From 2551a44527b30baf0cd171b98acaf5a59cb3f5f9 Mon Sep 17 00:00:00 2001
From: Danny Auble <da@llnl.gov>
Date: Wed, 10 Jun 2009 22:22:43 +0000
Subject: [PATCH] svn merge -r17789:17805
https://eris.llnl.gov/svn/slurm/branches/slurm-2.0
---
NEWS | 1 +
src/common/read_config.c | 142 +++++++++---------
src/common/read_config.h | 2 +
.../mysql/accounting_storage_mysql.c | 2 +-
.../pgsql/accounting_storage_pgsql.c | 2 +-
src/plugins/jobcomp/mysql/jobcomp_mysql.c | 2 +-
src/plugins/jobcomp/pgsql/jobcomp_pgsql.c | 2 +-
src/plugins/select/cons_res/select_cons_res.c | 134 +++++++++++++----
src/sacctmgr/config_functions.c | 4 +
src/slurmctld/node_mgr.c | 2 +-
src/slurmd/slurmstepd/io.c | 50 ++++--
src/slurmd/slurmstepd/io.h | 1 +
src/slurmd/slurmstepd/mgr.c | 9 +-
src/slurmdbd/read_config.c | 16 +-
testsuite/expect/test1.84 | 2 +
15 files changed, 242 insertions(+), 129 deletions(-)
diff --git a/NEWS b/NEWS
index dfff9749c85..e15bac8dbc9 100644
--- a/NEWS
+++ b/NEWS
@@ -72,6 +72,7 @@ documents those changes that are of interest to users and admins.
memory affinity that was applied to the batch script.
-- Ignore the extra processors on a node above configured size if either
sched/gang or select/cons_res is configured.
+ -- Fix bug in tracking memory allocated on a node for select/cons_res plugin.
-- Fixed a race condition when writing labelled output with a file per task
or per node, which potentially closed a file before all data was written.
diff --git a/src/common/read_config.c b/src/common/read_config.c
index e6ba2eb5aea..6306a1fb6d1 100644
--- a/src/common/read_config.c
+++ b/src/common/read_config.c
@@ -114,23 +114,23 @@ bool nodehash_initialized = false;
static names_ll_t *host_to_node_hashtbl[NAME_HASH_LEN] = {NULL};
static names_ll_t *node_to_host_hashtbl[NAME_HASH_LEN] = {NULL};
-static int parse_nodename(void **dest, slurm_parser_enum_t type,
- const char *key, const char *value,
- const char *line, char **leftover);
-static void destroy_nodename(void *ptr);
-static int parse_partitionname(void **dest, slurm_parser_enum_t type,
- const char *key, const char *value,
- const char *line, char **leftover);
-static void destroy_partitionname(void *ptr);
-static int parse_downnodes(void **dest, slurm_parser_enum_t type,
+static int _parse_nodename(void **dest, slurm_parser_enum_t type,
const char *key, const char *value,
const char *line, char **leftover);
-static void destroy_downnodes(void *ptr);
-static int defunct_option(void **dest, slurm_parser_enum_t type,
- const char *key, const char *value,
- const char *line, char **leftover);
-static void validate_and_set_defaults(slurm_ctl_conf_t *conf,
- s_p_hashtbl_t *hashtbl);
+static void _destroy_nodename(void *ptr);
+static int _parse_partitionname(void **dest, slurm_parser_enum_t type,
+ const char *key, const char *value,
+ const char *line, char **leftover);
+static void _destroy_partitionname(void *ptr);
+static int _parse_downnodes(void **dest, slurm_parser_enum_t type,
+ const char *key, const char *value,
+ const char *line, char **leftover);
+static void _destroy_downnodes(void *ptr);
+static int _defunct_option(void **dest, slurm_parser_enum_t type,
+ const char *key, const char *value,
+ const char *line, char **leftover);
+static void _validate_and_set_defaults(slurm_ctl_conf_t *conf,
+ s_p_hashtbl_t *hashtbl);
s_p_options_t slurm_conf_options[] = {
{"AccountingStorageEnforce", S_P_STRING},
@@ -169,13 +169,13 @@ s_p_options_t slurm_conf_options[] = {
{"FastSchedule", S_P_UINT16},
{"FirstJobId", S_P_UINT32},
{"GetEnvTimeout", S_P_UINT16},
- {"HashBase", S_P_LONG, defunct_option},
- {"HeartbeatInterval", S_P_LONG, defunct_option},
+ {"HashBase", S_P_LONG, _defunct_option},
+ {"HeartbeatInterval", S_P_LONG, _defunct_option},
{"HealthCheckInterval", S_P_UINT16},
{"HealthCheckProgram", S_P_STRING},
