diff --git a/NEWS b/NEWS
index 2ff7e341292630216d1e028839fbdd3523f7cb79..f65f94ed294d79b6cfbc4abb38a7dbc9626e7266 100644
--- a/NEWS
+++ b/NEWS
@@ -81,6 +81,10 @@ documents those changes that are of interest to users and administrators.
time.
-- MYSQL - Better locking around g_qos_count which was previously unprotected.
-- Correct size of buffer used for jobid2str to avoid truncation.
+ -- Fix allocation/distribution of tasks across multiple nodes when
+ --hint=nomultithread is requested.
+ -- If a reservation's nodes value is "all" then track the current nodes in the
+ system, even if those nodes change.
* Changes in Slurm 15.08.6
==========================
diff --git a/doc/html/meetings.shtml b/doc/html/meetings.shtml
index a783ae2021551812533e02c46657428e5a682d5d..eeee5e87d4b39bfd0e870d9022731d149a08a727 100644
--- a/doc/html/meetings.shtml
+++ b/doc/html/meetings.shtml
@@ -5,7 +5,7 @@
<p><b>Slurm User Group Meeting 2016</b><br>
26-27 September 2016<br>
Athens, Greece<br>
-Host: <a href="http://www.grnet.gre/">Greek Research and Technology Network (GRNET)</a></p>
+Host: <a href="http://www.grnet.gr/">Greek Research and Technology Network (GRNET)</a></p>
<p>More information coming soon.</p>
<!-- <a href="slurm_ug_cfp.html">Call for Abstracts: Due 1 June 2015</a><br> -->
<!-- <a href="slurm_ug_agenda.html">Meeting agenda, registration information, etc.</a><br> -->
@@ -44,6 +44,6 @@ Host: Bull</p>
Paris, France<br>
Host: CEA</p>
-<p style="text-align:center;">Last modified 23 December 2015</p>
+<p style="text-align:center;">Last modified 31 December 2015</p>
<!--#include virtual="footer.txt"-->
diff --git a/slurm/slurm.h.in b/slurm/slurm.h.in
index 2b105cf647bd037f7b3b6ccb8d83a4955b1b855a..a4896811300abc1c2a4c259dd2e95f67a9096b6f 100644
--- a/slurm/slurm.h.in
+++ b/slurm/slurm.h.in
@@ -2369,6 +2369,8 @@ void slurm_init_update_block_msg PARAMS((update_block_msg_t *update_block_msg));
* relative */
#define RESERVE_FLAG_REPLACE 0x00040000 /* Replace resources
* as assigned to jobs */
+#define RESERVE_FLAG_ALL_NODES 0x00080000 /* Use all compute
+ * nodes */
typedef struct reserve_info {
char *accounts; /* names of accounts permitted to use */
diff --git a/src/common/slurm_protocol_defs.c b/src/common/slurm_protocol_defs.c
index 5b5628bfc4095900be012a2c2a4b3df511a63ffd..c878fa4b6885edba9a8650806dccb03a6aaa990d 100644
--- a/src/common/slurm_protocol_defs.c
+++ b/src/common/slurm_protocol_defs.c
@@ -2063,6 +2063,11 @@ extern char *reservation_flags_string(uint32_t flags)
xstrcat(flag_str, ",");
xstrcat(flag_str, "SPEC_NODES");
}
+ if (flags & RESERVE_FLAG_ALL_NODES) {
+ if (flag_str[0])
+ xstrcat(flag_str, ",");
+ xstrcat(flag_str, "ALL_NODES");
+ }
if (flags & RESERVE_FLAG_ANY_NODES) {
if (flag_str[0])
xstrcat(flag_str, ",");
diff --git a/src/plugins/select/cons_res/dist_tasks.c b/src/plugins/select/cons_res/dist_tasks.c
index c42b3b7af1d5633cee0d5626e62766ddc9c2d966..6a42b46addf9b0096c7437e7b3d04495f2f9f84c 100644
--- a/src/plugins/select/cons_res/dist_tasks.c
+++ b/src/plugins/select/cons_res/dist_tasks.c
@@ -329,7 +329,7 @@ static void _block_sync_core_bitmap(struct job_record *job_ptr,
uint16_t cpus_per_task = job_ptr->details->cpus_per_task;
job_resources_t *job_res = job_ptr->job_resrcs;
