extern int parse_image(void **dest, slurm_parser_enum_t type,
const char *key, const char *value,
const char *line, char **leftover)
{
s_p_options_t image_options[] = {
{(char *)"GROUPS", S_P_STRING},
{NULL}
};
s_p_hashtbl_t *tbl = NULL;
char *tmp = NULL;
image_t *n = NULL;
image_group_t *image_group = NULL;
int i = 0, j = 0;
tbl = s_p_hashtbl_create(image_options);
s_p_parse_line(tbl, *leftover, leftover);
n = (image_t *)xmalloc(sizeof(image_t));
n->name = xstrdup(value);
n->def = false;
n->groups = list_create(destroy_image_group_list);
s_p_get_string(&tmp, "Groups", tbl);
if (tmp) {
for(i=0; i<(int)strlen(tmp); i++) {
if ((tmp[i] == ':') || (tmp[i] == ',')) {
image_group = (image_group_t *)
xmalloc(sizeof(image_group_t));
image_group->name = (char *)xmalloc(i-j+2);
snprintf(image_group->name,
(i-j)+1, "%s", tmp+j);
gid_from_string (image_group->name,
&image_group->gid);
list_append(n->groups, image_group);
j=i;
j++;
}
}
if (j != i) {
image_group = (image_group_t *)
xmalloc(sizeof(image_group_t));
image_group->name = (char *)xmalloc(i-j+2);
snprintf(image_group->name, (i-j)+1, "%s", tmp+j);
if (gid_from_string (image_group->name,
&image_group->gid) < 0)
fatal("Invalid bluegene.conf parameter "
"Groups=%s",
image_group->name);
list_append(n->groups, image_group);
}
xfree(tmp);
}
s_p_hashtbl_destroy(tbl);
*dest = (void *)n;
return 1;
}
static int _parse_switches(void **dest, slurm_parser_enum_t type,
const char *key, const char *value,
const char *line, char **leftover)
{
s_p_hashtbl_t *tbl;
slurm_conf_switches_t *s;
static s_p_options_t _switch_options[] = {
{"LinkSpeed", S_P_UINT32},
{"Nodes", S_P_STRING},
{"Switches", S_P_STRING},
{NULL}
};
tbl = s_p_hashtbl_create(_switch_options);
s_p_parse_line(tbl, *leftover, leftover);
s = xmalloc(sizeof(slurm_conf_switches_t));
s->switch_name = xstrdup(value);
if (!s_p_get_uint32(&s->link_speed, "LinkSpeed", tbl))
s->link_speed = 1;
s_p_get_string(&s->nodes, "Nodes", tbl);
s_p_get_string(&s->switches, "Switches", tbl);
s_p_hashtbl_destroy(tbl);
if (s->nodes && s->switches) {
error("switch %s has both child switches and nodes",
s->switch_name);
_destroy_switches(s);
return -1;
}
if (!s->nodes && !s->switches) {
error("switch %s has neither child switches nor nodes",
s->switch_name);
_destroy_switches(s);
return -1;
}
*dest = (void *)s;
return 1;
}
/* Determine the current actual burst buffer state.
* Run the program "get_sys_state" and parse stdout for details. */
static void _load_state(void)
{
static uint32_t last_total_space = 0;
char *save_ptr = NULL, *tok, *leftover = NULL, *resp, *tmp = NULL;
char *script_args[3] = { NULL, "get_sys", NULL };
s_p_hashtbl_t *state_hashtbl = NULL;
static s_p_options_t state_options[] = {
{"ENOENT", S_P_STRING},
{"UserID", S_P_ARRAY, _parse_job_info, _destroy_job_info},
{"TotalSize", S_P_STRING},
{NULL}
};
tok = strrchr(get_sys_state, '/');
if (tok)
script_args[0] = tok + 1;
else
script_args[0] = get_sys_state;
resp = _run_script("GetSysState", get_sys_state, script_args, 10);
if (resp == NULL)
return;
state_hashtbl = s_p_hashtbl_create(state_options);
tok = strtok_r(resp, "\n", &save_ptr);
while (tok) {
s_p_parse_line(state_hashtbl, tok, &leftover);
tok = strtok_r(NULL, "\n", &save_ptr);
}
if (s_p_get_string(&tmp, "TotalSize", state_hashtbl)) {
total_space = _get_size_num(tmp);
xfree(tmp);
} else {
error("%s: GetSysState failed to respond with TotalSize",
plugin_type);
}
s_p_hashtbl_destroy(state_hashtbl);
if (debug_flag && (total_space != last_total_space))
info("%s: total_space:%u", __func__, total_space);
last_total_space = total_space;
}
/*
* Determine the current actual burst buffer state.
