diff --git a/NEWS b/NEWS
index dc52c856fe8201e55cfa50905eedca17a66d4aff..b655765b693c62035565670dde91675a302a2e51 100644
--- a/NEWS
+++ b/NEWS
@@ -1,6 +1,9 @@
This file describes changes in recent versions of SLURM. It primarily
documents those changes that are of interest to users and admins.
+* Changes in SLURM 0.6.0-pre2
+=============================
+
* Changes in SLURM 0.6.0-pre1
=============================
-- Added bgl/partition_allocator/smap changes from 0.5.7.
diff --git a/src/plugins/proctrack/aix/proctrack_aix.c b/src/plugins/proctrack/aix/proctrack_aix.c
index 75f7060c041fb65681d76fdffefc0b125b658037..fa3777e66ab9ec73141f999e7890b3c7652a467c 100644
--- a/src/plugins/proctrack/aix/proctrack_aix.c
+++ b/src/plugins/proctrack/aix/proctrack_aix.c
@@ -110,7 +110,7 @@ extern int fini ( void )
* Uses job step process group id as a unique identifier. Job id
* and step id are not unique by themselves.
*/
-extern uint32_t slurm_create_container ( slurmd_job_t *job )
+extern uint32_t slurm_container_create ( slurmd_job_t *job )
{
int pgid = (int) job->pgid;
int i;
@@ -128,13 +128,13 @@ extern uint32_t slurm_create_container ( slurmd_job_t *job )
}
-extern int slurm_add_container ( uint32_t id )
+extern int slurm_container_add ( uint32_t id, pid_t pid )
{
- error("slurm_add_container not supported");
- return SLURM_ERROR;
+ debug("slurm_container_add not supported");
+ return SLURM_SUCCESS;
}
-extern int slurm_signal_container ( uint32_t id, int signal )
+extern int slurm_container_signal ( uint32_t id, int signal )
{
int jobid = (int) id;
if (!id) /* no container ID */
@@ -143,7 +143,7 @@ extern int slurm_signal_container ( uint32_t id, int signal )
return proctrack_job_kill(&jobid, &signal);
}
-extern int slurm_destroy_container ( uint32_t id )
+extern int slurm_container_destroy ( uint32_t id )
{
int jobid = (int) id;
@@ -158,7 +158,7 @@ extern int slurm_destroy_container ( uint32_t id )
}
extern uint32_t
-slurm_find_container(pid_t pid)
+slurm_container_find(pid_t pid)
{
int local_pid = (int) pid;
int cont_id = proctrack_get_job_id(&local_pid);
diff --git a/src/plugins/proctrack/linuxproc/proctrack_linuxproc.c b/src/plugins/proctrack/linuxproc/proctrack_linuxproc.c
index c4301bcca8c2c175895a897c0dae8b8d5ee2d7ed..f737f86bc0d1337e066c1d84a1334ad75bd10b73 100644
--- a/src/plugins/proctrack/linuxproc/proctrack_linuxproc.c
+++ b/src/plugins/proctrack/linuxproc/proctrack_linuxproc.c
@@ -92,27 +92,27 @@ extern int fini ( void )
/*
* Uses slurmd job-step manager's pid as the unique container id.
*/
-extern uint32_t slurm_create_container ( slurmd_job_t *job )
+extern uint32_t slurm_container_create ( slurmd_job_t *job )
{
return (uint32_t) job->jmgr_pid;
}
-extern int slurm_add_container ( uint32_t id )
+extern int slurm_container_add ( uint32_t id, pid_t pid )
{
return SLURM_SUCCESS;
}
-extern int slurm_signal_container ( uint32_t id, int signal )
+extern int slurm_container_signal ( uint32_t id, int signal )
{
return kill_proc_tree_not_top((pid_t)id, signal);
}
-extern int slurm_destroy_container ( uint32_t id )
+extern int slurm_container_destroy ( uint32_t id )
{
return SLURM_SUCCESS;
}
-extern uint32_t slurm_find_container(pid_t pid)
+extern uint32_t slurm_container_find(pid_t pid)
{
return (uint32_t) find_ancestor(pid, "slurmd");
}
diff --git a/src/plugins/proctrack/pgid/proctrack_pgid.c b/src/plugins/proctrack/pgid/proctrack_pgid.c
index bab31b769d5c47f5801ba400bdd1eb4cf70f0851..e0c2f73fb982143f445c1dd36ca8c1c99ca9c5bc 100644
--- a/src/plugins/proctrack/pgid/proctrack_pgid.c
+++ b/src/plugins/proctrack/pgid/proctrack_pgid.c
@@ -94,17 +94,17 @@ extern int fini ( void )
/*
* Uses job step process group id.
