From 5a8dbc3904c4e1acf9e128e21750de7796ee7342 Mon Sep 17 00:00:00 2001
From: Moe Jette <jette1@llnl.gov>
Date: Fri, 16 Dec 2005 17:21:21 +0000
Subject: [PATCH] Change suspend/resume from job step to job granularity.
---
doc/man/man1/scontrol.1 | 14 +++-------
doc/man/man3/slurm_resume.3 | 18 ++++---------
slurm/slurm.h.in | 10 +++----
src/api/suspend.c | 23 +++++++---------
src/common/slurm_protocol_defs.h | 5 ++--
src/common/slurm_protocol_pack.c | 2 --
src/scontrol/scontrol.c | 31 ++++++++++------------
src/slurmctld/job_mgr.c | 45 ++++++++++++++++++++++++++++++++
src/slurmctld/proc_req.c | 24 +++++------------
src/slurmctld/slurmctld.h | 4 +--
src/slurmctld/step_mgr.c | 45 --------------------------------
11 files changed, 92 insertions(+), 129 deletions(-)
diff --git a/doc/man/man1/scontrol.1 b/doc/man/man1/scontrol.1
index 5d74c8f405d..a9160d9bd58 100644
--- a/doc/man/man1/scontrol.1
+++ b/doc/man/man1/scontrol.1
@@ -126,11 +126,8 @@ parameters are to be changed: AuthType, BackupAddr, BackupController,
ControlAddr, ControlMach, PluginDir, StateSaveLocation, SlurmctldPort
or SlurmdPort.
.TP
-\fIresume\fP \fIID\fP
-Resume a previously suspended job or job step.
-\fIID\fP can be used to identify a specific job (e.g. "<job_id>",
-which applies to all of its existing steps)
-or a specific job step (e.g. "<job_id>.<step_id>").
+\fIresume\fP \fIjob_id\fP
+Resume a previously suspended job.
.TP
\fIshow\fP \fIENTITY\fP \fIID\fP
Display the state of the specified entity with the specified identification.
@@ -150,12 +147,9 @@ Instruct all Slurm daemons to save current state and terminate.
The Slurm controller (slurmctld) forwards the request all other daemons
(slurmd daemon on each compute node).
.TP
-\fIsuspend\fP \fIID\fP
-Suspend a running job or job step.
+\fIsuspend\fP \fIjob_id\fP
+Suspend a running job.
Use the \fIresume\fP command to resume its execution.
-\fIID\fP can be used to identify a specific job (e.g. "<job_id>",
-which applies to all of its existing steps)
-or a specific job step (e.g. "<job_id>.<step_id>").
.TP
\fIupdate\fP \fISPECIFICATION\fP
Update job, node or partition configuration per the supplied specification.
diff --git a/doc/man/man3/slurm_resume.3 b/doc/man/man3/slurm_resume.3
index ce247242b95..b6b18dc90e7 100644
--- a/doc/man/man3/slurm_resume.3
+++ b/doc/man/man3/slurm_resume.3
@@ -10,17 +10,13 @@ slurm_suspend, slurm_resume \- Slurm suspend and resume functions
.LP
int \fBslurm_suspend\fR (
.br
- uint32_t \fIjob_id\fP,
-.br
- uint32_t \fIstep_id\fP
+ uint32_t \fIjob_id\fP
.br
);
.LP
int \fBslurm_resume\fR (
.br
- uint32_t \fIjob_id\fP,
-.br
- uint32_t \fIstep_id\fP
+ uint32_t \fIjob_id\fP
.br
);
@@ -29,18 +25,14 @@ int \fBslurm_resume\fR (
.TP
\fIjob_id\fP
SLURM job ID to perform the operation upon.
-.TP
-\fIstep_id\fP
-SLURM job step ID to perform the operation upon.
-May be NO_VAL if the operation is to be performed on all steps of the specified job.
.SH "DESCRIPTION"
.LP
\fBslurm_suspend\fR
-Suspend the specified job or job step.
+Suspend the specified job.
.LP
\fBslurm_resume\fR
-Resume execution of a previously suspended job or job step.
+Resume execution of a previously suspended job.
