diff --git a/src/common/slurm_protocol_defs.c b/src/common/slurm_protocol_defs.c
index c9a1abc19348c50b954da679a27469d30ac6c257..e5f6bac7a7032a39587be90a25decf524b30eca3 100644
--- a/src/common/slurm_protocol_defs.c
+++ b/src/common/slurm_protocol_defs.c
@@ -1597,30 +1597,53 @@ extern char* node_use_string(enum node_use_type node_use)
extern char *bg_block_state_string(uint16_t state)
{
- static char tmp[16];
+ static char tmp[25];
+ char *state_str = NULL;
+ char *err_str = NULL;
+ if (state & BG_BLOCK_ERROR_FLAG) {
+ err_str = "Error";
+ state &= (~BG_BLOCK_ERROR_FLAG);
+ }
switch (state) {
case BG_BLOCK_NAV:
- return "NAV";
+ if (!err_str)
+ state_str = "NAV";
+ else {
+ err_str = NULL;
+ state_str = "Error";
+ }
+ break;
case BG_BLOCK_FREE:
- return "Free";
+ state_str = "Free";
+ break;
case BG_BLOCK_BUSY:
- return "Busy";
+ state_str = "Busy";
+ break;
case BG_BLOCK_BOOTING:
- return "Boot";
+ state_str = "Boot";
+ break;
case BG_BLOCK_REBOOTING:
- return "Reboot";
+ state_str = "Reboot";
+ break;
case BG_BLOCK_INITED:
- return "Ready";
+ state_str = "Ready";
+ break;
case BG_BLOCK_ALLOCATED:
- return "Alloc";
+ state_str = "Alloc";
+ break;
case BG_BLOCK_TERM:
- return "Term";
- case BG_BLOCK_ERROR:
- return "Error";
+ state_str = "Term";
+ break;
+ default:
+ state_str = "Unknown";
+ break;
}
- snprintf(tmp, sizeof(tmp), "%d", state);
+ if (err_str)
+ snprintf(tmp, sizeof(tmp), "%s(%s)", err_str, state_str);
+ else
+ return state_str;
return tmp;
}
diff --git a/src/plugins/select/bluegene/bg_core.c b/src/plugins/select/bluegene/bg_core.c
index 302f3fe1997bb97c0471c29b5220fcf2ff592733..fba317a8438458b5bb9d3fa91c5c732079152f23 100644
--- a/src/plugins/select/bluegene/bg_core.c
+++ b/src/plugins/select/bluegene/bg_core.c
@@ -95,8 +95,8 @@ static int _post_block_free(bg_record_t *bg_record, bool restore)
return SLURM_SUCCESS;
}
- if ((bg_record->state != BG_BLOCK_FREE)
- && (bg_record->state != BG_BLOCK_ERROR)) {
+ if (!(bg_record->state & BG_BLOCK_ERROR_FLAG)
+ && (bg_record->state != BG_BLOCK_FREE)) {
/* Something isn't right, go mark this one in an error
state. */
update_block_msg_t block_msg;
@@ -107,7 +107,7 @@ static int _post_block_free(bg_record_t *bg_record, bool restore)
bg_block_state_string(bg_record->state));
slurm_init_update_block_msg(&block_msg);
block_msg.bg_block_id = bg_record->bg_block_id;
- block_msg.state = BG_BLOCK_ERROR;
+ block_msg.state |= BG_BLOCK_ERROR_FLAG;
block_msg.reason = "Block would not deallocate";
slurm_mutex_unlock(&block_state_mutex);
select_g_update_block(&block_msg);
@@ -206,12 +206,12 @@ static void *_track_freeing_blocks(void *args)
/* Fake a free since we are n deallocating
state before this.
*/
- if ((bg_record->state != BG_BLOCK_ERROR)
+ if (!(bg_record->state & BG_BLOCK_ERROR_FLAG)
&& (retry_cnt >= 3))
bg_record->state = BG_BLOCK_FREE;
#endif
if ((bg_record->state == BG_BLOCK_FREE)
- || (bg_record->state == BG_BLOCK_ERROR))
+ || (bg_record->state & BG_BLOCK_ERROR_FLAG))
free_cnt++;
else if (bg_record->state != BG_BLOCK_TERM)
bg_free_block(bg_record, 0, 1);
@@ -371,7 +371,8 @@ extern int bg_free_block(bg_record_t *bg_record, bool wait, bool locked)
we set it ourselves so
break out.