bool alloc_cores = false, alloc_sockets = false;
- uint16_t ntasks_per_core = 0xffff;
+ uint16_t ncpus_per_core = 0xffff; /* Usable CPUs per core */
int count, cpu_min, b_min, elig, s_min, comb_idx, sock_idx;
int elig_idx, comb_brd_idx, sock_list_idx, comb_min, board_num;
int* boards_cpu_cnt;
@@ -361,12 +361,14 @@ static void _block_sync_core_bitmap(struct job_record *job_ptr,
if (job_ptr->details && job_ptr->details->mc_ptr) {
multi_core_data_t *mc_ptr = job_ptr->details->mc_ptr;
- if (mc_ptr->ntasks_per_core) {
- ntasks_per_core = mc_ptr->ntasks_per_core;
+ if ((mc_ptr->ntasks_per_core != (uint16_t) INFINITE) &&
+ (mc_ptr->ntasks_per_core)) {
+ ncpus_per_core = mc_ptr->ntasks_per_core;
+ ncpus_per_core *= cpus_per_task;
}
if ((mc_ptr->threads_per_core != (uint16_t) NO_VAL) &&
- (mc_ptr->threads_per_core < ntasks_per_core)) {
- ntasks_per_core = mc_ptr->threads_per_core;
+ (mc_ptr->threads_per_core < ncpus_per_core)) {
+ ncpus_per_core = mc_ptr->threads_per_core;
}
}
@@ -395,12 +397,7 @@ static void _block_sync_core_bitmap(struct job_record *job_ptr,
fatal("cons_res: _block_sync_core_bitmap index error");
cpus = job_res->cpus[i];
- if (ntasks_per_core == 0xffff) {
- vpus = select_node_record[n].vpus;
- } else {
- vpus = MIN(select_node_record[n].vpus,
- (ntasks_per_core * cpus_per_task));
- }
+ vpus = MIN(select_node_record[n].vpus, ncpus_per_core);
/* compute still required cores on the node */
req_cpus = cpus / vpus;
@@ -682,7 +679,8 @@ static int _cyclic_sync_core_bitmap(struct job_record *job_ptr,
bitstr_t *core_map;
bool *sock_used, *sock_avoid;
bool alloc_cores = false, alloc_sockets = false;
- uint16_t ntasks_per_core = 0xffff, ntasks_per_socket = 0xffff;
+ uint16_t ncpus_per_core = 0xffff; /* Usable CPUs per core */
+ uint16_t ntasks_per_socket = 0xffff;
int error_code = SLURM_SUCCESS;
if ((job_res == NULL) || (job_res->core_bitmap == NULL) ||
@@ -697,13 +695,15 @@ static int _cyclic_sync_core_bitmap(struct job_record *job_ptr,
core_map = job_res->core_bitmap;
if (job_ptr->details->mc_ptr) {
multi_core_data_t *mc_ptr = job_ptr->details->mc_ptr;
- if (mc_ptr->ntasks_per_core) {
- ntasks_per_core = mc_ptr->ntasks_per_core;
+ if ((mc_ptr->ntasks_per_core != (uint16_t) INFINITE) &&
+ (mc_ptr->ntasks_per_core)) {
+ ncpus_per_core = mc_ptr->ntasks_per_core;
+ ncpus_per_core *= cpus_per_task;
}
if ((mc_ptr->threads_per_core != (uint16_t) NO_VAL) &&
- (mc_ptr->threads_per_core < ntasks_per_core)) {
- ntasks_per_core = mc_ptr->threads_per_core;
+ (mc_ptr->threads_per_core < ncpus_per_core)) {
+ ncpus_per_core = mc_ptr->threads_per_core;
}
if (mc_ptr->ntasks_per_socket)
@@ -723,12 +723,7 @@ static int _cyclic_sync_core_bitmap(struct job_record *job_ptr,
continue;
sockets = select_node_record[n].sockets;
cps = select_node_record[n].cores;
- if (ntasks_per_core == 0xffff) {
- vpus = select_node_record[n].vpus;
- } else {
- vpus = MIN(select_node_record[n].vpus,
- (ntasks_per_core * cpus_per_task));
- }
+ vpus = MIN(select_node_record[n].vpus, ncpus_per_core);
if (select_debug_flags & DEBUG_FLAG_SELECT_TYPE) {
info("DEBUG: job %u node %s vpus %u cpus %u",
diff --git a/src/plugins/select/cons_res/job_test.c b/src/plugins/select/cons_res/job_test.c