.SH "RETURN VALUE"
.LP
@@ -48,7 +40,7 @@ Zero is returned upon success.
On error, -1 is returned, and the Slurm error code is set appropriately.
.SH "ERRORS"
.LP
-\fBESLURM_INVALID_JOB_ID\fR the requested job or job step id does not exist.
+\fBESLURM_INVALID_JOB_ID\fR the requested job id does not exist.
.LP
\fBESLURM_ACCESS_DENIED\fR the requesting user lacks authorization for the requested
action (e.g. trying to delete or modify another user's job).
diff --git a/slurm/slurm.h.in b/slurm/slurm.h.in
index de3912f874b..d796b63f6d3 100644
--- a/slurm/slurm.h.in
+++ b/slurm/slurm.h.in
@@ -1234,20 +1234,18 @@ extern int slurm_shutdown PARAMS(( uint16_t core ));
\*****************************************************************************/
/*
- * slurm_suspend - suspend execution of a job step.
+ * slurm_suspend - suspend execution of a job.
* IN job_id - job on which to perform operation
- * IN step_id - job step on which to perform operation
* RET 0 or a slurm error code
*/
-extern int slurm_suspend PARAMS(( uint32_t job_id, uint32_t step_id ));
+extern int slurm_suspend PARAMS(( uint32_t job_id ));
/*
- * slurm_resume - resume execution of a previously suspended job step.
+ * slurm_resume - resume execution of a previously suspended job.
* IN job_id - job on which to perform operation
- * IN step_id - job step on which to perform operation
* RET 0 or a slurm error code
*/
-extern int slurm_resume PARAMS(( uint32_t job_id, uint32_t step_id ));
+extern int slurm_resume PARAMS(( uint32_t job_id ));
/*****************************************************************************\
* SLURM JOB CHECKPOINT FUNCTIONS
diff --git a/src/api/suspend.c b/src/api/suspend.c
index ddf2790a9d1..a04adf756ac 100644
--- a/src/api/suspend.c
+++ b/src/api/suspend.c
@@ -32,15 +32,15 @@
#include <slurm/slurm.h>
#include "src/common/slurm_protocol_api.h"
-static int _suspend_op (uint16_t op, uint32_t job_id, uint32_t step_id);
+static int _suspend_op (uint16_t op, uint32_t job_id);
/*
- * _suspend_op - perform a suspend/resume operation for some job step.
+ * _suspend_op - perform a suspend/resume operation for some job.
* IN op - operation to perform
* IN job_id - job on which to perform operation
* IN step_id - job step on which to perform operation
* RET 0 or a slurm error code
*/
-static int _suspend_op (uint16_t op, uint32_t job_id, uint32_t step_id)
+static int _suspend_op (uint16_t op, uint32_t job_id)
{
int rc;
checkpoint_msg_t sus_req;
@@ -48,7 +48,6 @@ static int _suspend_op (uint16_t op, uint32_t job_id, uint32_t step_id)
sus_req.op = op;
sus_req.job_id = job_id;
- sus_req.step_id = step_id;
req_msg.msg_type = REQUEST_SUSPEND;
req_msg.data = &sus_req;
@@ -60,25 +59,21 @@ static int _suspend_op (uint16_t op, uint32_t job_id, uint32_t step_id)
}
/*
- * slurm_suspend - suspend execution of a job step.
+ * slurm_suspend - suspend execution of a job.
* IN job_id - job on which to perform operation
- * IN step_id - job step on which to perform operation or NO_VAL
- * for all of the job's steps
* RET 0 or a slurm error code
*/
-extern int slurm_suspend (uint32_t job_id, uint32_t step_id)
+extern int slurm_suspend (uint32_t job_id)
{
- return _suspend_op (SUSPEND_STEP, job_id, step_id);
+ return _suspend_op (SUSPEND_JOB, job_id);
}
/*
- * slurm_resume - resume execution of a previously suspended job step.
+ * slurm_resume - resume execution of a previously suspended job.