*/
- if (bg_record->state == BG_BLOCK_ERROR)
+ if (bg_record->state
+ & BG_BLOCK_ERROR_FLAG)
break;
#endif
if (bg_conf->slurm_debug_flags
@@ -400,7 +401,7 @@ extern int bg_free_block(bg_record_t *bg_record, bool wait, bool locked)
/* Fake a free since we are n deallocating
state before this.
*/
- if (bg_record->state == BG_BLOCK_ERROR)
+ if (bg_record->state & BG_BLOCK_ERROR_FLAG)
break;
else if (count >= 3)
bg_record->state = BG_BLOCK_FREE;
@@ -410,7 +411,7 @@ extern int bg_free_block(bg_record_t *bg_record, bool wait, bool locked)
if (!wait || (bg_record->state == BG_BLOCK_FREE)
#ifdef HAVE_BGL
- || (bg_record->state == BG_BLOCK_ERROR)
+ || (bg_record->state & BG_BLOCK_ERROR_FLAG)
#endif
) {
break;
@@ -427,7 +428,7 @@ extern int bg_free_block(bg_record_t *bg_record, bool wait, bool locked)
rc = SLURM_SUCCESS;
if ((bg_record->state == BG_BLOCK_FREE)
- || (bg_record->state == BG_BLOCK_ERROR))
+ || (bg_record->state & BG_BLOCK_ERROR_FLAG))
remove_from_bg_list(bg_lists->booted, bg_record);
else if (count >= MAX_FREE_RETRIES) {
/* Something isn't right, go mark this one in an error
@@ -440,7 +441,7 @@ extern int bg_free_block(bg_record_t *bg_record, bool wait, bool locked)
bg_block_state_string(bg_record->state));
slurm_init_update_block_msg(&block_msg);
block_msg.bg_block_id = bg_record->bg_block_id;
- block_msg.state = BG_BLOCK_ERROR;
+ block_msg.state |= BG_BLOCK_ERROR_FLAG;
block_msg.reason = "Block would not deallocate";
slurm_mutex_unlock(&block_state_mutex);
select_g_update_block(&block_msg);
@@ -645,7 +646,7 @@ extern int load_state_file(List curr_block_list, char *dir_name)
/* we only care about the states we need here
* everthing else should have been set up already */
- if (block_info->state == BG_BLOCK_ERROR) {
+ if (block_info->state & BG_BLOCK_ERROR_FLAG) {
slurm_mutex_lock(&block_state_mutex);
if ((bg_record = find_bg_record_in_list(
curr_block_list,
diff --git a/src/plugins/select/bluegene/bg_enums.h b/src/plugins/select/bluegene/bg_enums.h
index 6cb7cb7a89cd1b71d45d409dfb6b4058bebd0f76..330032b643328ae29391b645fc0e6161973d2880 100644
--- a/src/plugins/select/bluegene/bg_enums.h
+++ b/src/plugins/select/bluegene/bg_enums.h
@@ -98,7 +98,6 @@ typedef enum {
BG_BLOCK_INITED, // Block is initialized
BG_BLOCK_REBOOTING, // Block is rebooting
BG_BLOCK_TERM, // Block is terminating
- BG_BLOCK_ERROR, // Block is in error
BG_BLOCK_NAV, // Block state is undefined
} bg_block_status_t;
@@ -112,6 +111,9 @@ typedef enum {
BG_JOB_ERROR //!< Job is in error status.
