diff --git a/NEWS b/NEWS
index b45677413c4b9dde2e14a51920de9ce63df11b35..e8b3505728ce6b055c823e5b3b2865877c102837 100644
--- a/NEWS
+++ b/NEWS
@@ -7,6 +7,7 @@ documents those changes that are of interest to users and admins.
-- NOTE: "startclean" when transitioning from version 0.3, JOBS ARE LOST
-- Added support for job account information (arbitrary string)
-- Added support for job dependencies (start job X after job Y completes)
+ -- Added support for configuration parameter CheckpointType
* Changes in SLURM 0.3.7
========================
diff --git a/doc/man/man5/slurm.conf.5 b/doc/man/man5/slurm.conf.5
index bb44bc0fce7658f766836027b82b6a4faac5ecb8..2da94ce898fd465d29ee949e359c207f28f6bbdb 100644
--- a/doc/man/man5/slurm.conf.5
+++ b/doc/man/man5/slurm.conf.5
@@ -1,4 +1,4 @@
-.TH "slurm.conf" "5" "July 2004" "Morris Jette" "Slurm configuration file"
+.TH "slurm.conf" "5" "August 2004" "slurm.conf 0.4" "Slurm configuration file"
.SH "NAME"
slurm.conf \- Slurm configuration file
.SH "DESCRIPTION"
@@ -54,6 +54,11 @@ to a "standby" mode when the ControlMachine becomes available once again.
This should be a node name without the full domain name (e.g. "lx0002").
While not essential, it is recommended that you specify a backup controller.
.TP
+\fBCheckpointType\fR
+Define the system-initiated checkpoint method to be used for user jobs.
+The slurmctld daemon must be restarted for a change in CheckpointType
+to take effect.
+.TP
\fBControlAddr\fR
Name that \fBControlMachine\fR should be referred to in
establishing a communications path. This name will
diff --git a/slurm/slurm.h.in b/slurm/slurm.h.in
index b3ca0d82b11d9229c129a01cd1057b85d1690dde..232ccec277a1943b733804ad923f684667cb393e 100644
--- a/slurm/slurm.h.in
+++ b/slurm/slurm.h.in
@@ -407,6 +407,7 @@ typedef struct slurm_ctl_conf {
char *authtype; /* authentication type */
char *backup_addr; /* comm path of slurmctld secondary server */
char *backup_controller;/* name of slurmctld secondary server */
+ char *checkpoint_type; /* checkpoint plugin type */
char *control_addr; /* comm path of slurmctld primary server */
char *control_machine; /* name of slurmctld primary server */
char *epilog; /* pathname of job epilog */
diff --git a/slurm/slurm_errno.h b/slurm/slurm_errno.h
index 2702920a2033faa1a8d59b823bdecb6e0c87d963..befbbdd24dbfb463f51fba4011e4e9a8a7cf307c 100644
--- a/slurm/slurm_errno.h
+++ b/slurm/slurm_errno.h
@@ -125,6 +125,7 @@ enum {
ESLURM_INVALID_NODE_STATE,
ESLURM_INVALID_FEATURE,
ESLURM_INVALID_AUTHTYPE_CHANGE,
+ ESLURM_INVALID_CHECKPOINT_TYPE_CHANGE,
ESLURM_INVALID_SCHEDTYPE_CHANGE,
ESLURM_INVALID_SWITCHTYPE_CHANGE,
ESLURM_FRAGMENTATION,
diff --git a/src/api/config_info.c b/src/api/config_info.c
index 040d861c97090667a10fd475a2638d5e3bd90ecd..31caf68fe919ac49c005fe5bb30837e493307e53 100644
--- a/src/api/config_info.c
+++ b/src/api/config_info.c
@@ -4,7 +4,7 @@
*****************************************************************************
* Copyright (C) 2002 The Regents of the University of California.
* Produced at Lawrence Livermore National Laboratory (cf, DISCLAIMER).
- * Written by Moe Jette <jette1@llnl.gov> and Kevin Tew <tew1@llnl.gov>.
+ * Written by Morris Jette <jette1@llnl.gov> and Kevin Tew <tew1@llnl.gov>.
* UCRL-CODE-2002-040.
*
* This file is part of SLURM, a resource management program.
