diff --git a/src/common/gres.c b/src/common/gres.c
index fd2a1a2cb13d73b0bc240318dc8b3b7d07f5aac5..b6cdbec92bbd0cb88223af33d204a185b56a05d3 100644
--- a/src/common/gres.c
+++ b/src/common/gres.c
@@ -164,6 +164,8 @@ static void _set_gres_cnt(char *orig_config, char **new_config,
char *gres_name_colon, int gres_name_colon_len);
static int _step_state_validate(char *config, void **gres_data,
slurm_gres_context_t *context_ptr);
+static uint32_t _step_test(void *step_gres_data, void *job_gres_data,
+ int node_offset, bool ignore_alloc, char *gres_name);
static int _strcmp(const char *s1, const char *s2);
static int _unload_gres_plugin(slurm_gres_context_t *plugin_context);
static void _validate_config(slurm_gres_context_t *context_ptr);
@@ -2144,6 +2146,9 @@ List gres_plugin_job_state_dup(List gres_list)
* IN gres_list - generated by gres_plugin_job_config_validate()
* IN/OUT buffer - location to write state to
* IN job_id - job's ID
+ *
+ * NOTE: A job's allocation to steps is not recorded here, but recovered with
+ * the job step state information upon slurmctld restart.
*/
extern int gres_plugin_job_state_pack(List gres_list, Buf buffer,
uint32_t job_id)
@@ -2152,7 +2157,6 @@ extern int gres_plugin_job_state_pack(List gres_list, Buf buffer,
uint32_t top_offset, tail_offset;
uint32_t magic = GRES_MAGIC;
uint16_t rec_cnt = 0;
- uint8_t has_bitmap;
ListIterator gres_iter;
gres_state_t *gres_ptr;
gres_job_state_t *gres_job_ptr;
@@ -2174,15 +2178,13 @@ extern int gres_plugin_job_state_pack(List gres_list, Buf buffer,
pack32(gres_job_ptr->gres_cnt_alloc, buffer);
pack32(gres_job_ptr->node_cnt, buffer);
if (gres_job_ptr->gres_bit_alloc) {
- has_bitmap = 1;
- pack8(has_bitmap, buffer);
+ pack8((uint8_t) 1, buffer);
for (i=0; i<gres_job_ptr->node_cnt; i++) {
pack_bit_str(gres_job_ptr->gres_bit_alloc[i],
buffer);
}
} else {
- has_bitmap = 0;
- pack8(has_bitmap, buffer);
+ pack8((uint8_t) 0, buffer);
}
rec_cnt++;
}
@@ -2252,11 +2254,11 @@ extern int gres_plugin_job_state_unpack(List *gres_list, Buf buffer,
break;
}
if (i >= gres_context_cnt) {
+ /* A likely sign that GresPlugins has changed.
+ * Not a fatal error, skip over the data. */
error("gres_plugin_job_state_unpack: no plugin "
"configured to unpack data type %u from job %u",
plugin_id, job_id);
- /* A likely sign that GresPlugins has changed.
- * Not a fatal error, skip over the data. */
_job_state_delete(gres_job_ptr);
continue;
}
@@ -3123,24 +3125,6 @@ List gres_plugin_step_state_dup(List gres_list)
return new_gres_list;
}
-static int _step_state_pack(void *gres_data, Buf buffer)
-{
- int i;
- gres_step_state_t *gres_ptr = (gres_step_state_t *) gres_data;
-
- pack32(gres_ptr->gres_cnt_alloc, buffer);
- pack32(gres_ptr->node_cnt, buffer);
- if (gres_ptr->gres_bit_alloc) {
- pack8((uint8_t) 1, buffer);
- for (i=0; i<gres_ptr->node_cnt; i++)
- pack_bit_str(gres_ptr->gres_bit_alloc[i], buffer);
- } else {
- pack8((uint8_t) 0, buffer);
- }
-
- return SLURM_SUCCESS;
-}
-
/*
* Pack a step's current gres status, called from slurmctld for save/restore
* IN gres_list - generated by gres_plugin_step_allocate()
@@ -3150,13 +3134,12 @@ static int _step_state_pack(void *gres_data, Buf buffer)
extern int gres_plugin_step_state_pack(List gres_list, Buf buffer,
uint32_t job_id, uint32_t step_id)
{
- int i, rc = SLURM_SUCCESS, rc2;
- uint32_t top_offset, gres_size = 0;
- uint32_t header_offset, size_offset, data_offset, tail_offset;
- uint32_t magic = GRES_MAGIC;
+ int i, rc = SLURM_SUCCESS;
+ uint32_t top_offset, tail_offset, magic = GRES_MAGIC;
uint16_t rec_cnt = 0;
ListIterator gres_iter;
gres_state_t *gres_ptr;
+ gres_step_state_t *gres_step_ptr;
top_offset = get_buf_offset(buffer);
pack16(rec_cnt, buffer); /* placeholder if data */
@@ -3169,35 +3152,20 @@ extern int gres_plugin_step_state_pack(List gres_list, Buf buffer,
slurm_mutex_lock(&gres_context_lock);
gres_iter = list_iterator_create(gres_list);