index 4115c7a93b8e63be60072f7b445d41ffbbf2a1f0..918397f52aa02c2213a62a86cfad9235249ff7e2 100644
--- a/src/plugins/select/cons_res/job_test.c
+++ b/src/plugins/select/cons_res/job_test.c
@@ -214,7 +214,7 @@ static uint16_t _allocate_sc(struct job_record *job_ptr, bitstr_t *core_map,
uint16_t cores_per_socket = select_node_record[node_i].cores;
uint16_t threads_per_core = select_node_record[node_i].vpus;
uint16_t min_cores = 1, min_sockets = 1, ntasks_per_socket = 0;
- uint16_t ntasks_per_core = 0xffff;
+ uint16_t ncpus_per_core = 0xffff; /* Usable CPUs per core */
uint32_t free_cpu_count = 0, used_cpu_count = 0, *used_cpu_array = NULL;
if (job_ptr->details && job_ptr->details->mc_ptr) {
@@ -226,18 +226,20 @@ static uint16_t _allocate_sc(struct job_record *job_ptr, bitstr_t *core_map,
if (mc_ptr->sockets_per_node != (uint16_t) NO_VAL) {
min_sockets = mc_ptr->sockets_per_node;
}
- if (mc_ptr->ntasks_per_core) {
- ntasks_per_core = mc_ptr->ntasks_per_core;
+ if ((mc_ptr->ntasks_per_core != (uint16_t) INFINITE) &&
+ (mc_ptr->ntasks_per_core)) {
+ ncpus_per_core = mc_ptr->ntasks_per_core;
+ ncpus_per_core *= cpus_per_task;
}
if ((mc_ptr->threads_per_core != (uint16_t) NO_VAL) &&
- (mc_ptr->threads_per_core < ntasks_per_core)) {
- ntasks_per_core = mc_ptr->threads_per_core;
+ (mc_ptr->threads_per_core < ncpus_per_core)) {
+ ncpus_per_core = mc_ptr->threads_per_core;
}
ntasks_per_socket = mc_ptr->ntasks_per_socket;
- if ((ntasks_per_core != (uint16_t) NO_VAL) &&
- (ntasks_per_core != (uint16_t) INFINITE) &&
- (ntasks_per_core > threads_per_core)) {
+ if ((ncpus_per_core != (uint16_t) NO_VAL) &&
+ (ncpus_per_core != (uint16_t) INFINITE) &&
+ (ncpus_per_core > threads_per_core)) {
goto fini;
}
threads_per_socket = threads_per_core * cores_per_socket;
@@ -282,7 +284,7 @@ static uint16_t _allocate_sc(struct job_record *job_ptr, bitstr_t *core_map,
* Step 2: For core-level and socket-level: apply sockets_per_node
* and cores_per_socket to the "free" cores.
*
- * Step 3: Compute task-related data: ntasks_per_core,
+ * Step 3: Compute task-related data: ncpus_per_core,
* ntasks_per_socket, ntasks_per_node and cpus_per_task
* and determine the number of tasks to run on this node
*
@@ -381,15 +383,12 @@ static uint16_t _allocate_sc(struct job_record *job_ptr, bitstr_t *core_map,
* ntasks_per_socket, ntasks_per_node and cpus_per_task
* to determine the number of tasks to run on this node
*
- * Note: cpus_per_task and ntasks_per_core need to play nice
+ * Note: cpus_per_task and ncpus_per_core need to play nice
* 2 tasks_per_core vs. 2 cpus_per_task
*/
avail_cpus = 0;
num_tasks = 0;
- if (ntasks_per_core != 0xffff) {
- threads_per_core = MIN(threads_per_core,
- (ntasks_per_core * cpus_per_task));
- }
+ threads_per_core = MIN(threads_per_core, ncpus_per_core);
for (i = 0; i < sockets; i++) {
uint16_t tmp = free_cores[i] * threads_per_core;
diff --git a/src/slurmctld/reservation.c b/src/slurmctld/reservation.c
index 743ee70ef64a22930915069ee7ef277a9bd651a0..1be9c3681b027ef74c5b71198d8faba4acd781f6 100644
--- a/src/slurmctld/reservation.c
+++ b/src/slurmctld/reservation.c
@@ -2253,6 +2253,7 @@ extern int create_resv(resv_desc_msg_t *resv_desc_ptr)
resv_desc_ptr->partition);
node_bitmap = bit_copy(part_ptr->node_bitmap);