@@ -71,6 +71,8 @@ void slurm_print_ctl_conf ( FILE* out,
slurm_ctl_conf_ptr->backup_addr);
fprintf(out, "BackupController = %s\n",
slurm_ctl_conf_ptr->backup_controller);
+ fprintf(out, "CheckpointType = %s\n",
+ slurm_ctl_conf_ptr->checkpoint_type);
fprintf(out, "ControlAddr = %s\n",
slurm_ctl_conf_ptr->control_addr);
fprintf(out, "ControlMachine = %s\n",
diff --git a/src/common/checkpoint.c b/src/common/checkpoint.c
index cadb31a1cf322874bcfd9e6108c843e39ad62b7c..fa906fad3dc83bf59019f44072973b50442dbb2b 100644
--- a/src/common/checkpoint.c
+++ b/src/common/checkpoint.c
@@ -180,20 +180,14 @@ _slurm_checkpoint_get_ops( slurm_checkpoint_context_t c )
/* initialize checkpoint plugin */
extern int
-checkpoint_init(void)
+checkpoint_init(char *checkpoint_type)
{
int retval = SLURM_SUCCESS;
- char *checkpoint_type;
slurm_mutex_lock( &context_lock );
if ( g_context )
_slurm_checkpoint_context_destroy(g_context);
-#if 0
- checkpoint_type = slurm_get_checkpoint_type();
-#else
- checkpoint_type = xstrdup("checkpoint/none");
-#endif
g_context = _slurm_checkpoint_context_create( checkpoint_type );
if ( g_context == NULL ) {
error( "cannot create a context for %s", checkpoint_type );
@@ -209,7 +203,6 @@ checkpoint_init(void)
retval = SLURM_ERROR;
}
verbose("Checkpoint plugin loaded: %s", checkpoint_type);
- xfree(checkpoint_type);
done:
slurm_mutex_unlock( &context_lock );
diff --git a/src/common/checkpoint.h b/src/common/checkpoint.h
index c983b222a9fb1b941e76d66844fd5a1c3faaa29c..90dfd06984d18e3673ba43bbbd2f0c8c97ac78a2 100644
--- a/src/common/checkpoint.h
+++ b/src/common/checkpoint.h
@@ -53,7 +53,7 @@ enum check_opts {
typedef struct slurm_checkpoint_context * slurm_checkpoint_context_t;
/* initialize checkpoint plugin */
-extern int checkpoint_init(void);
+extern int checkpoint_init(char *checkpoint_type);
/* shutdown checkpoint plugin */
extern void checkpoint_fini(void);
diff --git a/src/common/read_config.c b/src/common/read_config.c
index 6f85f4bc894b15c59256d25ff2fe848fb9a745f9..0420ddce738e70706a32405021bb796df360e362 100644
--- a/src/common/read_config.c
+++ b/src/common/read_config.c
@@ -3,7 +3,7 @@
*****************************************************************************
* Copyright (C) 2002 The Regents of the University of California.
* Produced at Lawrence Livermore National Laboratory (cf, DISCLAIMER).
- * Written by Moe Mette <jette1@llnl.gov>.
+ * Written by Morris Jette <jette1@llnl.gov>.
* UCRL-CODE-2002-040.
*
* This file is part of SLURM, a resource management program.