{"InactiveLimit", S_P_UINT16},
{"JobAcctGatherType", S_P_STRING},
- {"JobAcctFrequency", S_P_UINT16, defunct_option},
+ {"JobAcctFrequency", S_P_UINT16, _defunct_option},
{"JobAcctGatherFrequency", S_P_UINT16},
{"JobAcctLogFile", S_P_STRING},
{"JobAcctType", S_P_STRING},
@@ -190,7 +190,7 @@ s_p_options_t slurm_conf_options[] = {
{"JobCredentialPublicCertificate", S_P_STRING},
{"JobFileAppend", S_P_UINT16},
{"JobRequeue", S_P_UINT16},
- {"KillTree", S_P_UINT16, defunct_option},
+ {"KillTree", S_P_UINT16, _defunct_option},
{"KillOnBadExit", S_P_UINT16},
{"KillWait", S_P_UINT16},
{"Licenses", S_P_STRING},
@@ -201,7 +201,7 @@ s_p_options_t slurm_conf_options[] = {
{"MaxMemPerTask", S_P_UINT32}, /* defunct */
{"MessageTimeout", S_P_UINT16},
{"MinJobAge", S_P_UINT16},
- {"MpichGmDirectSupport", S_P_LONG, defunct_option},
+ {"MpichGmDirectSupport", S_P_LONG, _defunct_option},
{"MpiDefault", S_P_STRING},
{"MpiParams", S_P_STRING},
{"OverTimeLimit", S_P_UINT16},
@@ -230,7 +230,7 @@ s_p_options_t slurm_conf_options[] = {
{"ResvOverRun", S_P_UINT16},
{"ReturnToService", S_P_UINT16},
{"SallocDefaultCommand", S_P_STRING},
- {"SchedulerAuth", S_P_STRING, defunct_option},
+ {"SchedulerAuth", S_P_STRING, _defunct_option},
{"SchedulerParameters", S_P_STRING},
{"SchedulerPort", S_P_UINT16},
{"SchedulerRootFilter", S_P_UINT16},
@@ -275,16 +275,16 @@ s_p_options_t slurm_conf_options[] = {
{"UsePAM", S_P_BOOLEAN},
{"WaitTime", S_P_UINT16},
- {"NodeName", S_P_ARRAY, parse_nodename, destroy_nodename},
- {"PartitionName", S_P_ARRAY, parse_partitionname,
- destroy_partitionname},
- {"DownNodes", S_P_ARRAY, parse_downnodes, destroy_downnodes},
+ {"NodeName", S_P_ARRAY, _parse_nodename, _destroy_nodename},
+ {"PartitionName", S_P_ARRAY, _parse_partitionname,
+ _destroy_partitionname},
+ {"DownNodes", S_P_ARRAY, _parse_downnodes, _destroy_downnodes},
{NULL}
};
-static int defunct_option(void **dest, slurm_parser_enum_t type,
+static int _defunct_option(void **dest, slurm_parser_enum_t type,
const char *key, const char *value,
const char *line, char **leftover)
{
@@ -325,9 +325,9 @@ static void _set_node_prefix(const char *nodenames)
#endif /* HAVE_BG */
-static int parse_nodename(void **dest, slurm_parser_enum_t type,
- const char *key, const char *value,
- const char *line, char **leftover)
+static int _parse_nodename(void **dest, slurm_parser_enum_t type,
+ const char *key, const char *value,
+ const char *line, char **leftover)
{
s_p_hashtbl_t *tbl, *dflt;
slurm_conf_node_t *n;
@@ -499,7 +499,7 @@ static int parse_nodename(void **dest, slurm_parser_enum_t type,
/* should not get here */
}
-static void destroy_nodename(void *ptr)
+static void _destroy_nodename(void *ptr)
{
slurm_conf_node_t *n = (slurm_conf_node_t *)ptr;
xfree(n->nodenames);
@@ -526,7 +526,7 @@ int slurm_conf_nodename_array(slurm_conf_node_t **ptr_array[])
}
-static int parse_partitionname(void **dest, slurm_parser_enum_t type,
+static int _parse_partitionname(void **dest, slurm_parser_enum_t type,
const char *key, const char *value,
const char *line, char **leftover)
{
@@ -604,7 +604,7 @@ static int parse_partitionname(void **dest, slurm_parser_enum_t type,
int max_time = time_str2mins(tmp);
if ((max_time < 0) && (max_time != INFINITE)) {
error("Bad value \"%s\" for MaxTime", tmp);
- destroy_partitionname(p);
+ _destroy_partitionname(p);
s_p_hashtbl_destroy(tbl);
xfree(tmp);
return -1;
@@ -620,7 +620,7 @@ static int parse_partitionname(void **dest, slurm_parser_enum_t type,
int default_time = time_str2mins(tmp);
if ((default_time < 0) && (default_time != INFINITE)) {
error("Bad value \"%s\" for DefaultTime", tmp);
- destroy_partitionname(p);
+ _destroy_partitionname(p);
s_p_hashtbl_destroy(tbl);
xfree(tmp);
return -1;
@@ -687,7 +687,7 @@ static int parse_partitionname(void **dest, slurm_parser_enum_t type,
#endif
else {