} bg_job_status_t;
+#define BG_BLOCK_ERROR_FLAG 0x1000 // Block is in error
+
+
#define BG_SWITCH_NONE 0x0000
#define BG_SWITCH_OUT 0x0001
#define BG_SWITCH_IN 0x0002
diff --git a/src/plugins/select/bluegene/bg_job_place.c b/src/plugins/select/bluegene/bg_job_place.c
index b06737879672efef2a926c472918a8b792f9ce80..ad80e635a44b299734a36d136d61ece9f918bb71 100644
--- a/src/plugins/select/bluegene/bg_job_place.c
+++ b/src/plugins/select/bluegene/bg_job_place.c
@@ -307,10 +307,10 @@ static bg_record_t *_find_matching_block(List block_list,
if (bg_record->job_ptr)
bg_record->job_running = bg_record->job_ptr->job_id;
- /*block is messed up some how (BLOCK_ERROR_STATE)
+ /*block is messed up some how (BLOCK_ERROR_STATE_FLAG)
* ignore it or if state == BG_BLOCK_ERROR */
if ((bg_record->job_running == BLOCK_ERROR_STATE)
- || (bg_record->state == BG_BLOCK_ERROR)) {
+ || (bg_record->state & BG_BLOCK_ERROR_FLAG)) {
if (bg_conf->slurm_debug_flags & DEBUG_FLAG_BG_PICK)
info("block %s is in an error "
"state (can't use)",
@@ -597,10 +597,12 @@ static int _check_for_booted_overlapping_blocks(
|| SELECT_IS_MODE_RUN_NOW(query_mode))
&& (bg_conf->layout_mode != LAYOUT_DYNAMIC)))
&& ((found_record->job_running != NO_JOB_RUNNING)
- || (found_record->state == BG_BLOCK_ERROR))) {
+ || (found_record->state
+ & BG_BLOCK_ERROR_FLAG))) {
if ((found_record->job_running
== BLOCK_ERROR_STATE)
- || (found_record->state == BG_BLOCK_ERROR))
+ || (found_record->state
+ & BG_BLOCK_ERROR_FLAG))
error("can't use %s, "
"overlapping block %s "
"is in an error state.",
@@ -1026,7 +1028,7 @@ static int _find_best_block_match(List block_list,
*/
bg_record->job_running =
BLOCK_ERROR_STATE;
- bg_record->state = BG_BLOCK_ERROR;
+ bg_record->state |= BG_BLOCK_ERROR_FLAG;
error("_find_best_block_match: Picked "
"block (%s) had some issues with "
"hardware, trying a different "
diff --git a/src/plugins/select/bluegene/bg_node_info.c b/src/plugins/select/bluegene/bg_node_info.c
index e0ac56af4c635aa848c468d4837635b01ce9da5c..c9462bfcf2c7ed21213c277629a89446a659f614 100644
--- a/src/plugins/select/bluegene/bg_node_info.c
+++ b/src/plugins/select/bluegene/bg_node_info.c
@@ -209,12 +209,12 @@ unpack_error:
extern select_nodeinfo_t *select_nodeinfo_alloc(uint32_t size)
{
select_nodeinfo_t *nodeinfo = xmalloc(sizeof(struct select_nodeinfo));
- uint32_t cluster_flags = slurmdb_setup_cluster_flags();
+ //uint32_t cluster_flags = slurmdb_setup_cluster_flags();
if (!g_bitmap_size) {
- if (cluster_flags & CLUSTER_FLAG_BGQ)
- g_bitmap_size = bg_conf->mp_cnode_cnt;
- else
+ /* if (cluster_flags & CLUSTER_FLAG_BGQ) */
+ /* g_bitmap_size = bg_conf->mp_cnode_cnt; */
+ /* else */
g_bitmap_size = bg_conf->ionodes_per_mp;
}
@@ -249,15 +249,15 @@ extern int select_nodeinfo_set_all(time_t last_query_time)
int i=0;
bg_record_t *bg_record = NULL;
static time_t last_set_all = 0;
- uint32_t cluster_flags = slurmdb_setup_cluster_flags();
+ //uint32_t cluster_flags = slurmdb_setup_cluster_flags();
if (!blocks_are_created)
return SLURM_NO_CHANGE_IN_DATA;
if (!g_bitmap_size) {
- if (cluster_flags & CLUSTER_FLAG_BGQ)
- g_bitmap_size = bg_conf->mp_cnode_cnt;
- else
+ /* if (cluster_flags & CLUSTER_FLAG_BGQ) */
+ /* g_bitmap_size = bg_conf->mp_cnode_cnt; */
+ /* else */
g_bitmap_size = bg_conf->ionodes_per_mp;
}
@@ -297,7 +297,7 @@ extern int select_nodeinfo_set_all(time_t last_query_time)