while ((gres_ptr = (gres_state_t *) list_next(gres_iter))) {
- for (i=0; i<gres_context_cnt; i++) {
- if (gres_ptr->plugin_id !=
- gres_context[i].plugin_id)
- continue;
- header_offset = get_buf_offset(buffer);
- pack32(magic, buffer);
- pack32(gres_ptr->plugin_id, buffer);
- size_offset = get_buf_offset(buffer);
- pack32(gres_size, buffer); /* placeholder */
- data_offset = get_buf_offset(buffer);
- rc2 = _step_state_pack(gres_ptr->gres_data, buffer);
- if (rc2 != SLURM_SUCCESS) {
- rc = rc2;
- set_buf_offset(buffer, header_offset);
- continue;
- }
- tail_offset = get_buf_offset(buffer);
- set_buf_offset(buffer, size_offset);
- gres_size = tail_offset - data_offset;
- pack32(gres_size, buffer);
- set_buf_offset(buffer, tail_offset);
- rec_cnt++;
- break;
- }
- if (i >= gres_context_cnt) {
- error("Could not find plugin id %u to pack record for "
- "step %u.%u",
- gres_ptr->plugin_id, job_id, step_id);
+ gres_step_ptr = (gres_step_state_t *) gres_ptr->gres_data;
+ pack32(magic, buffer);
+ pack32(gres_ptr->plugin_id, buffer);
+ pack32(gres_step_ptr->gres_cnt_alloc, buffer);
+ pack32(gres_step_ptr->node_cnt, buffer);
+ if (gres_step_ptr->gres_bit_alloc) {
+ pack8((uint8_t) 1, buffer);
+ for (i=0; i<gres_step_ptr->node_cnt; i++)
+ pack_bit_str(gres_step_ptr->gres_bit_alloc[i],
+ buffer);
+ } else {
+ pack8((uint8_t) 0, buffer);
}
+ rec_cnt++;
}
list_iterator_destroy(gres_iter);
slurm_mutex_unlock(&gres_context_lock);
@@ -3210,42 +3178,6 @@ extern int gres_plugin_step_state_pack(List gres_list, Buf buffer,
return rc;
}
-static int _step_state_unpack(void **gres_data, Buf buffer, char *gres_name)
-{
- int i;
- gres_step_state_t *gres_ptr;
- uint8_t gres_bit_flag;
-
- gres_ptr = xmalloc(sizeof(gres_step_state_t));
-
- if (buffer) {
- safe_unpack32(&gres_ptr->gres_cnt_alloc, buffer);
- safe_unpack32(&gres_ptr->node_cnt, buffer);
- safe_unpack8(&gres_bit_flag, buffer);
- if (gres_bit_flag) {
- gres_ptr->gres_bit_alloc = xmalloc(sizeof(bitstr_t *) *
- gres_ptr->node_cnt);
- for (i=0; i<gres_ptr->node_cnt; i++)
- unpack_bit_str(&gres_ptr->gres_bit_alloc[i],
- buffer);
- }
- }
-
- *gres_data = gres_ptr;
- return SLURM_SUCCESS;
-
-unpack_error:
- error("Unpacking gres/%s step state info", gres_name);
- if (gres_ptr->gres_bit_alloc) {
- for (i=0; i<gres_ptr->node_cnt; i++)
- FREE_NULL_BITMAP(gres_ptr->gres_bit_alloc[i]);
- xfree(gres_ptr->gres_bit_alloc);
- }
- xfree(gres_ptr);
- *gres_data = NULL;
- return SLURM_ERROR;
-}
-
/*
* Unpack a step's current gres status, called from slurmctld for save/restore
* OUT gres_list - restored state stored by gres_plugin_step_state_pack()
@@ -3255,11 +3187,12 @@ unpack_error:
extern int gres_plugin_step_state_unpack(List *gres_list, Buf buffer,
uint32_t job_id, uint32_t step_id)
{
- int i, rc, rc2;
- uint32_t gres_size, magic, tail_offset, plugin_id;
+ int i, rc;
+ uint32_t magic, plugin_id;
uint16_t rec_cnt;
+ uint8_t has_file;
gres_state_t *gres_ptr;
- void *gres_data;
+ gres_step_state_t *gres_step_ptr = NULL;
safe_unpack16(&rec_cnt, buffer);
if (rec_cnt == 0)
@@ -3282,41 +3215,49 @@ extern int gres_plugin_step_state_unpack(List *gres_list, Buf buffer,
if (magic != GRES_MAGIC)
goto unpack_error;
safe_unpack32(&plugin_id, buffer);
- safe_unpack32(&gres_size, buffer);
+ gres_step_ptr = xmalloc(sizeof(gres_step_state_t));
+ safe_unpack32(&gres_step_ptr->gres_cnt_alloc, buffer);
+ safe_unpack32(&gres_step_ptr->node_cnt, buffer);
+ safe_unpack8(&has_file, buffer);
+ if (has_file) {
+ gres_step_ptr->gres_bit_alloc =
+ xmalloc(sizeof(bitstr_t) *
+ gres_step_ptr->node_cnt);
+ for (i=0; i<gres_step_ptr->node_cnt; i++) {
+ unpack_bit_str(&gres_step_ptr->gres_bit_alloc[i],
+ buffer);
+ }
+ }
+
for (i=0; i<gres_context_cnt; i++) {
if (gres_context[i].plugin_id == plugin_id)
break;
}
if (i >= gres_context_cnt) {
- error("gres_plugin_step_state_unpack: no plugin "
+ /* A likely sign that GresPlugins has changed.