@@ -94,6 +94,7 @@ void
free_slurm_conf (slurm_ctl_conf_t *ctl_conf_ptr)
{
xfree (ctl_conf_ptr->authtype);
+ xfree (ctl_conf_ptr->checkpoint_type);
xfree (ctl_conf_ptr->backup_addr);
xfree (ctl_conf_ptr->backup_controller);
xfree (ctl_conf_ptr->control_addr);
@@ -130,6 +131,7 @@ init_slurm_conf (slurm_ctl_conf_t *ctl_conf_ptr)
{
ctl_conf_ptr->last_update = time(NULL);
xfree (ctl_conf_ptr->authtype);
+ xfree (ctl_conf_ptr->checkpoint_type);
xfree (ctl_conf_ptr->backup_addr);
xfree (ctl_conf_ptr->backup_controller);
xfree (ctl_conf_ptr->control_addr);
@@ -199,7 +201,8 @@ parse_config_spec (char *in_line, slurm_ctl_conf_t *ctl_conf_ptr)
int max_job_cnt = -1, min_job_age = -1, wait_time = -1;
int slurmctld_port = -1, slurmd_port = -1;
char *backup_addr = NULL, *backup_controller = NULL;
- char *control_addr = NULL, *control_machine = NULL, *epilog = NULL;
+ char *checkpoint_type = NULL, *control_addr = NULL;
+ char *control_machine = NULL, *epilog = NULL;
char *prolog = NULL;
char *sched_type = NULL, *sched_auth = NULL;
char *state_save_location = NULL, *tmp_fs = NULL;
@@ -215,6 +218,7 @@ parse_config_spec (char *in_line, slurm_ctl_conf_t *ctl_conf_ptr)
error_code = slurm_parser (in_line,
"AuthType=", 's', &auth_type,
+ "CheckpointType=", 's', &checkpoint_type,
"BackupAddr=", 's', &backup_addr,
"BackupController=", 's', &backup_controller,
"ControlAddr=", 's', &control_addr,
@@ -268,6 +272,14 @@ parse_config_spec (char *in_line, slurm_ctl_conf_t *ctl_conf_ptr)
ctl_conf_ptr->authtype = auth_type;
}
+ if ( checkpoint_type ) {
+ if ( ctl_conf_ptr->checkpoint_type ) {
+ error( MULTIPLE_VALUE_MSG, "CheckpointType" );
+ xfree( ctl_conf_ptr->checkpoint_type );
+ }
+ ctl_conf_ptr->checkpoint_type = checkpoint_type;
+ }
+
if ( backup_addr ) {
if ( ctl_conf_ptr->backup_addr ) {
error (MULTIPLE_VALUE_MSG, "BackupAddr");
@@ -821,6 +833,10 @@ validate_config (slurm_ctl_conf_t *ctl_conf_ptr)
if (ctl_conf_ptr->authtype == NULL)
ctl_conf_ptr->authtype = xstrdup(DEFAULT_AUTH_TYPE);
+ if (ctl_conf_ptr->checkpoint_type == NULL)
+ ctl_conf_ptr->checkpoint_type =
+ xstrdup(DEFAULT_CHECKPOINT_TYPE);
+
if (ctl_conf_ptr->fast_schedule == (uint16_t) NO_VAL)
ctl_conf_ptr->fast_schedule = DEFAULT_FAST_SCHEDULE;
diff --git a/src/common/read_config.h b/src/common/read_config.h
index 65deae016d2d0061830149577aac0a4a8f4d4ac0..dbde82090d0d167751069e0e634fe28c3da50d06 100644
--- a/src/common/read_config.h
+++ b/src/common/read_config.h
@@ -4,7 +4,7 @@
*****************************************************************************
* Copyright (C) 2002 The Regents of the University of California.
* Produced at Lawrence Livermore National Laboratory (cf, DISCLAIMER).
- * Written by Moe Mette <jette1@llnl.gov>.
+ * Written by Morris Mette <jette1@llnl.gov>.
* UCRL-CODE-2002-040.
*
* This file is part of SLURM, a resource management program.
@@ -31,6 +31,7 @@
#include "src/common/slurm_protocol_defs.h"
#define DEFAULT_AUTH_TYPE "auth/none"
+#define DEFAULT_CHECKPOINT_TYPE "checkpoint/none"
#define DEFAULT_FAST_SCHEDULE 1
#define DEFAULT_FIRST_JOB_ID 1
#define DEFAULT_HASH_BASE 10
diff --git a/src/common/slurm_errno.c b/src/common/slurm_errno.c