if (bg_record->job_running == NO_JOB_RUNNING)
continue;
- if (bg_record->state == BG_BLOCK_ERROR)
+ if (bg_record->state & BG_BLOCK_ERROR_FLAG)
state = NODE_STATE_ERROR;
else if (bg_record->job_running > NO_JOB_RUNNING) {
/* we don't need to set the allocated here
@@ -311,10 +311,10 @@ extern int select_nodeinfo_set_all(time_t last_query_time)
bg_block_state_string(bg_record->state));
continue;
}
- if ((cluster_flags & CLUSTER_FLAG_BGQ)
- && (state != NODE_STATE_ERROR))
- bitmap = bg_record->cnodes_used_bitmap;
- else
+ /* if ((cluster_flags & CLUSTER_FLAG_BGQ) */
+ /* && (state != NODE_STATE_ERROR)) */
+ /* bitmap = bg_record->cnodes_used_bitmap; */
+ /* else */
bitmap = bg_record->ionode_bitmap;
for (i=0; i<node_record_count; i++) {
@@ -331,11 +331,11 @@ extern int select_nodeinfo_set_all(time_t last_query_time)
state, g_bitmap_size);
if (subgrp->cnode_cnt < bg_conf->mp_cnode_cnt) {
- if (cluster_flags & CLUSTER_FLAG_BGQ) {
- bit_or(subgrp->bitmap, bitmap);
- subgrp->cnode_cnt +=
- bit_set_count(bitmap);
- } else if (bg_record->cnode_cnt
+ /* if (cluster_flags & CLUSTER_FLAG_BGQ) { */
+ /* bit_or(subgrp->bitmap, bitmap); */
+ /* subgrp->cnode_cnt += */
+ /* bit_set_count(bitmap); */
+ /* } else */ if (bg_record->cnode_cnt
< bg_conf->mp_cnode_cnt) {
bit_or(subgrp->bitmap, bitmap);
subgrp->cnode_cnt +=
diff --git a/src/plugins/select/bluegene/bg_record_functions.c b/src/plugins/select/bluegene/bg_record_functions.c
index df7e73262f650df775ac63a48f9b7fa6b61019f8..cc539d12aaa9fd75c2bf9416ded3d8f2ae3bf627 100644
--- a/src/plugins/select/bluegene/bg_record_functions.c
+++ b/src/plugins/select/bluegene/bg_record_functions.c
@@ -1088,7 +1088,7 @@ extern int down_nodecard(char *mp_name, bitoff_t io_start,
*/
if (smallest_bg_record
&& (smallest_bg_record->cnode_cnt < bg_conf->mp_cnode_cnt)){
- if (smallest_bg_record->state == BG_BLOCK_ERROR) {
+ if (smallest_bg_record->state & BG_BLOCK_ERROR_FLAG) {
rc = SLURM_NO_CHANGE_IN_DATA;
goto cleanup;
}
@@ -1157,7 +1157,7 @@ extern int down_nodecard(char *mp_name, bitoff_t io_start,
} else if (smallest_bg_record) {
debug2("smallest dynamic block is %s",
smallest_bg_record->bg_block_id);
- if (smallest_bg_record->state == BG_BLOCK_ERROR) {
+ if (smallest_bg_record->state & BG_BLOCK_ERROR_FLAG) {
rc = SLURM_NO_CHANGE_IN_DATA;
goto cleanup;
}
@@ -1167,7 +1167,8 @@ extern int down_nodecard(char *mp_name, bitoff_t io_start,
if (smallest_bg_record->cnode_cnt == create_size) {
rc = put_block_in_error_state(
- smallest_bg_record, BLOCK_ERROR_STATE, reason);
+ smallest_bg_record, BLOCK_ERROR_STATE,
+ reason);
goto cleanup;
}
@@ -1411,7 +1412,7 @@ extern int put_block_in_error_state(bg_record_t *bg_record,
list_push(bg_lists->booted, bg_record);
bg_record->job_running = state;
- bg_record->state = BG_BLOCK_ERROR;
+ bg_record->state |= BG_BLOCK_ERROR_FLAG;
xfree(bg_record->user_name);
xfree(bg_record->target_name);
@@ -1442,21 +1443,23 @@ extern int resume_block(bg_record_t *bg_record)
if (bg_record->job_running > NO_JOB_RUNNING)
return SLURM_SUCCESS;
- if (bg_record->state == BG_BLOCK_ERROR)
+ if (bg_record->state & BG_BLOCK_ERROR_FLAG) {
+ bg_record->state &= (~BG_BLOCK_ERROR_FLAG);
info("Block %s put back into service after "
"being in an error state.",
bg_record->bg_block_id);
+ }
if (remove_from_bg_list(bg_lists->job_running, bg_record)