error("Bad value \"%s\" for Shared", tmp);
- destroy_partitionname(p);
+ _destroy_partitionname(p);
s_p_hashtbl_destroy(tbl);
xfree(tmp);
return -1;
@@ -711,7 +711,7 @@ static int parse_partitionname(void **dest, slurm_parser_enum_t type,
/* should not get here */
}
-static void destroy_partitionname(void *ptr)
+static void _destroy_partitionname(void *ptr)
{
slurm_conf_partition_t *p = (slurm_conf_partition_t *)ptr;
@@ -737,7 +737,7 @@ int slurm_conf_partition_array(slurm_conf_partition_t **ptr_array[])
}
}
-static int parse_downnodes(void **dest, slurm_parser_enum_t type,
+static int _parse_downnodes(void **dest, slurm_parser_enum_t type,
const char *key, const char *value,
const char *line, char **leftover)
{
@@ -771,7 +771,7 @@ static int parse_downnodes(void **dest, slurm_parser_enum_t type,
return 1;
}
-static void destroy_downnodes(void *ptr)
+static void _destroy_downnodes(void *ptr)
{
slurm_conf_downnodes_t *n = (slurm_conf_downnodes_t *)ptr;
xfree(n->nodenames);
@@ -1494,8 +1494,7 @@ init_slurm_conf (slurm_ctl_conf_t *ctl_conf_ptr)
}
/* caller must lock conf_lock */
-static void
-_init_slurm_conf(const char *file_name)
+static void _init_slurm_conf(const char *file_name)
{
char *name = (char *)file_name;
/* conf_ptr = (slurm_ctl_conf_t *)xmalloc(sizeof(slurm_ctl_conf_t)); */
@@ -1513,7 +1512,7 @@ _init_slurm_conf(const char *file_name)
if(s_p_parse_file(conf_hashtbl, name) == SLURM_ERROR)
fatal("something wrong with opening/reading conf file");
/* s_p_dump_values(conf_hashtbl, slurm_conf_options); */
- validate_and_set_defaults(conf_ptr, conf_hashtbl);
+ _validate_and_set_defaults(conf_ptr, conf_hashtbl);
conf_ptr->slurm_conf = xstrdup(name);
}
@@ -1681,7 +1680,7 @@ static void _normalize_debug_level(uint16_t *level)
* NOTE: if control_addr is NULL, it is over-written by control_machine
*/
static void
-validate_and_set_defaults(slurm_ctl_conf_t *conf, s_p_hashtbl_t *hashtbl)
+_validate_and_set_defaults(slurm_ctl_conf_t *conf, s_p_hashtbl_t *hashtbl)
{
char *temp_str = NULL;
long long_suspend_time;
@@ -1719,7 +1718,7 @@ validate_and_set_defaults(slurm_ctl_conf_t *conf, s_p_hashtbl_t *hashtbl)
conf->complete_wait = DEFAULT_COMPLETE_WAIT;
if (!s_p_get_string(&conf->control_machine, "ControlMachine", hashtbl))
- fatal ("validate_and_set_defaults: "
+ fatal ("_validate_and_set_defaults: "
"ControlMachine not specified.");
else if (strcasecmp("localhost", conf->control_machine) == 0) {
xfree (conf->control_machine);
@@ -1882,33 +1881,30 @@ validate_and_set_defaults(slurm_ctl_conf_t *conf, s_p_hashtbl_t *hashtbl)
if (!s_p_get_string(&conf->job_comp_host, "JobCompHost",
hashtbl)) {
if(default_storage_host)
- conf->job_comp_host =
- xstrdup(default_storage_host);
+ conf->job_comp_host = xstrdup(default_storage_host);
else
conf->job_comp_host = xstrdup(DEFAULT_STORAGE_HOST);
}
if (!s_p_get_string(&conf->job_comp_user, "JobCompUser",
hashtbl)) {
if(default_storage_user)
- conf->job_comp_user =
- xstrdup(default_storage_user);
+ conf->job_comp_user = xstrdup(default_storage_user);
else
conf->job_comp_user = xstrdup(DEFAULT_STORAGE_USER);
}
if (!s_p_get_string(&conf->job_comp_pass, "JobCompPass",
hashtbl)) {
if(default_storage_pass)
- conf->job_comp_pass =
- xstrdup(default_storage_pass);
+ conf->job_comp_pass = xstrdup(default_storage_pass);
}
if (!s_p_get_uint32(&conf->job_comp_port, "JobCompPort",
hashtbl)) {
if(default_storage_port)
conf->job_comp_port = default_storage_port;
else if(!strcmp(conf->job_comp_type, "job_comp/mysql"))
- conf->job_comp_port = 3306;
+ conf->job_comp_port = DEFAULT_MYSQL_PORT;
else if(!strcmp(conf->job_comp_type, "job_comp/pgsql"))
- conf->job_comp_port = 5432;
+ conf->job_comp_port = DEFAULT_PGSQL_PORT;
else
conf->job_comp_port = DEFAULT_STORAGE_PORT;
}
@@ -2033,6 +2029,15 @@ validate_and_set_defaults(slurm_ctl_conf_t *conf, s_p_hashtbl_t *hashtbl)
