage_start,
+ (int)job_cond->usage_end);
+ }
+ } else if (job_cond->usage_end) {
+ xstrfmtcat(query, " where (time_start && "
+ "time_start < %d)",
+ (int)job_cond->usage_end);
+ }
+
+ debug3("%d(%s:%d) query\n%s",
+ mysql_conn->conn, THIS_FILE, __LINE__, query);
+ result = mysql_db_query_ret(mysql_conn, query, 0);
+ xfree(query);
+ if (!result)
+ return SLURM_ERROR;
+ set = 0;
+ while ((row = mysql_fetch_row(result))) {
+ if (set)
+ xstrfmtcat(*extra,
+ " || t1.job_db_inx=%s",
+ row[0]);
+ else {
+ set = 1;
+ if (*extra)
+ xstrfmtcat(
+ *extra,
+ " || (t1.job_db_inx=%s",
+ row[0]);
+ else
+ xstrfmtcat(*extra, " where "
+ "(t1.job_db_inx=%s",
+ row[0]);
+ }
+ }
+ mysql_free_result(result);
+ if (set)
+ xstrcat(*extra, ")");
+ }
+ }
+
return SLURM_SUCCESS;
}
@@ -1355,8 +1330,8 @@ extern int setup_job_cond_limits(mysql_conn_t *mysql_conn,
while ((object = list_next(itr))) {
if (set)
xstrcat(*extra, " || ");
- _state_time_string(mysql_conn, job_cond,
- extra, (uint32_t)slurm_atoul(object),
+
+ _state_time_string(extra, (uint32_t)slurm_atoul(object),
job_cond->usage_start,
job_cond->usage_end);
set = 1;
diff --git a/src/plugins/accounting_storage/mysql/as_mysql_jobacct_process.h b/src/plugins/accounting_storage/mysql/as_mysql_jobacct_process.h
index 47cdd3bf949..719d02306de 100644
--- a/src/plugins/accounting_storage/mysql/as_mysql_jobacct_process.h
+++ b/src/plugins/accounting_storage/mysql/as_mysql_jobacct_process.h
@@ -52,9 +52,9 @@ extern List setup_cluster_list_with_inx(mysql_conn_t *mysql_conn,
extern int good_nodes_from_inx(List local_cluster_list,
void **object, char *node_inx,
int submit);
-extern char *setup_job_cluster_cond_limits(mysql_conn_t *mysql_conn,
- slurmdb_job_cond_t *job_cond,
- char *cluster_name, char **extra);
+extern int setup_job_cluster_cond_limits(mysql_conn_t *mysql_conn,
+ slurmdb_job_cond_t *job_cond,
+ char *cluster_name, char **extra);
extern int setup_job_cond_limits(mysql_conn_t *mysql_conn,
slurmdb_job_cond_t *job_cond,
char **extra);
--
GitLab