== SLURM_SUCCESS)
num_unused_cpus += bg_record->cpu_cnt;
+
if (bg_record->state != BG_BLOCK_INITED)
remove_from_bg_list(bg_lists->booted, bg_record);
+ else if (!block_ptr_exist_in_list(bg_lists->booted, bg_record))
+ list_push(bg_lists->booted, bg_record);
bg_record->job_running = NO_JOB_RUNNING;
-#ifndef HAVE_BG_FILES
- bg_record->state = BG_BLOCK_FREE;
-#endif
xfree(bg_record->reason);
last_bg_update = time(NULL);
@@ -1548,7 +1551,7 @@ static int _check_all_blocks_error(int node_inx, time_t event_time,
itr = list_iterator_create(bg_lists->main);
while ((bg_record = list_next(itr))) {
/* only look at other nodes in error state */
- if (bg_record->state != BG_BLOCK_ERROR)
+ if (!(bg_record->state & BG_BLOCK_ERROR_FLAG))
continue;
if (!bit_test(bg_record->bitmap, node_inx))
continue;
diff --git a/src/plugins/select/bluegene/bg_status.c b/src/plugins/select/bluegene/bg_status.c
index d3e8f965cd07968475c5190ff4b1ef6dd64c9fc4..d3b69ceb0f1f07ef4640f233e2d9a68a13c9fe36 100644
--- a/src/plugins/select/bluegene/bg_status.c
+++ b/src/plugins/select/bluegene/bg_status.c
@@ -119,15 +119,15 @@ static int _block_is_deallocating(bg_record_t *bg_record, List kill_job_list)
}
extern int bg_status_update_block_state(bg_record_t *bg_record,
- bg_block_status_t state,
+ uint16_t state,
List kill_job_list)
{
bool skipped_dealloc = false;
kill_job_struct_t *freeit = NULL;
int updated = 0;
+ uint16_t real_state = bg_record->state & (~BG_BLOCK_ERROR_FLAG);
- if (bg_record->job_running == BLOCK_ERROR_STATE
- || bg_record->state == state)
+ if (real_state == state)
return 0;
debug("state of Block %s was %s and now is %s",
@@ -139,8 +139,8 @@ extern int bg_status_update_block_state(bg_record_t *bg_record,
check to make sure block went
through freeing correctly
*/
- if ((bg_record->state != BG_BLOCK_TERM
- && bg_record->state != BG_BLOCK_ERROR)
+ if ((real_state != BG_BLOCK_TERM
+ && !(bg_record->state & BG_BLOCK_ERROR_FLAG))
&& state == BG_BLOCK_FREE)
skipped_dealloc = 1;
else if ((bg_record->state == BG_BLOCK_INITED)
@@ -166,7 +166,12 @@ extern int bg_status_update_block_state(bg_record_t *bg_record,
state and act like this didn't happen. */
goto nochange_state;
- bg_record->state = state;
+ if (bg_record->state & BG_BLOCK_ERROR_FLAG)
+ state |= BG_BLOCK_ERROR_FLAG;
+ else if (state & BG_BLOCK_ERROR_FLAG)
+ bg_record->state |= state;
+ else
+ bg_record->state = state;
if (bg_record->state == BG_BLOCK_TERM || skipped_dealloc)
_block_is_deallocating(bg_record, kill_job_list);
@@ -179,7 +184,7 @@ extern int bg_status_update_block_state(bg_record_t *bg_record,
num_unused_cpus += bg_record->cpu_cnt;
remove_from_bg_list(bg_lists->booted,
bg_record);
- } else if (bg_record->state == BG_BLOCK_ERROR) {
+ } else if (bg_record->state & BG_BLOCK_ERROR_FLAG) {
if (bg_record->boot_state)
error("Block %s in an error state while booting.",
bg_record->bg_block_id);
@@ -199,7 +204,19 @@ nochange_state:
debug3("boot state for block %s is %d",
bg_record->bg_block_id, bg_record->boot_state);
if (bg_record->boot_state) {
- switch(bg_record->state) {
+ if (bg_record->state & BG_BLOCK_ERROR_FLAG) {
+ /* If we get an error on boot that
+ * means it is a transparent L3 error
+ * and should be trying to fix
+ * itself. If this is the case we
+ * just hang out waiting for the state
+ * to go to free where we will try to
+ * boot again below.