index 6290cca0ff8ea5296902e9b198d5dc6a486fdb6a..35967e9f6dae4667198e673af05247be1468d92d 100644
--- a/src/common/slurm_errno.c
+++ b/src/common/slurm_errno.c
@@ -154,6 +154,8 @@ static slurm_errtab_t slurm_errtab[] = {
"Invalid feature specification" },
{ ESLURM_INVALID_AUTHTYPE_CHANGE,
"AuthType change requires restart of all SLURM daemons and commands"},
+ { ESLURM_INVALID_CHECKPOINT_TYPE_CHANGE,
+ "Invalid change in CheckpointType requested" },
{ ESLURM_INVALID_SCHEDTYPE_CHANGE,
"Invalid change in SchedulerType requested" },
{ ESLURM_INVALID_SWITCHTYPE_CHANGE,
diff --git a/src/common/slurm_protocol_pack.c b/src/common/slurm_protocol_pack.c
index 8548998fced128fcdd0f4db550ef5bd7ee34a513..61ae7d4d4a3215544aefd9b845b0addd5b61724c 100644
--- a/src/common/slurm_protocol_pack.c
+++ b/src/common/slurm_protocol_pack.c
@@ -1677,6 +1677,7 @@ _pack_slurm_ctl_conf_msg(slurm_ctl_conf_info_msg_t * build_ptr, Buf buffer)
packstr(build_ptr->authtype, buffer);
packstr(build_ptr->backup_addr, buffer);
packstr(build_ptr->backup_controller, buffer);
+ packstr(build_ptr->checkpoint_type, buffer);
packstr(build_ptr->control_addr, buffer);
packstr(build_ptr->control_machine, buffer);
packstr(build_ptr->epilog, buffer);
@@ -1736,6 +1737,8 @@ _unpack_slurm_ctl_conf_msg(slurm_ctl_conf_info_msg_t **
safe_unpackstr_xmalloc(&build_ptr->backup_addr, &uint16_tmp, buffer);
safe_unpackstr_xmalloc(&build_ptr->backup_controller, &uint16_tmp,
buffer);
+ safe_unpackstr_xmalloc(&build_ptr->checkpoint_type, &uint16_tmp,
+ buffer);
safe_unpackstr_xmalloc(&build_ptr->control_addr, &uint16_tmp, buffer);
safe_unpackstr_xmalloc(&build_ptr->control_machine, &uint16_tmp,
buffer);
diff --git a/src/slurmctld/controller.c b/src/slurmctld/controller.c
index ff2272ccfe3e35ca8a08fc324a23e4320c695e8f..f433581615848760efefdae96f079ad681197672 100644
--- a/src/slurmctld/controller.c
+++ b/src/slurmctld/controller.c
@@ -214,7 +214,8 @@ int main(int argc, char *argv[])
*/
if ( slurm_sched_init() != SLURM_SUCCESS )
fatal( "failed to initialize scheduling plugin" );
- if ( checkpoint_init() != SLURM_SUCCESS )
+ if ( checkpoint_init(slurmctld_conf.checkpoint_type) !=
+ SLURM_SUCCESS )
fatal( "failed to initialize checkpoint plugin" );
while (1) {
diff --git a/src/slurmctld/proc_req.c b/src/slurmctld/proc_req.c
index 3787f84f4f7d77f90f0384ad4ebe5b2603a7bf58..b11fb86ce9419d1ba66c5f23af4b762eb8cf7a78 100644
--- a/src/slurmctld/proc_req.c
+++ b/src/slurmctld/proc_req.c
@@ -63,7 +63,6 @@
#define BUF_SIZE 1024 /* Temporary buffer size */
static void _fill_ctld_conf(slurm_ctl_conf_t * build_ptr);
-static void _free_ctld_conf(slurm_ctl_conf_t * build_ptr);
static inline bool _is_super_user(uid_t uid);
static void _kill_job_on_msg_fail(uint32_t job_id);
static int _make_step_cred(struct step_record *step_rec,
@@ -245,6 +244,7 @@ void _fill_ctld_conf(slurm_ctl_conf_t * conf_ptr)
conf_ptr->backup_addr = xstrdup(slurmctld_conf.backup_addr);
conf_ptr->backup_controller = xstrdup(slurmctld_conf.
backup_controller);
+ conf_ptr->checkpoint_type = xstrdup(slurmctld_conf.checkpoint_type);
conf_ptr->control_addr = xstrdup(slurmctld_conf.control_addr);
conf_ptr->control_machine = xstrdup(slurmctld_conf.