*/
-extern uint32_t slurm_create_container ( slurmd_job_t *job )
+extern uint32_t slurm_container_create ( slurmd_job_t *job )
{
return (uint32_t) job->pgid;
}
-extern int slurm_add_container ( uint32_t id )
+extern int slurm_container_add ( uint32_t id, pid_t pid )
{
return SLURM_SUCCESS;
}
-extern int slurm_signal_container ( uint32_t id, int signal )
+extern int slurm_container_signal ( uint32_t id, int signal )
{
pid_t pid = (pid_t) id;
@@ -119,13 +119,13 @@ extern int slurm_signal_container ( uint32_t id, int signal )
return (int)killpg(pid, signal);
}
-extern int slurm_destroy_container ( uint32_t id )
+extern int slurm_container_destroy ( uint32_t id )
{
return SLURM_SUCCESS;
}
extern uint32_t
-slurm_find_container(pid_t pid)
+slurm_container_find(pid_t pid)
{
pid_t rc = getpgid(pid);
diff --git a/src/slurmd/mgr.c b/src/slurmd/mgr.c
index 9a8cf9d73a1384346622eba4f1455321364c76d6..e3cab987fce04d41b91baa3888798f18116085ff 100644
--- a/src/slurmd/mgr.c
+++ b/src/slurmd/mgr.c
@@ -696,9 +696,9 @@ _fork_all_tasks(slurmd_job_t *job)
* will wait for our signal before calling exec.
*/
shm_update_step_pgid(job->jobid, job->stepid, job->pgid);
- cont_id = slurm_create_container(job);
+ cont_id = slurm_container_create(job);
if (cont_id == 0) {
- error("slurm_create_container: %m");
+ error("slurm_container_create: %m");
exit(3);
}
shm_update_step_cont_id(job->jobid, job->stepid, cont_id);
@@ -876,12 +876,12 @@ _kill_running_tasks(slurmd_job_t *job)
return;
if (s->cont_id) {
- slurm_signal_container(s->cont_id, SIGKILL);
+ slurm_container_signal(s->cont_id, SIGKILL);
/* Try destroying the container up to 30 times */
- while (slurm_destroy_container(s->cont_id) != SLURM_SUCCESS
+ while (slurm_container_destroy(s->cont_id) != SLURM_SUCCESS
&& limit < 30) {
- slurm_signal_container(s->cont_id, SIGKILL);
+ slurm_container_signal(s->cont_id, SIGKILL);
sleep(1);
limit++;
}
diff --git a/src/slurmd/proctrack.c b/src/slurmd/proctrack.c
index e3615ed814742930d59049326b4fb750e8b088c8..a7daa31efd625e0194f15220920dc3f910beeafd 100644
--- a/src/slurmd/proctrack.c
+++ b/src/slurmd/proctrack.c
@@ -38,7 +38,7 @@
/* ************************************************************************ */
typedef struct slurm_proctrack_ops {
uint32_t (*create) ( slurmd_job_t *job );
- int (*add) ( uint32_t id );
+ int (*add) ( uint32_t id, pid_t pid );
int (*signal) ( uint32_t id, int signal );
int (*destroy) ( uint32_t id );
uint32_t (*find_cont) ( pid_t pid );
@@ -70,11 +70,11 @@ _proctrack_get_ops( slurm_proctrack_context_t *c )
* Must be synchronized with slurm_proctrack_ops_t above.