if(default_storage_loc)
conf->accounting_storage_loc =
xstrdup(default_storage_loc);
+ else if(!strcmp(conf->accounting_storage_type,
+ "accounting_storage/mysql")
+ || !strcmp(conf->accounting_storage_type,
+ "accounting_storage/pgsql"))
+ conf->accounting_storage_loc =
+ xstrdup(DEFAULT_ACCOUNTING_DB);
+ else
+ conf->accounting_storage_loc =
+ xstrdup(DEFAULT_STORAGE_LOC);
}
if (!s_p_get_string(&conf->accounting_storage_user,
"AccountingStorageUser", hashtbl)) {
@@ -2053,39 +2058,26 @@ validate_and_set_defaults(slurm_ctl_conf_t *conf, s_p_hashtbl_t *hashtbl)
"AccountingStoragePort", hashtbl)) {
if(default_storage_port)
conf->accounting_storage_port = default_storage_port;
+ else if(!strcmp(conf->accounting_storage_type,
+ "accounting_storage/slurmdbd"))
+ conf->accounting_storage_port = SLURMDBD_PORT;
+ else if(!strcmp(conf->accounting_storage_type,
+ "accounting_storage/mysql"))
+ conf->accounting_storage_port = DEFAULT_MYSQL_PORT;
+ else if(!strcmp(conf->accounting_storage_type,
+ "accounting_storage/pgsql"))
+ conf->accounting_storage_port = DEFAULT_PGSQL_PORT;
else
- conf->accounting_storage_port = NO_VAL;
+ conf->accounting_storage_port = DEFAULT_STORAGE_PORT;
}
- /* set correct defaults so scontrol show config works
- correctly */
+
+ /* remove the user and loc if using slurmdbd */
if(!strcmp(conf->accounting_storage_type,
"accounting_storage/slurmdbd")) {
xfree(conf->accounting_storage_loc);
conf->accounting_storage_loc = xstrdup("N/A");
xfree(conf->accounting_storage_user);
conf->accounting_storage_user = xstrdup("N/A");
- if(conf->accounting_storage_port == NO_VAL)
- conf->accounting_storage_port = SLURMDBD_PORT;
- } else if(!strcmp(conf->accounting_storage_type,
- "accounting_storage/mysql")) {
- if(conf->accounting_storage_port == NO_VAL)
- conf->accounting_storage_port = 3306;
- if(!conf->accounting_storage_loc)
- conf->accounting_storage_loc =
- xstrdup(DEFAULT_ACCOUNTING_DB);
- } else if(!strcmp(conf->accounting_storage_type,
- "accounting_storage/pgsql")) {
- if(conf->accounting_storage_port == NO_VAL)
- conf->accounting_storage_port = 5432;
- if(!conf->accounting_storage_loc)
- conf->accounting_storage_loc =
- xstrdup(DEFAULT_ACCOUNTING_DB);
- } else {
- if(conf->accounting_storage_port == NO_VAL)
- conf->accounting_storage_port = DEFAULT_STORAGE_PORT;
- if(!conf->accounting_storage_loc)
- conf->accounting_storage_loc =
- xstrdup(DEFAULT_STORAGE_LOC);
}
s_p_get_uint16(&conf->over_time_limit, "OverTimeLimit", hashtbl);
diff --git a/src/common/read_config.h b/src/common/read_config.h
index ce0973684c5..924d8517592 100644
--- a/src/common/read_config.h
+++ b/src/common/read_config.h
@@ -121,6 +121,8 @@ extern char *default_plugstack;
#define DEFAULT_STORAGE_LOC "/var/log/slurm_jobacct.log"
#define DEFAULT_STORAGE_USER "root"
#define DEFAULT_STORAGE_PORT 0
+#define DEFAULT_PGSQL_PORT 5432
+#define DEFAULT_MYSQL_PORT 3306
#define DEFAULT_SUSPEND_RATE 60
#define DEFAULT_SUSPEND_TIME 0
#define DEFAULT_SUSPEND_TIMEOUT 30
diff --git a/src/plugins/accounting_storage/mysql/accounting_storage_mysql.c b/src/plugins/accounting_storage/mysql/accounting_storage_mysql.c
index 5f2be55ddf1..431850f4599 100644
--- a/src/plugins/accounting_storage/mysql/accounting_storage_mysql.c
+++ b/src/plugins/accounting_storage/mysql/accounting_storage_mysql.c
@@ -2805,7 +2805,7 @@ static mysql_db_info_t *_mysql_acct_create_db_info()
mysql_db_info_t *db_info = xmalloc(sizeof(mysql_db_info_t));
db_info->port = slurm_get_accounting_storage_port();
if(!db_info->port) {
- db_info->port = 3306;
+ db_info->port = DEFAULT_MYSQL_PORT;
slurm_set_accounting_storage_port(db_info->port);
}
db_info->host = slurm_get_accounting_storage_host();
diff --git a/src/plugins/accounting_storage/pgsql/accounting_storage_pgsql.c b/src/plugins/accounting_storage/pgsql/accounting_storage_pgsql.c