control_machine);
@@ -298,35 +298,6 @@ void _fill_ctld_conf(slurm_ctl_conf_t * conf_ptr)
return;
}
-/* _free_ctld_conf - free memory allocated by _fill_ctld_conf */
-static void _free_ctld_conf(slurm_ctl_conf_t * conf_ptr)
-{
- xfree(conf_ptr->authtype);
- xfree(conf_ptr->backup_addr);
- xfree(conf_ptr->backup_controller);
- xfree(conf_ptr->control_addr);
- xfree(conf_ptr->control_machine);
- xfree(conf_ptr->epilog);
- xfree(conf_ptr->job_comp_loc);
- xfree(conf_ptr->job_comp_type);
- xfree(conf_ptr->job_credential_private_key);
- xfree(conf_ptr->job_credential_public_certificate);
- xfree(conf_ptr->plugindir);
- xfree(conf_ptr->prolog);
- xfree(conf_ptr->schedauth);
- xfree(conf_ptr->schedtype);
- xfree(conf_ptr->slurm_user_name);
- xfree(conf_ptr->slurmctld_logfile);
- xfree(conf_ptr->slurmctld_pidfile);
- xfree(conf_ptr->slurmd_logfile);
- xfree(conf_ptr->slurmd_pidfile);
- xfree(conf_ptr->slurmd_spooldir);
- xfree(conf_ptr->slurm_conf);
- xfree(conf_ptr->state_save_location);
- xfree(conf_ptr->switch_type);
- xfree(conf_ptr->tmp_fs);
-}
-
/* return true if supplied uid is a super-user: root, self, or SlurmUser */
static inline bool _is_super_user(uid_t uid)
{
@@ -608,7 +579,7 @@ static void _slurm_rpc_dump_conf(slurm_msg_t * msg)
/* send message */
slurm_send_node_msg(msg->conn_fd, &response_msg);
- _free_ctld_conf(&config_tbl);
+ free_slurm_conf(&config_tbl);
}
}
diff --git a/src/slurmctld/read_config.c b/src/slurmctld/read_config.c
index e6712281fb81b7f81f188debd2155103bae9d9d5..6b0c132e058f57e676543ae285d6a6e1bc8871b4 100644
--- a/src/slurmctld/read_config.c
+++ b/src/slurmctld/read_config.c
@@ -65,8 +65,8 @@ static void _purge_old_node_state(struct node_record *old_node_table_ptr,
static void _restore_node_state(struct node_record *old_node_table_ptr,
int old_node_record_count);
static int _preserve_plugins(slurm_ctl_conf_t * ctl_conf_ptr,
- char *old_auth_type, char *old_sched_type,
- char *old_switch_type);
+ char *old_auth_type, char *old_checkpoint_type,
+ char *old_sched_type, char *old_switch_type);
static int _sync_nodes_to_comp_job(void);
static int _sync_nodes_to_jobs(void);
static int _sync_nodes_to_active_job(struct job_record *job_ptr);
@@ -725,9 +725,10 @@ int read_slurm_conf(int recover)
int i, j, error_code;
int old_node_record_count;
struct node_record *old_node_table_ptr;
- char *old_auth_type = xstrdup(slurmctld_conf.authtype);
- char *old_sched_type = xstrdup(slurmctld_conf.schedtype);
- char *old_switch_type = xstrdup(slurmctld_conf.switch_type);
+ char *old_auth_type = xstrdup(slurmctld_conf.authtype);
+ char *old_checkpoint_type = xstrdup(slurmctld_conf.checkpoint_type);
+ char *old_sched_type = xstrdup(slurmctld_conf.schedtype);
+ char *old_switch_type = xstrdup(slurmctld_conf.switch_type);
/* initialization */
START_TIMER;
@@ -862,7 +863,8 @@ int read_slurm_conf(int recover)
/* Update plugins as possible */
error_code = _preserve_plugins(&slurmctld_conf,
- old_auth_type, old_sched_type, old_switch_type);
+ old_auth_type, old_checkpoint_type,
+ old_sched_type, old_switch_type);
slurmctld_conf.last_update = time(NULL);
END_TIMER;
@@ -915,8 +917,8 @@ static void _purge_old_node_state(struct node_record *old_node_table_ptr,
* RET zero or error code
*/
static int _preserve_plugins(slurm_ctl_conf_t * ctl_conf_ptr,
- char *old_auth_type, char *old_sched_type,
- char *old_switch_type)
+ char *old_auth_type, char *old_checkpoint_type,
+ char *old_sched_type, char *old_switch_type)
{
int rc = SLURM_SUCCESS;
@@ -929,6 +931,16 @@ static int _preserve_plugins(slurm_ctl_conf_t * ctl_conf_ptr,
xfree(old_auth_type);
}
+ if (old_checkpoint_type) {
+ if (strcmp(old_checkpoint_type,
+ ctl_conf_ptr->checkpoint_type)) {
+ xfree(ctl_conf_ptr->checkpoint_type);
+ ctl_conf_ptr->checkpoint_type = old_checkpoint_type;
+ rc = ESLURM_INVALID_CHECKPOINT_TYPE_CHANGE;
+ } else /* free duplicate value */
+ xfree(old_checkpoint_type);
+ }
+
if (old_sched_type) {
if (strcmp(old_sched_type, ctl_conf_ptr->schedtype)) {
xfree(ctl_conf_ptr->schedtype);