* IN job_id - job on which to perform operation
- * IN step_id - job step on which to perform operation or NO_VAL
- * for all of the job's steps
* RET 0 or a slurm error code
*/
-extern int slurm_resume (uint32_t job_id, uint32_t step_id)
+extern int slurm_resume (uint32_t job_id)
{
- return _suspend_op (RESUME_STEP, job_id, step_id);
+ return _suspend_op (RESUME_JOB, job_id);
}
diff --git a/src/common/slurm_protocol_defs.h b/src/common/slurm_protocol_defs.h
index ae9f88618bf..a10462dd1d4 100644
--- a/src/common/slurm_protocol_defs.h
+++ b/src/common/slurm_protocol_defs.h
@@ -66,8 +66,8 @@ enum part_shared {
};
enum suspend_opts {
- SUSPEND_STEP, /* Suspend a job step now */
- RESUME_STEP /* Resume a job step now */
+ SUSPEND_JOB, /* Suspend a job now */
+ RESUME_JOB /* Resume a job now */
};
/* SLURM Message types */
@@ -466,7 +466,6 @@ typedef struct checkpoint_resp_msg {
typedef struct suspend_msg {
uint16_t op; /* suspend operation, see enum suspend_opts */
uint32_t job_id; /* slurm job_id */
- uint32_t step_id; /* slurm step_id */
} suspend_msg_t;
typedef struct jobacct_msg {
diff --git a/src/common/slurm_protocol_pack.c b/src/common/slurm_protocol_pack.c
index 179fc186465..18011f0fb33 100644
--- a/src/common/slurm_protocol_pack.c
+++ b/src/common/slurm_protocol_pack.c
@@ -3226,7 +3226,6 @@ static void _pack_suspend_msg(suspend_msg_t *msg, Buf buffer)
pack16 ( msg -> op, buffer ) ;
pack32 ( msg -> job_id, buffer ) ;
- pack32 ( msg -> step_id, buffer ) ;
}
static int _unpack_suspend_msg(suspend_msg_t **msg_ptr, Buf buffer)
@@ -3239,7 +3238,6 @@ static int _unpack_suspend_msg(suspend_msg_t **msg_ptr, Buf buffer)
safe_unpack16 ( & msg -> op , buffer ) ;
safe_unpack32 ( & msg -> job_id , buffer ) ;
- safe_unpack32 ( & msg -> step_id , buffer ) ;
return SLURM_SUCCESS;
unpack_error:
diff --git a/src/scontrol/scontrol.c b/src/scontrol/scontrol.c
index e879f868122..b7f11852ada 100644
--- a/src/scontrol/scontrol.c
+++ b/src/scontrol/scontrol.c
@@ -110,7 +110,7 @@ static void _print_ping (void);
static void _print_step (char *job_step_id_str);
static void _print_version( void );
static int _process_command (int argc, char *argv[]);
-static int _suspend(char *op, char *job_step_id_str);
+static int _suspend(char *op, char *job_id_str);
static void _update_it (int argc, char *argv[]);
static int _update_job (int argc, char *argv[]);
static int _update_node (int argc, char *argv[]);
@@ -1913,8 +1913,8 @@ scontrol [<OPTION>] [<COMMAND>] \n\
show <ENTITY> [<ID>] display state of identified entity, default \n\
is all records. \n\
shutdown shutdown slurm controller. \n\
- suspend <job[.step]> susend specified job or job step \n\
- resume <job[.step]> resume previously suspended job or job step \n\
+ suspend <job_id> susend specified job \n\
+ resume <job_id> resume previously suspended job \n\
update <SPECIFICATIONS> update job, node, or partition configuration \n\
verbose enable detailed logging. \n\
version display tool version number. \n\
@@ -2030,36 +2030,33 @@ static int _checkpoint(char *op, char *job_step_id_str)
/*
* _suspend - perform some suspend/resume operation