index 3b6b3659c61..391d3be0576 100644
--- a/src/plugins/accounting_storage/pgsql/accounting_storage_pgsql.c
+++ b/src/plugins/accounting_storage/pgsql/accounting_storage_pgsql.c
@@ -137,7 +137,7 @@ static pgsql_db_info_t *_pgsql_acct_create_db_info()
/* it turns out it is better if using defaults to let postgres
handle them on it's own terms */
if(!db_info->port) {
- db_info->port = 5432;
+ db_info->port = DEFAULT_PGSQL_PORT;
slurm_set_accounting_storage_port(db_info->port);
}
db_info->host = slurm_get_accounting_storage_host();
diff --git a/src/plugins/jobcomp/mysql/jobcomp_mysql.c b/src/plugins/jobcomp/mysql/jobcomp_mysql.c
index 8e3182c83c7..59b88b1ba69 100644
--- a/src/plugins/jobcomp/mysql/jobcomp_mysql.c
+++ b/src/plugins/jobcomp/mysql/jobcomp_mysql.c
@@ -132,7 +132,7 @@ static mysql_db_info_t *_mysql_jobcomp_create_db_info()
mysql_db_info_t *db_info = xmalloc(sizeof(mysql_db_info_t));
db_info->port = slurm_get_jobcomp_port();
if(!db_info->port) {
- db_info->port = 3306;
+ db_info->port = DEFAULT_MYSQL_PORT;
slurm_set_jobcomp_port(db_info->port);
}
db_info->host = slurm_get_jobcomp_host();
diff --git a/src/plugins/jobcomp/pgsql/jobcomp_pgsql.c b/src/plugins/jobcomp/pgsql/jobcomp_pgsql.c
index 56df166c3b2..569fbf27b0e 100644
--- a/src/plugins/jobcomp/pgsql/jobcomp_pgsql.c
+++ b/src/plugins/jobcomp/pgsql/jobcomp_pgsql.c
@@ -131,7 +131,7 @@ static pgsql_db_info_t *_pgsql_jobcomp_create_db_info()
/* it turns out it is better if using defaults to let postgres
handle them on it's own terms */
if(!db_info->port) {
- db_info->port = 5432;
+ db_info->port = DEFAULT_PGSQL_PORT;
slurm_set_jobcomp_port(db_info->port);
}
db_info->host = slurm_get_jobcomp_host();
diff --git a/src/plugins/select/cons_res/select_cons_res.c b/src/plugins/select/cons_res/select_cons_res.c
index 14a4754daad..19b00a0d6f3 100644
--- a/src/plugins/select/cons_res/select_cons_res.c
+++ b/src/plugins/select/cons_res/select_cons_res.c
@@ -413,20 +413,6 @@ static struct node_use_record *_dup_node_usage(struct node_use_record *orig_ptr)
return new_use_ptr;
}
-/* Restore a node_state information */
-static void _restore_node_usage(struct node_use_record *orig_ptr)
-{
- uint32_t i;
-
- if (orig_ptr == NULL)
- return;
-
- for (i = 0; i < select_node_cnt; i++) {
- select_node_usage[i].node_state = orig_ptr[i].node_state;
- select_node_usage[i].alloc_memory = orig_ptr[i].alloc_memory;
- }
-}
-
/* delete the given row data */
static void _destroy_row_data(struct part_row_data *row, uint16_t num_rows) {
uint16_t i;
@@ -859,10 +845,12 @@ static int _add_job_to_res(struct job_record *job_ptr, int action)
select_node_usage[i].alloc_memory +=
job->memory_allocated[n];
if (select_node_usage[i].alloc_memory >
- select_node_record[i].real_memory) {
- error("error: node %s mem is overallocated(%u)",
- select_node_record[i].node_ptr->name,
- select_node_usage[i].alloc_memory);
+ select_node_record[i].real_memory) {
+ error("error: node %s mem is overallocated "
+ "(%u) for job %u",
+ select_node_record[i].node_ptr->name,
+ select_node_usage[i].alloc_memory,
+ job_ptr->job_id);
}
n++;
@@ -949,11 +937,13 @@ static int _rm_job_from_res(struct part_res_record *part_record_ptr,
if (!bit_test(job->node_bitmap, i))
continue;
if (node_usage[i].alloc_memory <
- job->memory_allocated[n]) {
- error("error: %s mem is underalloc'd(%u-%u)",
- select_node_record[i].node_ptr->name,
- node_usage[i].alloc_memory,
- job->memory_allocated[n]);
+ job->memory_allocated[n]) {
+ error("error: node %s mem is underallocated "
+ "(%u-%u) for job %u",
+ select_node_record[i].node_ptr->name,
+ node_usage[i].alloc_memory,
+ job->memory_allocated[n],
+ job_ptr->job_id);
node_usage[i].alloc_memory = 0;
} else {
node_usage[i].alloc_memory -=
@@ -1017,14 +1007,14 @@ static int _rm_job_from_res(struct part_res_record *part_record_ptr,