*/
static const char *syms[] = {
- "slurm_create_container",
- "slurm_add_container",
- "slurm_signal_container",
- "slurm_destroy_container",
- "slurm_find_container"
+ "slurm_container_create",
+ "slurm_container_add",
+ "slurm_container_signal",
+ "slurm_container_destroy",
+ "slurm_container_find"
};
int n_syms = sizeof( syms ) / sizeof( char * );
@@ -223,7 +223,7 @@ slurm_proctrack_fini( void )
* Returns container ID or zero on error
*/
extern uint32_t
-slurm_create_container(slurmd_job_t *job)
+slurm_container_create(slurmd_job_t *job)
{
if ( slurm_proctrack_init() < 0 )
return 0;
@@ -232,30 +232,31 @@ slurm_create_container(slurmd_job_t *job)
}
/*
- * Add this process to the specified container
- * cont_id IN - container ID as returned by slurm_create_container()
+ * Add a process to the specified container
+ * cont_id IN - container ID as returned by slurm_container_create()
+ * pid IN - process ID to be added to the container
*
* Returns a SLURM errno.
*/
extern int
-slurm_add_container(uint32_t cont_id)
+slurm_container_add(uint32_t cont_id, pid_t pid)
{
if ( slurm_proctrack_init() < 0 )
return SLURM_ERROR;
- return (*(g_proctrack_context->ops.add))( cont_id );
+ return (*(g_proctrack_context->ops.add))( cont_id , pid );
}
/*
* Signal all processes within a container
- * cont_id IN - container ID as returned by slurm_create_container()
+ * cont_id IN - container ID as returned by slurm_container_create()
* signal IN - signal to send, if zero then perform error checking
* but do not send signal
*
* Returns a SLURM errno.
*/
extern int
-slurm_signal_container(uint32_t cont_id, int signal)
+slurm_container_signal(uint32_t cont_id, int signal)
{
if ( slurm_proctrack_init() < 0 )
return SLURM_ERROR;
@@ -265,12 +266,12 @@ slurm_signal_container(uint32_t cont_id, int signal)
/*
* Destroy a container, any processes within the container are not effected
- * cont_id IN - container ID as returned by slurm_create_container()
+ * cont_id IN - container ID as returned by slurm_container_create()
*
* Returns a SLURM errno.
*/
extern int
-slurm_destroy_container(uint32_t cont_id)
+slurm_container_destroy(uint32_t cont_id)
{
if ( slurm_proctrack_init() < 0 )
return SLURM_ERROR;
@@ -284,7 +285,7 @@ slurm_destroy_container(uint32_t cont_id)
* Returns a SLURM errno.
*/
extern uint32_t
-slurm_find_container(pid_t pid)
+slurm_container_find(pid_t pid)
{
if ( slurm_proctrack_init() < 0 )
return SLURM_ERROR;
diff --git a/src/slurmd/proctrack.h b/src/slurmd/proctrack.h
index 6d04643ed62239886f5c393f62ea8ceeced6f7a8..f104cf067d0842ff38d613db82680fb939967aaa 100644
--- a/src/slurmd/proctrack.h
+++ b/src/slurmd/proctrack.h
@@ -55,41 +55,42 @@ extern int slurm_proctrack_fini(void);
*
* Returns container ID or zero on error
*/
-extern uint32_t slurm_create_container(slurmd_job_t *job);
+extern uint32_t slurm_container_create(slurmd_job_t *job);
/*
- * Add this process to the specified container
- * cont_id IN - container ID as returned by slurm_create_container()
+ * Add a process to the specified container
+ * cont_id IN - container ID as returned by slurm_container_create()
+ * pid IN - process ID to be added to the container
*
* Returns a SLURM errno.
*/
-extern int slurm_add_container(uint32_t cont_id);
+extern int slurm_container_add(uint32_t cont_id, pid_t pid);
/*
* Signal all processes within a container
- * cont_id IN - container ID as returned by slurm_create_container()
+ * cont_id IN - container ID as returned by slurm_container_create()
* signal IN - signal to send, if zero then perform error checking
* but do not send signal
*
* Returns a SLURM errno.