+ */
+ return updated;
+ }
+
+ switch (bg_record->state) {
case BG_BLOCK_BOOTING:
debug3("checking to make sure user %s "
"is the user.",
@@ -213,16 +230,6 @@ nochange_state:
last_job_update = time(NULL);
}
break;
- case BG_BLOCK_ERROR:
- /* If we get an error on boot that
- * means it is a transparent L3 error
- * and should be trying to fix
- * itself. If this is the case we
- * just hang out waiting for the state
- * to go to free where we will try to
- * boot again below.
- */
- break;
case BG_BLOCK_FREE:
if (bg_record->boot_count < RETRY_BOOT_COUNT) {
boot_block(bg_record);
diff --git a/src/plugins/select/bluegene/bg_status.h b/src/plugins/select/bluegene/bg_status.h
index ea0b41d78371413c1dbdf3c5f8874a1cf7c84352..2e854152f57edad5ecaf3d7c591ecc0fcb90cac9 100644
--- a/src/plugins/select/bluegene/bg_status.h
+++ b/src/plugins/select/bluegene/bg_status.h
@@ -41,7 +41,7 @@
#include "bg_core.h"
extern int bg_status_update_block_state(bg_record_t *bg_record,
- bg_block_status_t state,
+ uint16_t state,
List kill_job_list);
extern List bg_status_create_kill_job_list(void);
extern void bg_status_process_kill_job_list(List kill_job_list);
diff --git a/src/plugins/select/bluegene/bl/bridge_linker.c b/src/plugins/select/bluegene/bl/bridge_linker.c
index 9799ba07bdbcb4abf40441d65e78bd4497842024..a0a6b05cb49c197623308f367802679cfea3ffdd 100644
--- a/src/plugins/select/bluegene/bl/bridge_linker.c
+++ b/src/plugins/select/bluegene/bl/bridge_linker.c
@@ -2152,7 +2152,7 @@ extern status_t bridge_get_data(rm_element_t* element,
*state = BG_BLOCK_TERM;
break;
case RM_PARTITION_ERROR:
- *state = BG_BLOCK_ERROR;
+ *state |= BG_BLOCK_ERROR_FLAG;
break;
case RM_PARTITION_NAV:
*state = BG_BLOCK_NAV;
diff --git a/src/plugins/select/bluegene/bl_bgq/bridge_helper.cc b/src/plugins/select/bluegene/bl_bgq/bridge_helper.cc
index e61ac8cc7155baa301a743c2d39982777432848f..16e7f4816ac9237b19d5f1f2a9299935bf6c6447 100644
--- a/src/plugins/select/bluegene/bl_bgq/bridge_helper.cc
+++ b/src/plugins/select/bluegene/bl_bgq/bridge_helper.cc
@@ -250,8 +250,7 @@ extern int bridge_handle_runtime_errors(const char *function,
return rc;
}
-extern bg_block_status_t bridge_translate_status(
- bgsched::Block::Status state_in)
+extern uint16_t bridge_translate_status(bgsched::Block::Status state_in)
{
switch (state_in) {
case Block::Allocated:
@@ -270,7 +269,7 @@ extern bg_block_status_t bridge_translate_status(
return BG_BLOCK_TERM;
break;
default:
- return BG_BLOCK_ERROR;
+ return BG_BLOCK_ERROR_FLAG;
break;
}
error("unknown block state %d", state_in);
diff --git a/src/plugins/select/bluegene/select_bluegene.c b/src/plugins/select/bluegene/select_bluegene.c
index a4002347073fd5f196b15df95df6c6d4b3d9e242..1aadaffbee9066a606748836f544aa5579404de6 100644
--- a/src/plugins/select/bluegene/select_bluegene.c
+++ b/src/plugins/select/bluegene/select_bluegene.c
@@ -755,7 +755,7 @@ extern int select_p_state_save(char *dir_name)
* the blocks in an error state