for (n = 0; n < select_node_cnt; n++) {
if (bit_test(job->node_bitmap, n) == 0)
continue;
- if (select_node_usage[n].node_state >=
+ if (node_usage[n].node_state >=
job->node_req) {
- select_node_usage[n].node_state -=
+ node_usage[n].node_state -=
job->node_req;
} else {
error("cons_res:_rm_job_from_res: "
"node_state mis-count");
- select_node_usage[n].node_state =
+ node_usage[n].node_state =
NODE_CR_AVAILABLE;
}
}
@@ -1511,6 +1501,98 @@ extern int select_p_job_list_test(List req_list)
return EINVAL;
}
+<<<<<<< .working
+=======
+
+/* _will_run_test - determine when and where a pending job can start, removes
+ * jobs from node table at termination time and run _test_job() after
+ * each one. */
+static int _will_run_test(struct job_record *job_ptr, bitstr_t *bitmap,
+ uint32_t min_nodes, uint32_t max_nodes,
+ uint32_t req_nodes, uint16_t job_node_req)
+{
+ struct part_res_record *future_part;
+ struct node_use_record *future_usage;
+ struct job_record *tmp_job_ptr, **tmp_job_pptr;
+ List cr_job_list;
+ ListIterator job_iterator;
+ bitstr_t *orig_map;
+ int rc = SLURM_ERROR;
+ time_t now = time(NULL);
+
+ orig_map = bit_copy(bitmap);
+
+ /* Try to run with currently available nodes */
+ rc = cr_job_test(job_ptr, bitmap, min_nodes, max_nodes, req_nodes,
+ SELECT_MODE_WILL_RUN, cr_type, job_node_req,
+ select_node_cnt, select_part_record,
+ select_node_usage);
+ if (rc == SLURM_SUCCESS) {
+ bit_free(orig_map);
+ job_ptr->start_time = time(NULL);
+ return SLURM_SUCCESS;
+ }
+
+ /* Job is still pending. Simulate termination of jobs one at a time
+ * to determine when and where the job can start. */
+
+ future_part = _dup_part_data(select_part_record);
+ if (future_part == NULL) {
+ bit_free(orig_map);
+ return SLURM_ERROR;
+ }
+ future_usage = _dup_node_usage(select_node_usage);
+ if (future_usage == NULL) {
+ _destroy_part_data(future_part);
+ bit_free(orig_map);
+ return SLURM_ERROR;
+ }
+
+ /* Build list of running jobs */
+ cr_job_list = list_create(NULL);
+ job_iterator = list_iterator_create(job_list);
+ while ((tmp_job_ptr = (struct job_record *) list_next(job_iterator))) {
+ if (tmp_job_ptr->job_state != JOB_RUNNING)
+ continue;
+ if (tmp_job_ptr->end_time == 0) {
+ error("Job %u has zero end_time", tmp_job_ptr->job_id);
+ continue;
+ }
+ tmp_job_pptr = xmalloc(sizeof(struct job_record *));
+ *tmp_job_pptr = tmp_job_ptr;
+ list_append(cr_job_list, tmp_job_pptr);
+ }
+ list_iterator_destroy(job_iterator);
+ list_sort(cr_job_list, _cr_job_list_sort);
+
+ /* Remove the running jobs one at a time from exp_node_cr and try
+ * scheduling the pending job after each one */
+ job_iterator = list_iterator_create(cr_job_list);
+ while ((tmp_job_pptr = (struct job_record **) list_next(job_iterator))) {
+ tmp_job_ptr = *tmp_job_pptr;
+ _rm_job_from_res(future_part, future_usage, tmp_job_ptr, 0);
+ bit_or(bitmap, orig_map);
+ rc = cr_job_test(job_ptr, bitmap, min_nodes, max_nodes,
+ req_nodes, SELECT_MODE_WILL_RUN, cr_type,
+ job_node_req, select_node_cnt, future_part,
+ future_usage);
+ if (rc == SLURM_SUCCESS) {
+ if (tmp_job_ptr->end_time <= now)
+ job_ptr->start_time = now + 1;
+ else
+ job_ptr->start_time = tmp_job_ptr->end_time;
+ break;
+ }
+ }
+ list_iterator_destroy(job_iterator);
+ list_destroy(cr_job_list);
+ _destroy_part_data(future_part);
+ _destroy_node_data(future_usage, NULL);
+ bit_free(orig_map);
+ return rc;
+}
+
+>>>>>>> .merge-right.r17805
extern int select_p_job_begin(struct job_record *job_ptr)
{
return SLURM_SUCCESS;
diff --git a/src/sacctmgr/config_functions.c b/src/sacctmgr/config_functions.c
index 76cda0381db..642d993a0d0 100644
--- a/src/sacctmgr/config_functions.c
+++ b/src/sacctmgr/config_functions.c
@@ -43,6 +43,7 @@
#include "src/common/xstring.h"
#include "src/sacctmgr/sacctmgr.h"
+static char *acct_storage_backup_host = NULL;