*/
-extern int slurm_signal_container(uint32_t cont_id, int signal);
+extern int slurm_container_signal(uint32_t cont_id, int signal);
/*
* Destroy a container, any processes within the container are not effected
- * cont_id IN - container ID as returned by slurm_create_container()
+ * cont_id IN - container ID as returned by slurm_container_create()
*
* Returns a SLURM errno.
*/
-extern int slurm_destroy_container(uint32_t cont_id);
+extern int slurm_container_destroy(uint32_t cont_id);
/*
* Get container ID for give process ID
*
* Returns a SLURM errno.
*/
-extern uint32_t slurm_find_container(pid_t pid);
+extern uint32_t slurm_container_find(pid_t pid);
/* Wait for all processes within a container to exit */
/* Add process to a container */
diff --git a/src/slurmd/req.c b/src/slurmd/req.c
index dcd4546ef8ae1c120197da60300cd3d94bf7e6d8..1cfce6be5943043bc575e2ce768372fc45ba16fc 100644
--- a/src/slurmd/req.c
+++ b/src/slurmd/req.c
@@ -760,11 +760,11 @@ _rpc_kill_tasks(slurm_msg_t *msg, slurm_addr *cli_addr)
* Assume step termination request.
* Send SIGCONT just in case the processes are stopped.
*/
- slurm_signal_container(step->cont_id, SIGCONT);
- if (slurm_signal_container(step->cont_id, req->signal) < 0)
+ slurm_container_signal(step->cont_id, SIGCONT);
+ if (slurm_container_signal(step->cont_id, req->signal) < 0)
rc = errno;
} else if (req->signal == 0) {
- if (slurm_signal_container(step->cont_id, req->signal) < 0)
+ if (slurm_container_signal(step->cont_id, req->signal) < 0)
rc = errno;
/* SIGMIGRATE and SIGSOUND are used to initiate job checkpoint on AIX.
* These signals are not sent to the entire process group, but just a
@@ -839,10 +839,10 @@ static void _rpc_pid2jid(slurm_msg_t *msg, slurm_addr *cli)
slurm_msg_t resp_msg;
job_id_response_msg_t resp;
bool found = false;
- uint32_t my_cont = slurm_find_container(req->job_pid);
+ uint32_t my_cont = slurm_container_find(req->job_pid);
if (my_cont == 0) {
- verbose("slurm_find_container(%u): process not found",
+ verbose("slurm_container_find(%u): process not found",
(uint32_t) req->job_pid);
} else {
List steps = shm_get_steps();
@@ -1012,7 +1012,7 @@ _kill_all_active_steps(uint32_t jobid, int sig, bool batch)
debug2("signal %d to job %u (cont_id:%u)",
sig, jobid, s->cont_id);
- if (slurm_signal_container(s->cont_id, sig) < 0)
+ if (slurm_container_signal(s->cont_id, sig) < 0)
error("kill jid %d cont_id %u: %m",
s->jobid, s->cont_id);
}
diff --git a/src/slurmd/shm.c b/src/slurmd/shm.c
index 7e47c9c59d568f9af5b355fa5042b4ccbef9f6f8..fbf0f3086621a93f4df86cbbc21d2bde22bfc3b7 100644
--- a/src/slurmd/shm.c
+++ b/src/slurmd/shm.c
@@ -880,7 +880,7 @@ _shm_clear_stale_entries(void)
if ((s->state == SLURMD_JOB_UNUSED) /* unused */
|| (s->cont_id == 0) /* empty */
- || (slurm_signal_container(s->cont_id, 0) == 0)) /* active */
+ || (slurm_container_signal(s->cont_id, 0) == 0)) /* active */
continue;
while (t && !active_tasks) {
@@ -1135,7 +1135,7 @@ static bool
_valid_slurmd_cont_id(uint32_t cont_id)
{
/* Check if container has processes */
- if (slurm_signal_container(cont_id, 0) != 0)
+ if (slurm_container_signal(cont_id, 0) != 0)
return false;
return true;