} else {
+ resv_desc_ptr->flags |= RESERVE_FLAG_ALL_NODES;
node_bitmap = bit_alloc(node_record_count);
bit_nset(node_bitmap, 0,(node_record_count-1));
}
@@ -2710,6 +2711,7 @@ extern int update_resv(resv_desc_msg_t *resv_desc_ptr)
if (resv_desc_ptr->node_list &&
(resv_desc_ptr->node_list[0] == '\0')) { /* Clear bitmap */
resv_ptr->flags &= (~RESERVE_FLAG_SPEC_NODES);
+ resv_ptr->flags &= (~RESERVE_FLAG_ALL_NODES);
xfree(resv_desc_ptr->node_list);
xfree(resv_ptr->node_list);
FREE_NULL_BITMAP(resv_ptr->node_bitmap);
@@ -2738,14 +2740,18 @@ extern int update_resv(resv_desc_msg_t *resv_desc_ptr)
xfree(resv_desc_ptr->node_list);
resv_ptr->node_list = xstrdup(part_ptr->nodes);
} else {
+ resv_ptr->flags |= RESERVE_FLAG_ALL_NODES;
node_bitmap = bit_alloc(node_record_count);
bit_nset(node_bitmap, 0,(node_record_count-1));
resv_ptr->flags &= (~RESERVE_FLAG_PART_NODES);
xfree(resv_ptr->node_list);
- resv_ptr->node_list = resv_desc_ptr->node_list;
+ xfree(resv_desc_ptr->node_list);
+ resv_ptr->node_list =
+ bitmap2node_name(node_bitmap);
}
} else {
resv_ptr->flags &= (~RESERVE_FLAG_PART_NODES);
+ resv_ptr->flags &= (~RESERVE_FLAG_ALL_NODES);
if (node_name2bitmap(resv_desc_ptr->node_list,
false, &node_bitmap)) {
info("Reservation %s request has invalid node name (%s)",
@@ -2767,6 +2773,7 @@ extern int update_resv(resv_desc_msg_t *resv_desc_ptr)
if (resv_desc_ptr->node_cnt) {
uint32_t total_node_cnt = 0;
resv_ptr->flags &= (~RESERVE_FLAG_PART_NODES);
+ resv_ptr->flags &= (~RESERVE_FLAG_ALL_NODES);
#ifdef HAVE_BG
if (!cnodes_per_mp) {
@@ -3158,6 +3165,8 @@ extern int dump_all_resv_state(void)
static bool _validate_one_reservation(slurmctld_resv_t *resv_ptr)
{
bool account_not = false, user_not = false;
+ slurmctld_resv_t old_resv_ptr;
+ bitstr_t *node_bitmap;
if ((resv_ptr->name == NULL) || (resv_ptr->name[0] == '\0')) {
error("Read reservation without name");
@@ -3221,9 +3230,7 @@ static bool _validate_one_reservation(slurmctld_resv_t *resv_ptr)
}
if ((resv_ptr->flags & RESERVE_FLAG_PART_NODES) &&
resv_ptr->part_ptr && resv_ptr->part_ptr->node_bitmap) {
- slurmctld_resv_t old_resv_ptr;
memset(&old_resv_ptr, 0, sizeof(slurmctld_resv_t));
-
xfree(resv_ptr->node_list);
resv_ptr->node_list = xstrdup(resv_ptr->part_ptr->nodes);
FREE_NULL_BITMAP(resv_ptr->node_bitmap);
@@ -3235,8 +3242,20 @@ static bool _validate_one_reservation(slurmctld_resv_t *resv_ptr)
_set_tres_cnt(resv_ptr, &old_resv_ptr);
xfree(old_resv_ptr.tres_str);
last_resv_update = time(NULL);
+ } else if (resv_ptr->flags & RESERVE_FLAG_ALL_NODES) {
+ memset(&old_resv_ptr, 0, sizeof(slurmctld_resv_t));
+ FREE_NULL_BITMAP(resv_ptr->node_bitmap);
+ resv_ptr->node_bitmap = bit_alloc(node_record_count);
+ bit_nset(resv_ptr->node_bitmap, 0, (node_record_count - 1));
+ xfree(resv_ptr->node_list);
+ resv_ptr->node_list = bitmap2node_name(resv_ptr->node_bitmap);
+ resv_ptr->node_cnt = bit_set_count(resv_ptr->node_bitmap);
+ old_resv_ptr.tres_str = resv_ptr->tres_str;
+ resv_ptr->tres_str = NULL;
+ _set_tres_cnt(resv_ptr, &old_resv_ptr);
+ xfree(old_resv_ptr.tres_str);
+ last_resv_update = time(NULL);
} else if (resv_ptr->node_list) { /* Change bitmap last */
- bitstr_t *node_bitmap;
#ifdef HAVE_BG
int inx;
char save = '\0';