static char *acct_storage_host = NULL;
static char *acct_storage_loc = NULL;
static char *acct_storage_pass = NULL;
@@ -91,6 +92,7 @@ static void _free_dbd_config(void)
static void _load_slurm_config(void)
{
+ acct_storage_backup_host = slurm_get_accounting_storage_backup_host();
acct_storage_host = slurm_get_accounting_storage_host();
acct_storage_loc = slurm_get_accounting_storage_loc();
acct_storage_pass = slurm_get_accounting_storage_pass();
@@ -107,6 +109,7 @@ static void _load_slurm_config(void)
static void _free_slurm_config(void)
{
+ xfree(acct_storage_backup_host);
xfree(acct_storage_host);
xfree(acct_storage_loc);
xfree(acct_storage_pass);
@@ -123,6 +126,7 @@ static void _print_slurm_config(void)
slurm_make_time_str(&now, tmp_str, sizeof(tmp_str));
printf("Configuration data as of %s\n", tmp_str);
+ printf("AccountingStorageBackupHost = %s\n", acct_storage_backup_host);
printf("AccountingStorageHost = %s\n", acct_storage_host);
printf("AccountingStorageLoc = %s\n", acct_storage_loc);
printf("AccountingStoragePass = %s\n", acct_storage_pass);
diff --git a/src/slurmctld/node_mgr.c b/src/slurmctld/node_mgr.c
index 386d5948cfc..ad04a6cea23 100644
--- a/src/slurmctld/node_mgr.c
+++ b/src/slurmctld/node_mgr.c
@@ -1648,7 +1648,7 @@ extern int validate_node_specs(slurm_node_registration_status_msg_t *reg_msg)
threads1 = cores1 * reg_msg->threads;
cores2 = config_ptr->sockets * config_ptr->cores;
threads2 = cores2 * config_ptr->threads;
- if ((cores1 < cores2) || (threads1 < threads2)) {
+ if (threads1 < threads2) {
error("Node %s has low socket*core*thread count %u",
reg_msg->node_name, threads1);
error_code = EINVAL;
diff --git a/src/slurmd/slurmstepd/io.c b/src/slurmd/slurmstepd/io.c
index 3fecbd60c2d..061c14b8cbe 100644
--- a/src/slurmd/slurmstepd/io.c
+++ b/src/slurmd/slurmstepd/io.c
@@ -126,6 +126,9 @@ struct client_io_info {
int ltaskid_stdout, ltaskid_stderr;
bool labelio;
int label_width;
+
+ /* true if writing to a file, false if writing to a socket */
+ bool is_local_file;
};
@@ -475,20 +478,9 @@ static bool
_local_file_writable(eio_obj_t *obj)
{
struct client_io_info *client = (struct client_io_info *) obj->arg;
- int rc;
xassert(client->magic == CLIENT_IO_MAGIC);
- if (obj->shutdown) {
- if (obj->fd >= 0) {
- do {
- rc = close(obj->fd);
- } while (rc == -1 && errno == EINTR);
- obj->fd = -1;
- }
- return false;
- }
-
if (client->out_eof == true)
return false;
@@ -1297,6 +1289,33 @@ io_close_all(slurmd_job_t *job)
eio_signal_shutdown(job->eio);
}
+void
+io_close_local_fds(slurmd_job_t *job)
+{
+ ListIterator clients;
+ eio_obj_t *eio;
+ int rc;
+ struct client_io_info *client;
+
+ if (job == NULL || job->clients == NULL)
+ return;
+
+ clients = list_iterator_create(job->clients);
+ while((eio = list_next(clients))) {
+ client = (struct client_io_info *)eio->arg;
+ if (client->is_local_file) {
+ if (eio->fd >= 0) {
+ do {
+ rc = close(eio->fd);
+ } while (rc == -1 && errno == EINTR);
+ eio->fd = -1;
+ }
+ }
+ }
+}
+
+
+
static void *
_io_thr(void *arg)
{
@@ -1337,11 +1356,6 @@ io_create_local_client(const char *filename, int file_flags,
fd = open(filename, file_flags, 0666);
if (fd == -1) {
- /* error("Could not open stdout file: %m");
- task->ofname = fname_create(job, "slurm-%J.out", 0);
- fd = open(task->ofname, file_flags, 0666);
- if (fd == -1)
- return SLURM_ERROR; */
return ESLURMD_IO_ERROR;
}
fd_set_close_on_exec(fd);
@@ -1357,12 +1371,14 @@ io_create_local_client(const char *filename, int file_flags,
client->ltaskid_stdout = stdout_tasks;
client->ltaskid_stderr = stderr_tasks;
client->labelio = labelio;
+ client->is_local_file = true;
client->label_width = 1;
tmp = job->ntasks-1;
while ((tmp /= 10) > 0)
client->label_width++;
+
obj = eio_obj_create(fd, &local_file_ops, (void *)client);
list_append(job->clients, (void *)obj);