* Run the program "get_sys_state" and parse stdout for details.
* job_id IN - specific job to get information about, or 0 for all jobs
*/
static void _load_state(uint32_t job_id)
{
static uint64_t last_total_space = 0;
char *save_ptr = NULL, *tok, *leftover = NULL, *resp, *tmp = NULL;
char *script_args[4], job_id_str[32];
s_p_hashtbl_t *state_hashtbl = NULL;
static s_p_options_t state_options[] = {
{"ENOENT", S_P_STRING},
{"UserID", S_P_ARRAY, _parse_job_info, _destroy_job_info},
{"TotalSize", S_P_STRING},
{NULL}
};
int status = 0;
DEF_TIMERS;
if (!bb_state.bb_config.get_sys_state)
return;
bb_state.last_load_time = time(NULL);
tok = strrchr(bb_state.bb_config.get_sys_state, '/');
if (tok)
script_args[0] = tok + 1;
else
script_args[0] = bb_state.bb_config.get_sys_state;
if (job_id) {
script_args[1] = "get_job";
snprintf(job_id_str, sizeof(job_id_str), "%u", job_id);
script_args[3] = NULL;
} else {
script_args[1] = "get_sys";
script_args[2] = NULL;
}
START_TIMER;
resp = bb_run_script("GetSysState", bb_state.bb_config.get_sys_state,
script_args, 2000, &status);
if (resp == NULL)
return;
END_TIMER;
if (DELTA_TIMER > 200000) /* 0.2 secs */
info("%s: GetSysState ran for %s", __func__, TIME_STR);
else if (bb_state.bb_config.debug_flag)
debug("%s: GetSysState ran for %s", __func__, TIME_STR);
state_hashtbl = s_p_hashtbl_create(state_options);
tok = strtok_r(resp, "\n", &save_ptr);
while (tok) {
s_p_parse_line(state_hashtbl, tok, &leftover);
tok = strtok_r(NULL, "\n", &save_ptr);
}
if (s_p_get_string(&tmp, "TotalSize", state_hashtbl)) {
bb_state.total_space = bb_get_size_num(tmp,
bb_state.bb_config.granularity);
xfree(tmp);
if (bb_state.bb_config.debug_flag &&
(bb_state.total_space != last_total_space)) {
info("%s: total_space:%"PRIu64"", __func__,
bb_state.total_space);
}
last_total_space = bb_state.total_space;
} else if (job_id == 0) {
error("%s: GetSysState failed to respond with TotalSize",
plugin_type);
}
s_p_hashtbl_destroy(state_hashtbl);
xfree(resp);
}
static int _parse_job_info(void **dest, slurm_parser_enum_t type,
const char *key, const char *value,
const char *line, char **leftover)
{
s_p_hashtbl_t *job_tbl;
char *name = NULL, *tmp = NULL, local_name[64] = "";
uint64_t size = 0;
uint32_t job_id = 0, user_id = 0;
uint16_t state = 0;
bb_alloc_t *bb_ptr;
struct job_record *job_ptr = NULL;
bb_job_t *bb_spec;
static s_p_options_t _job_options[] = {
{"JobID",S_P_STRING},
{"Name", S_P_STRING},
{"Size", S_P_STRING},
{"State", S_P_STRING},
{NULL}
};
*dest = NULL;
user_id = strtol(value, NULL, 10);
job_tbl = s_p_hashtbl_create(_job_options);
s_p_parse_line(job_tbl, *leftover, leftover);
if (s_p_get_string(&tmp, "JobID", job_tbl)) {
job_id = strtol(tmp, NULL, 10);
xfree(tmp);
}
if (s_p_get_string(&name, "Name", job_tbl)) {
snprintf(local_name, sizeof(local_name), "%s", name);