* IN op - suspend/resume operation
- * IN job_step_id_str - either a job name (for all steps of the given job) or
- * a step name: "<jid>.<step_id>"
+ * IN job_id_str - a job id
* RET 0 if no slurm error, errno otherwise. parsing error prints
* error message and returns 0
*/
-static int _suspend(char *op, char *job_step_id_str)
+static int _suspend(char *op, char *job_id_str)
{
int rc = SLURM_SUCCESS;
- uint32_t job_id = 0, step_id = 0;
+ uint32_t job_id = 0;
char *next_str;
- if (job_step_id_str) {
- job_id = (uint32_t) strtol (job_step_id_str, &next_str, 10);
- if (next_str[0] == '.') {
- step_id = (uint32_t) strtol (&next_str[1], &next_str, 10);
- } else
- step_id = NO_VAL;
+ if (job_id_str) {
+ job_id = (uint32_t) strtol (job_id_str, &next_str, 10);
if (next_str[0] != '\0') {
- fprintf(stderr, "Invalid job step name\n");
+ fprintf(stderr, "Invalid job id specified\n");
+ exit_code = 1;
return 0;
}
} else {
- fprintf(stderr, "Invalid job step name\n");
+ fprintf(stderr, "Invalid job id specified\n");
+ exit_code = 1;
return 0;
}
if (strncasecmp(op, "suspend", 3) == 0)
- rc = slurm_suspend (job_id, step_id);
+ rc = slurm_suspend (job_id);
else
- rc = slurm_resume (job_id, step_id);
+ rc = slurm_resume (job_id);
return rc;
}
diff --git a/src/slurmctld/job_mgr.c b/src/slurmctld/job_mgr.c
index 8e2da4b6166..b7594a520a4 100644
--- a/src/slurmctld/job_mgr.c
+++ b/src/slurmctld/job_mgr.c
@@ -3869,3 +3869,48 @@ extern int job_node_ready(uint32_t job_id, int *ready)
return SLURM_SUCCESS;
}
+/*
+ * job__suspend - perform some suspend/resume operation
+ * IN sus_ptr - suspend/resume request message
+ * IN uid - user id of the user issuing the RPC
+ * IN conn_fd - file descriptor on which to send reply
+ * RET 0 on success, otherwise ESLURM error code
+ */
+extern int job_suspend(suspend_msg_t *ckpt_ptr, uid_t uid,
+ slurm_fd conn_fd)
+{
+ int rc = SLURM_SUCCESS;
+ struct job_record *job_ptr;
+ struct step_record *step_ptr;
+ slurm_msg_t resp_msg;
+ return_code_msg_t rc_msg;
+
+ /* find the job */
+ job_ptr = find_job_record (ckpt_ptr->job_id);
+ if (job_ptr == NULL) {
+ rc = ESLURM_INVALID_JOB_ID;
+ goto reply;
+ }
+ if ((uid != job_ptr->user_id) && (uid != 0)) {
+ rc = ESLURM_ACCESS_DENIED ;
+ goto reply;
+ }
+ if (job_ptr->job_state == JOB_PENDING) {
+ rc = ESLURM_JOB_PENDING;
+ goto reply;
+ } else if (job_ptr->job_state != JOB_RUNNING) {
+ rc = ESLURM_ALREADY_DONE;
+ goto reply;
+ }
+
+ /* Not fully supported yet */
+ rc = ESLURM_NOT_SUPPORTED;
+
+ reply:
+ rc_msg.return_code = rc;
+ resp_msg.msg_type = RESPONSE_SLURM_RC;
+ resp_msg.data = &rc_msg;
+ (void) slurm_send_node_msg(conn_fd, &resp_msg);
+ return rc;
+}
+
diff --git a/src/slurmctld/proc_req.c b/src/slurmctld/proc_req.c
index ea4d331dcbe..4ebde8b4ec6 100644
--- a/src/slurmctld/proc_req.c
+++ b/src/slurmctld/proc_req.c
@@ -1874,10 +1874,10 @@ inline static void _slurm_rpc_suspend(slurm_msg_t * msg)
START_TIMER;
switch (sus_ptr->op) {
- case SUSPEND_STEP:
+ case SUSPEND_JOB:
op = "suspend";
break;
- case RESUME_STEP:
+ case RESUME_JOB:
op = "resume";
break;
default:
