diff --git a/NEWS b/NEWS
index a120c1fdea265fa166bf52f3ffcb63231c1d26c2..0671c1bb8c71187e39fb89846b5af9722a6c6861 100644
--- a/NEWS
+++ b/NEWS
@@ -188,6 +188,7 @@ documents those changes that are of interest to users and admins.
* Changes in SLURM 1.1.24
=========================
- In sched/wiki2, add support for direct "srun --dependency=" use.
+ - mpi/mvapich: Add support for MVAPICH protocol version 6.
* Changes in SLURM 1.1.23
=========================
diff --git a/src/plugins/mpi/mvapich/mvapich.c b/src/plugins/mpi/mvapich/mvapich.c
index cbdf7a21c8190cbe60615d8fe2e2cacc02898c93..5891018d44604941624a5b657666f84117a26c3a 100644
--- a/src/plugins/mpi/mvapich/mvapich.c
+++ b/src/plugins/mpi/mvapich/mvapich.c
@@ -126,7 +126,7 @@ static struct mvapich_info **mvarray = NULL;
static int mvapich_fd = -1;
static int nprocs = -1;
static int protocol_version = -1;
-static int v5_phase = 0;
+static int protocol_phase = 0;
static int connect_once = 1;
static int mvapich_verbose = 0;
static int do_timing = 0;
@@ -164,11 +164,20 @@ static void mvapich_info_destroy (struct mvapich_info *mvi)
static int mvapich_requires_pids (void)
{
if ( protocol_version == MVAPICH_VERSION_REQUIRES_PIDS
- || protocol_version == 5)
+ || protocol_version == 5
+ || protocol_version == 6 )
return (1);
return (0);
}
+/*
+ * Return non-zero if protocol version has two phases.
+ */
+static int mvapich_dual_phase (void)
+{
+ return (protocol_version == 5 || protocol_version == 6);
+}
+
static int mvapich_abort_sends_rank (void)
{
if (protocol_version >= 3)
@@ -230,9 +239,9 @@ static int mvapich_get_hostid (struct mvapich_info *mvi)
static int mvapich_get_task_header (int fd, int *version, int *rank)
{
/*
- * V5 only sends version on first pass
+ * dual phase only sends version on first pass
*/
- if (protocol_version != 5 || v5_phase == 0) {
+ if (!mvapich_dual_phase () || protocol_phase == 0) {
if (fd_read_n (fd, version, sizeof (int)) < 0)
return error ("mvapich: Unable to read version from task: %m");
}
@@ -240,7 +249,7 @@ static int mvapich_get_task_header (int fd, int *version, int *rank)
if (fd_read_n (fd, rank, sizeof (int)) < 0)
return error ("mvapich: Unable to read task rank: %m");
- if (protocol_version == 5 && v5_phase > 0)
+ if (mvapich_dual_phase () && protocol_phase > 0)
return (0);
if (protocol_version == -1)
@@ -264,7 +273,8 @@ static int mvapich_handle_task (int fd, struct mvapich_info *mvi)
case 3:
return mvapich_get_task_info (mvi);
case 5:
- if (v5_phase == 0)
+ case 6:
+ if (protocol_phase == 0)
return mvapich_get_hostid (mvi);
else
return mvapich_get_task_info (mvi);
@@ -367,7 +377,7 @@ static void mvapich_bcast_hostids (void)
static void mvapich_bcast (void)
{
- if (protocol_version < 5 || v5_phase > 0)
+ if (!mvapich_dual_phase () || protocol_phase > 0)
return mvapich_bcast_addrs ();
else
return mvapich_bcast_hostids ();
@@ -405,7 +415,8 @@ static void mvapich_barrier (void)
return;
}
-static void mvapich_print_abort_message (slurm_step_layout_t *sl, int rank)
+static void
+mvapich_print_abort_message (slurm_step_layout_t *sl, int rank, int dest)
{
char *host;
@@ -417,8 +428,15 @@ static void mvapich_print_abort_message (slurm_step_layout_t *sl, int rank)
host = slurm_step_layout_host_name(
sl, slurm_step_layout_host_id(sl, rank));
- info ("mvapich: Received ABORT message from MPI rank %d [on %s]",
- rank, host);
+ if (dest >= 0) {
+ info ("mvapich: %M: ABORT from MPI rank %d [on %s] dest rank %d [on %s]",
+ rank, host, dest,
+ slurm_step_layout_host_name (sl, dest));
+ }
+ else {
+ info ("mvapich: %M: ABORT from MPI rank %d [on %s]",
+ rank, host);
+ }
return;
}
@@ -427,6 +445,7 @@ static void mvapich_wait_for_abort(srun_job_t *job)
{
int rlen;
char rbuf[1024];
+ int *p = (int *) rbuf;
/*
* Wait for abort notification from any process.