*/
#if defined HAVE_BG_FILES
- if (bg_record->state != BG_BLOCK_ERROR)
+ if (!(bg_record->state & BG_BLOCK_ERROR_FLAG))
continue;
#endif
xassert(bg_record->bg_block_id != NULL);
@@ -877,7 +877,7 @@ extern int select_p_state_restore(char *dir_name)
no threads are started before this function. */
itr = list_iterator_create(bg_lists->main);
while ((bg_record = list_next(itr))) {
- if (bg_record->state == BG_BLOCK_ERROR)
+ if (bg_record->state & BG_BLOCK_ERROR_FLAG)
put_block_in_error_state(bg_record,
BLOCK_ERROR_STATE, NULL);
}
@@ -1337,7 +1337,7 @@ extern int select_p_update_block(update_block_msg_t *block_desc_ptr)
bg_record->job_ptr = NULL;
}
- if (block_desc_ptr->state == BG_BLOCK_ERROR) {
+ if (block_desc_ptr->state == BG_BLOCK_ERROR_FLAG) {
bg_record_t *found_record = NULL;
ListIterator itr;
List delete_list = list_create(NULL);
@@ -1470,7 +1470,7 @@ extern int select_p_update_block(update_block_msg_t *block_desc_ptr)
to a normal state in accounting first */
itr = list_iterator_create(delete_list);
while ((found_record = list_next(itr))) {
- if (found_record->state == BG_BLOCK_ERROR)
+ if (found_record->state & BG_BLOCK_ERROR_FLAG)
resume_block(found_record);
}
list_iterator_destroy(itr);
@@ -1485,7 +1485,7 @@ extern int select_p_update_block(update_block_msg_t *block_desc_ptr)
/* make sure if we are removing a block to put it back
to a normal state in accounting first */
- if (bg_record->state == BG_BLOCK_ERROR)
+ if (bg_record->state & BG_BLOCK_ERROR_FLAG)
resume_block(bg_record);
term_jobs_on_block(bg_record->bg_block_id);
@@ -1658,7 +1658,7 @@ extern int select_p_update_sub_node (update_block_msg_t *block_desc_ptr)
}
node_name = xstrdup_printf("%s%s", bg_conf->slurm_node_prefix, coord);
/* find out how many nodecards to get for each ionode */
- if (block_desc_ptr->state == BG_BLOCK_ERROR) {
+ if (block_desc_ptr->state == BG_BLOCK_ERROR_FLAG) {
info("Admin setting %s[%s] in an error state",
node_name, ionodes);
for(i = 0; i<bg_conf->ionodes_per_mp; i++) {
diff --git a/src/scontrol/scontrol.c b/src/scontrol/scontrol.c
index 6c151350998aed1f333e559eee0c9ed8c954ac32..8e36d80d8d2c315a5dc45ee9765938c01c2b64c7 100644
--- a/src/scontrol/scontrol.c
+++ b/src/scontrol/scontrol.c
@@ -1474,7 +1474,7 @@ _update_bluegene_block (int argc, char *argv[])
block_msg.bg_block_id = val;
} else if (!strncasecmp(tag, "State", MAX(tag_len, 2))) {
if (!strncasecmp(val, "ERROR", MAX(vallen, 1)))
- block_msg.state = BG_BLOCK_ERROR;
+ block_msg.state = BG_BLOCK_ERROR_FLAG;
else if (!strncasecmp(val, "FREE", MAX(vallen, 1)))
block_msg.state = BG_BLOCK_FREE;
else if (!strncasecmp(val, "RECREATE", MAX(vallen, 3)))
@@ -1560,7 +1560,7 @@ _update_bluegene_subbp (int argc, char *argv[])
block_msg.mp_str = val;
else if (!strncasecmp(tag, "State", MAX(tag_len, 2))) {
if (!strncasecmp(val, "ERROR", MAX(vallen, 1)))
- block_msg.state = BG_BLOCK_ERROR;
+ block_msg.state = BG_BLOCK_ERROR_FLAG;
else if (!strncasecmp(val, "FREE", MAX(vallen, 1)))