eio_new_initial_obj(job->eio, (void *)obj);
@@ -1430,6 +1446,7 @@ io_initial_client_connect(srun_info_t *srun, slurmd_job_t *job,
client->ltaskid_stderr = stderr_tasks;
client->labelio = false;
client->label_width = 0;
+ client->is_local_file = false;
obj = eio_obj_create(sock, &client_ops, (void *)client);
list_append(job->clients, (void *)obj);
@@ -1489,6 +1506,7 @@ io_client_connect(srun_info_t *srun, slurmd_job_t *job)
client->ltaskid_stderr = -1; /* accept from all tasks */
client->labelio = false;
client->label_width = 0;
+ client->is_local_file = false;
/* client object adds itself to job->clients in _client_writable */
diff --git a/src/slurmd/slurmstepd/io.h b/src/slurmd/slurmstepd/io.h
index 7e9894e4aaa..53026256ccc 100644
--- a/src/slurmd/slurmstepd/io.h
+++ b/src/slurmd/slurmstepd/io.h
@@ -130,6 +130,7 @@ void io_close_task_fds(slurmd_job_t *job);
void io_close_all(slurmd_job_t *job);
+void io_close_local_fds(slurmd_job_t *job);
/*
diff --git a/src/slurmd/slurmstepd/mgr.c b/src/slurmd/slurmstepd/mgr.c
index 8a4c259d81c..c6267cf5868 100644
--- a/src/slurmd/slurmstepd/mgr.c
+++ b/src/slurmd/slurmstepd/mgr.c
@@ -919,11 +919,9 @@ job_manager(slurmd_job_t *job)
/*
* Wait for io thread to complete (if there is one)
*/
- if (!job->batch && !job->user_managed_io && io_initialized) {
- eio_signal_shutdown(job->eio);
+ if (!job->batch && !job->user_managed_io && io_initialized)
_wait_for_io(job);
- }
-
+
debug2("Before call to spank_fini()");
if (spank_fini (job) < 0) {
error ("spank_fini failed\n");
@@ -1472,6 +1470,9 @@ _wait_for_io(slurmd_job_t *job)
} else
info("_wait_for_io: ioid==0");
+ /* Close any files for stdout/stderr opened by the stepd */
+ io_close_local_fds(job);
+
return;
}
diff --git a/src/slurmdbd/read_config.c b/src/slurmdbd/read_config.c
index 4e6c88d1350..e5819561f68 100644
--- a/src/slurmdbd/read_config.c
+++ b/src/slurmdbd/read_config.c
@@ -314,23 +314,33 @@ extern int read_slurmdbd_conf(void)
if (slurmdbd_conf->storage_type == NULL)
fatal("StorageType must be specified");
+ if (!slurmdbd_conf->storage_host)
+ slurmdbd_conf->storage_host = xstrdup(DEFAULT_STORAGE_HOST);
+
+ if (!slurmdbd_conf->storage_user)
+ slurmdbd_conf->storage_user = xstrdup(getlogin());
+
if(!strcmp(slurmdbd_conf->storage_type,
"accounting_storage/mysql")) {
if(!slurmdbd_conf->storage_port)
- slurmdbd_conf->storage_port = 3306;
+ slurmdbd_conf->storage_port = DEFAULT_MYSQL_PORT;
if(!slurmdbd_conf->storage_loc)
slurmdbd_conf->storage_loc =
xstrdup(DEFAULT_ACCOUNTING_DB);
} else if(!strcmp(slurmdbd_conf->storage_type,
"accounting_storage/pgsql")) {
if(!slurmdbd_conf->storage_port)
- slurmdbd_conf->storage_port = 5432;
+ slurmdbd_conf->storage_port = DEFAULT_PGSQL_PORT;
if(!slurmdbd_conf->storage_loc)
slurmdbd_conf->storage_loc =
xstrdup(DEFAULT_ACCOUNTING_DB);
- } else
+ } else {
if(!slurmdbd_conf->storage_port)
slurmdbd_conf->storage_port = DEFAULT_STORAGE_PORT;
+ if(!slurmdbd_conf->storage_loc)
+ slurmdbd_conf->storage_loc =
+ xstrdup(DEFAULT_STORAGE_LOC);
+ }
if (slurmdbd_conf->archive_dir) {
if(stat(slurmdbd_conf->archive_dir, &buf) < 0)
diff --git a/testsuite/expect/test1.84 b/testsuite/expect/test1.84
index aadf24cbf77..b301a657066 100755
--- a/testsuite/expect/test1.84
+++ b/testsuite/expect/test1.84
@@ -113,6 +113,7 @@ if {[string compare $host ""] == 0} {
if {$cpu_cnt != $task_cnt} {
send_user "FAILURE: should have run $cpu_cnt tasks (one per CPU) "
send_user "instead of $task_cnt tasks\n"
+ send_user "NOTE: This could be due to memory limit per allocated CPU\n\n"
set exit_code 1
}
if {$cpu_cnt < 2} {
@@ -151,6 +152,7 @@ expect {
#
if {$task_cnt != [expr $cpu_cnt / 2]} {
send_user "\nFAILURE: Improper task count for given cpus-per-task\n"
+ send_user "NOTE: This could be due to memory limit per allocated CPU\n\n"
set exit_code 1
}
--
GitLab