@@ -452,7 +471,10 @@ static void mvapich_wait_for_abort(srun_job_t *job)
}
close(newfd);
- mvapich_print_abort_message(job->step_layout, *((int *) rbuf));
+ if (rlen > sizeof (int))
+ mvapich_print_abort_message (job->step_layout, p[1], p[0]);
+ else
+ mvapich_print_abort_message (job->step_layout, p[0], -1);
fwd_signal(job, SIGKILL, opt.max_threads);
}
@@ -489,10 +511,12 @@ static int mvapich_handle_connection (int fd)
{
int version, rank;
- if (v5_phase == 0 || !connect_once) {
+ if (protocol_phase == 0 || !connect_once) {
if (mvapich_get_task_header (fd, &version, &rank) < 0)
return (-1);
+ mvarray [rank]->rank = rank;
+
if (rank > nprocs - 1)
return (error ("mvapich: task reported invalid rank (%d)", rank));
} else {
@@ -542,7 +566,7 @@ static int mvapich_get_next_connection (int listenfd)
slurm_addr addr;
int fd;
- if (connect_once && v5_phase > 0) {
+ if (connect_once && protocol_phase > 0) {
return (poll_mvapich_fds ());
}
@@ -620,8 +644,8 @@ again:
mvapich_debug ("bcasting mvapich info to %d tasks", nprocs);
mvapich_bcast ();
- if (protocol_version == 5 && v5_phase == 0) {
- v5_phase = 1;
+ if (mvapich_dual_phase () && protocol_phase == 0) {
+ protocol_phase = 1;
goto again;
}
diff --git a/src/plugins/select/bluegene/plugin/bg_job_place.c b/src/plugins/select/bluegene/plugin/bg_job_place.c
index 0498c1a266501c2dddb0856596b1d61b1b792d69..a18322bfdb3f03ed53d50e793cd913f1fba2baf7 100644
--- a/src/plugins/select/bluegene/plugin/bg_job_place.c
+++ b/src/plugins/select/bluegene/plugin/bg_job_place.c
@@ -111,7 +111,6 @@ static int _find_best_block_match(struct job_record* job_ptr,
int i;
int rot_cnt = 0;
int created = 0;
- int found = 0;
int allow = 0;
int check_image = 1;
uint32_t max_procs = NO_VAL;
@@ -449,15 +448,16 @@ try_again:
debug3("%s job_running = %d",
record->bg_block_id, record->job_running);
/*block is messed up some how (BLOCK_ERROR_STATE) ignore it*/
- if(record->job_running == BLOCK_ERROR_STATE)
+ if(record->job_running == BLOCK_ERROR_STATE) {
+ debug("block %s is in an error state (can't use)",
+ record->bg_block_id);
continue;
- else if((record->job_running != NO_JOB_RUNNING)
- && !test_only) {
+ } else if((record->job_running != NO_JOB_RUNNING)
+ && !test_only) {
debug("block %s in use by %s job %d",
record->bg_block_id,
record->user_name,
record->job_running);
- found = 1;
continue;
}
@@ -574,8 +574,8 @@ try_again:
}
}
list_iterator_destroy(itr2);
+
if(found_record) {
- found = 1;
continue;
}
@@ -641,7 +641,6 @@ try_again:
continue; /* Not usable */
}
*found_bg_record = record;
- found = 1;
debug2("we found one! %s", (*found_bg_record)->bg_block_id);
break;
}
diff --git a/src/srun/opt.c b/src/srun/opt.c
index 4d15b05d4c2c562798487e8222c56af38601163f..93baa9bad37ca52a6e3f5c2bbf2f19d05d886c04 100644
--- a/src/srun/opt.c
+++ b/src/srun/opt.c
@@ -1868,7 +1868,7 @@ void set_options(const int argc, char **argv, int first)
opt.begin = parse_time(optarg);
break;
case LONG_OPT_MAIL_TYPE:
- opt.mail_type = _parse_mail_type(optarg);
+ opt.mail_type |= _parse_mail_type(optarg);
if (opt.mail_type == 0)
fatal("--mail-type=%s invalid", optarg);
break;
@@ -2375,6 +2375,7 @@ static char *_print_mail_type(const uint16_t type)
{
if (type == 0)
return "NONE";
+
if (type == MAIL_JOB_BEGIN)
return "BEGIN";
if (type == MAIL_JOB_END)
@@ -2384,7 +2385,7 @@ static char *_print_mail_type(const uint16_t type)
if (type == (MAIL_JOB_BEGIN | MAIL_JOB_END | MAIL_JOB_FAIL))
return "ALL";
- return "UNKNOWN";
+ return "MULTIPLE";
}
static void