block_msg.state = BG_BLOCK_FREE;
else {
diff --git a/src/slurmctld/proc_req.c b/src/slurmctld/proc_req.c
index 4b911fc532f14d38de8548cfd8d4b314d7879e05..c3a001b8a006ad701a0bc864390c93550b943587 100644
--- a/src/slurmctld/proc_req.c
+++ b/src/slurmctld/proc_req.c
@@ -1560,7 +1560,7 @@ static void _slurm_rpc_complete_batch_script(slurm_msg_t * msg)
#ifdef HAVE_BG
if (block_desc.bg_block_id) {
block_desc.reason = slurm_strerror(comp_msg->slurm_rc);
- block_desc.state = BG_BLOCK_ERROR;
+ block_desc.state = BG_BLOCK_ERROR_FLAG;
i = select_g_update_block(&block_desc);
error_code = MAX(error_code, i);
xfree(block_desc.bg_block_id);
diff --git a/src/sview/block_info.c b/src/sview/block_info.c
index b2fc2cbea9e7ce7a30c58cb92c62a54ad4cb8edc..b3e28e609f99606aba21842403fbeaf589a0c053 100644
--- a/src/sview/block_info.c
+++ b/src/sview/block_info.c
@@ -627,7 +627,7 @@ need_refresh:
state */
enum node_states state = NODE_STATE_UNKNOWN;
- if (block_ptr->state == BG_BLOCK_ERROR)
+ if (block_ptr->state & BG_BLOCK_ERROR_FLAG)
state = NODE_STATE_ERROR;
else if (block_ptr->job_running > NO_JOB_RUNNING)
state = NODE_STATE_ALLOCATED;
@@ -779,7 +779,7 @@ extern int update_state_block(GtkDialog *dialog,
"Are you sure you want to put block %s "
"in an error state?",
blockid);
- block_msg.state = BG_BLOCK_ERROR;
+ block_msg.state = BG_BLOCK_ERROR_FLAG;
} else if (!strcasecmp("Recreate block", type)) {
snprintf(tmp_char, sizeof(tmp_char),
"Are you sure you want to recreate block %s?",
@@ -1252,7 +1252,7 @@ display_it:
}
list_push(send_block_list, block_ptr);
- if (block_ptr->state == BG_BLOCK_ERROR)
+ if (block_ptr->state & BG_BLOCK_ERROR_FLAG)
state = NODE_STATE_ERROR;
else if (block_ptr->job_running > NO_JOB_RUNNING)
state = NODE_STATE_ALLOCATED;
diff --git a/src/sview/grid.c b/src/sview/grid.c
index fcc9fec30ce9d87fbad67fa7c73f9aa6bf4f02bd..750e9463b9f55bde0698b1e2efbbc62d53d753fd 100644
--- a/src/sview/grid.c
+++ b/src/sview/grid.c
@@ -1361,7 +1361,7 @@ extern void add_extra_bluegene_buttons(List *button_list, int inx,
bg_info_ptr->ionode_str);
mp_str = tmp_nodes;
}
- if (bg_info_ptr->state == BG_BLOCK_ERROR)
+ if (bg_info_ptr->state & BG_BLOCK_ERROR_FLAG)
grid_button->state = NODE_STATE_ERROR;
else if (bg_info_ptr->job_running > NO_JOB_RUNNING)
grid_button->state = NODE_STATE_ALLOCATED;
diff --git a/src/sview/popups.c b/src/sview/popups.c
index 2e5717b93c2f1cad2dcad6b27fb841b7407d7f38..942752b93bd54ed53473bc542b12546d375f4162 100644
--- a/src/sview/popups.c
+++ b/src/sview/popups.c
@@ -759,7 +759,7 @@ extern void create_search_popup(GtkAction *action, gpointer user_data)
{G_TYPE_NONE, BG_BLOCK_INITED, "Inited", TRUE, -1},
{G_TYPE_NONE, BG_BLOCK_ALLOCATED, NULL, TRUE, -1},
{G_TYPE_NONE, BG_BLOCK_TERM, "Terminating", TRUE, -1},
- {G_TYPE_NONE, BG_BLOCK_ERROR, "Error", TRUE, -1},
+ {G_TYPE_NONE, BG_BLOCK_ERROR_FLAG, "Error", TRUE, -1},
{G_TYPE_NONE, -1, NULL, FALSE, -1}
};
display_data_t *display_data = pulldown_display_data;