@@ -1887,26 +1887,16 @@ inline static void _slurm_rpc_suspend(slurm_msg_t * msg)
uid = g_slurm_auth_get_uid(msg->cred);
lock_slurmctld(job_write_lock);
- error_code = job_step_suspend(sus_ptr, uid, msg->conn_fd);
+ error_code = job_suspend(sus_ptr, uid, msg->conn_fd);
unlock_slurmctld(job_write_lock);
END_TIMER;
if (error_code) {
- if (sus_ptr->step_id == NO_VAL)
- info("_slurm_rpc_suspend %s %u: %s", op,
- sus_ptr->job_id, slurm_strerror(error_code));
- else
- info("_slurm_rpc_suspend %s %u.%u %s", op,
- sus_ptr->job_id, sus_ptr->step_id,
- slurm_strerror(error_code));
+ info("_slurm_rpc_suspend %s %u: %s", op,
+ sus_ptr->job_id, slurm_strerror(error_code));
} else {
- if (sus_ptr->step_id == NO_VAL)
- info("_slurm_rpc_suspend %s for %u %s", op,
- sus_ptr->job_id, TIME_STR);
- else
- info("_slurm_rpc_suspend %s for %u.%u %s", op,
- sus_ptr->job_id, sus_ptr->step_id,
- TIME_STR);
+ info("_slurm_rpc_suspend %s for %u %s", op,
+ sus_ptr->job_id, TIME_STR);
/* NOTE: This function provides it own locks */
schedule_job_save();
}
diff --git a/src/slurmctld/slurmctld.h b/src/slurmctld/slurmctld.h
index 61260d5667a..efdb8502668 100644
--- a/src/slurmctld/slurmctld.h
+++ b/src/slurmctld/slurmctld.h
@@ -738,8 +738,8 @@ extern int job_step_checkpoint_comp(checkpoint_comp_msg_t *ckpt_ptr,
* IN conn_fd - file descriptor on which to send reply
* RET 0 on success, otherwise ESLURM error code
*/
-extern int job_step_suspend(suspend_msg_t *ckpt_ptr,
- uid_t uid, slurm_fd conn_fd);
+extern int job_suspend(suspend_msg_t *ckpt_ptr, uid_t uid,
+ slurm_fd conn_fd);
/*
* job_complete - note the normal termination the specified job
diff --git a/src/slurmctld/step_mgr.c b/src/slurmctld/step_mgr.c
index 1da9a30eeb9..ab68f2c30ff 100644
--- a/src/slurmctld/step_mgr.c
+++ b/src/slurmctld/step_mgr.c
@@ -951,48 +951,3 @@ extern int job_step_checkpoint_comp(checkpoint_comp_msg_t *ckpt_ptr,
(void) slurm_send_node_msg(conn_fd, &resp_msg);
return rc;
}
-
-/*
- * job_step_suspend - perform some suspend/resume operation
- * IN sus_ptr - suspend/resume request message
- * IN uid - user id of the user issuing the RPC
- * IN conn_fd - file descriptor on which to send reply
- * RET 0 on success, otherwise ESLURM error code
- */
-extern int job_step_suspend(suspend_msg_t *ckpt_ptr,
- uid_t uid, slurm_fd conn_fd)
-{
- int rc = SLURM_SUCCESS;
- struct job_record *job_ptr;
- struct step_record *step_ptr;
- slurm_msg_t resp_msg;
- return_code_msg_t rc_msg;
-
- /* find the job */
- job_ptr = find_job_record (ckpt_ptr->job_id);
- if (job_ptr == NULL) {
- rc = ESLURM_INVALID_JOB_ID;
- goto reply;
- }
- if ((uid != job_ptr->user_id) && (uid != 0)) {
- rc = ESLURM_ACCESS_DENIED ;
- goto reply;
- }
- if (job_ptr->job_state == JOB_PENDING) {
- rc = ESLURM_JOB_PENDING;
- goto reply;
- } else if (job_ptr->job_state != JOB_RUNNING) {
- rc = ESLURM_ALREADY_DONE;
- goto reply;
- }
-
- /* Not fully supported yet */
- rc = ESLURM_NOT_SUPPORTED;
-
- reply:
- rc_msg.return_code = rc;
- resp_msg.msg_type = RESPONSE_SLURM_RC;
- resp_msg.data = &rc_msg;
- (void) slurm_send_node_msg(conn_fd, &resp_msg);
- return rc;
-}
--
GitLab