+ * Not a fatal error, skip over the data. */
+ info("gres_plugin_step_state_unpack: no plugin "
"configured to unpack data type %u from "
"step %u.%u",
plugin_id, job_id, step_id);
- /* A likely sign that GresPlugins has changed.
- * Not a fatal error, skip over the data. */
- tail_offset = get_buf_offset(buffer);
- tail_offset += gres_size;
- set_buf_offset(buffer, tail_offset);
+ _step_state_delete(gres_step_ptr);
+ gres_step_ptr = NULL;
continue;
}
- rc2 = _step_state_unpack(&gres_data, buffer,
- gres_context[i].gres_name);
- if (rc2 != SLURM_SUCCESS) {
- rc = rc2;
- } else {
- gres_ptr = xmalloc(sizeof(gres_state_t));
- gres_ptr->plugin_id = gres_context[i].plugin_id;
- gres_ptr->gres_data = gres_data;
- list_append(*gres_list, gres_ptr);
- }
+ gres_ptr = xmalloc(sizeof(gres_state_t));
+ gres_ptr->plugin_id = gres_context[i].plugin_id;
+ gres_ptr->gres_data = gres_step_ptr;
+ gres_step_ptr = NULL;
+ list_append(*gres_list, gres_ptr);
}
slurm_mutex_unlock(&gres_context_lock);
-
return rc;
unpack_error:
- error("gres_plugin_job_state_unpack: unpack error from step %u.%u",
+ error("gres_plugin_step_state_unpack: unpack error from step %u.%u",
job_id, step_id);
+ if (gres_step_ptr)
+ _step_state_delete(gres_step_ptr);
slurm_mutex_unlock(&gres_context_lock);
return SLURM_ERROR;
}
@@ -3333,7 +3274,9 @@ static void _step_state_log(void *gres_data, uint32_t job_id, uint32_t step_id,
info(" gres_cnt:%u node_cnt:%u", gres_ptr->gres_cnt_alloc,
gres_ptr->node_cnt);
- if (gres_ptr->node_cnt && gres_ptr->gres_bit_alloc) {
+ if (gres_ptr->gres_bit_alloc == NULL)
+ info(" gres_bit_alloc:NULL");
+ else {
for (i=0; i<gres_ptr->node_cnt; i++) {
if (gres_ptr->gres_bit_alloc[i]) {
bit_fmt(tmp_str, sizeof(tmp_str),
@@ -3342,8 +3285,6 @@ static void _step_state_log(void *gres_data, uint32_t job_id, uint32_t step_id,
} else
info(" gres_bit_alloc[%d]:NULL", i);
}
- } else {
- info(" gres_bit_alloc:NULL");
}
}
@@ -3368,8 +3309,7 @@ extern void gres_plugin_step_state_log(List gres_list, uint32_t job_id,
gres_iter = list_iterator_create(gres_list);
while ((gres_ptr = (gres_state_t *) list_next(gres_iter))) {
for (i=0; i<gres_context_cnt; i++) {
- if (gres_ptr->plugin_id !=
- gres_context[i].plugin_id)
+ if (gres_ptr->plugin_id != gres_context[i].plugin_id)
continue;
_step_state_log(gres_ptr->gres_data, job_id, step_id,
gres_context[i].gres_name);
diff --git a/src/common/gres.h b/src/common/gres.h
index df3a5d36584f4980d44f12e46d0fb4010d14ffa1..a4a91ad0b27148e4a0f06fb67745991e259608b1 100644
--- a/src/common/gres.h
+++ b/src/common/gres.h
@@ -113,7 +113,6 @@ typedef struct gres_step_state {
/* Resources currently allocated to the job step on each node */
uint32_t node_cnt;
bitstr_t **gres_bit_alloc;
- uint32_t *gres_cnt_step_alloc;
} gres_step_state_t;
/*
@@ -296,6 +295,9 @@ List gres_plugin_job_state_dup(List gres_list);
* IN gres_list - generated by gres_plugin_job_config_validate()
* IN/OUT buffer - location to write state to
* IN job_id - job's ID
+ *
+ * NOTE: A job's allocation to steps is not recorded here, but recovered with
+ * the job step state information upon slurmctld restart.
*/
extern int gres_plugin_job_state_pack(List gres_list, Buf buffer,
uint32_t job_id);