xfree(name);
}
if (s_p_get_string(&tmp, "Size", job_tbl)) {
size = bb_get_size_num(tmp, bb_state.bb_config.granularity);
xfree(tmp);
}
if (s_p_get_string(&tmp, "State", job_tbl)) {
state = bb_state_num(tmp);
xfree(tmp);
}
s_p_hashtbl_destroy(job_tbl);
#if 0
info("%s: JobID:%u Name:%s Size:%"PRIu64" State:%u UserID:%u",
__func__, job_id, local_name, size, state, user_id);
#endif
if (job_id) {
job_ptr = find_job_record(job_id);
if (!job_ptr && (state == BB_STATE_STAGED_OUT)) {
struct job_record job_rec;
job_rec.job_id = job_id;
job_rec.user_id = user_id;
bb_ptr = bb_find_alloc_rec(&bb_state, &job_rec);
_stop_stage_out(job_id); /* Purge buffer */
if (bb_ptr) {
bb_ptr->cancelled = true;
bb_ptr->end_time = 0;
} else {
/* Slurm knows nothing about this job,
* may be result of slurmctld cold start */
error("%s: Vestigial buffer for purged job %u",
plugin_type, job_id);
}
return SLURM_SUCCESS;
} else if (!job_ptr &&
((state == BB_STATE_STAGING_IN) ||
(state == BB_STATE_STAGED_IN))) {
struct job_record job_rec;
job_rec.job_id = job_id;
job_rec.user_id = user_id;
bb_ptr = bb_find_alloc_rec(&bb_state, &job_rec);
_stop_stage_in(job_id); /* Purge buffer */
if (bb_ptr) {
bb_ptr->cancelled = true;
bb_ptr->end_time = 0;
} else {
/* Slurm knows nothing about this job,
* may be result of slurmctld cold start */
error("%s: Vestigial buffer for purged job %u",
plugin_type, job_id);
}
return SLURM_SUCCESS;
} else if (!job_ptr) {
error("%s: Vestigial buffer for job ID %u. "
"Clear manually",
plugin_type, job_id);
}
snprintf(local_name, sizeof(local_name), "VestigialJob%u",
job_id);
}
if (job_ptr) {
bb_ptr = bb_find_alloc_rec(&bb_state, job_ptr);
if (bb_ptr == NULL) {
bb_spec = xmalloc(sizeof(bb_job_t));
bb_spec->total_size = _get_bb_size(job_ptr);
bb_ptr = bb_alloc_job_rec(&bb_state, job_ptr, bb_spec);
xfree(bb_spec);
bb_ptr->state = state;
/* bb_ptr->state_time set in bb_alloc_job_rec() */
}
} else {
if ((bb_ptr = _find_bb_name_rec(local_name, user_id)) == NULL) {
bb_ptr = bb_alloc_name_rec(&bb_state, local_name,
user_id);
bb_ptr->size = size;
bb_ptr->state = state;
//FIXME: VESTIGIAL: Use bb_limit_add
// bb_add_user_load(bb_ptr, &bb_state);
return SLURM_SUCCESS;
}
}
bb_ptr->seen_time = time(NULL); /* used to purge defunct recs */
/* UserID set to 0 on some failure modes */
if ((bb_ptr->user_id != user_id) && (user_id != 0)) {
error("%s: User ID mismatch (%u != %u). "
"BB UserID=%u JobID=%u Name=%s",
plugin_type, bb_ptr->user_id, user_id,
bb_ptr->user_id, bb_ptr->job_id, bb_ptr->name);
}
if ((bb_ptr->state == BB_STATE_RUNNING) &&
(state == BB_STATE_STAGED_IN))
state = BB_STATE_RUNNING; /* More precise state info */
if (bb_ptr->state != state) {
/* State is subject to real-time changes */
debug("%s: State changed (%s to %s). "
"BB UserID=%u JobID=%u Name=%s",
plugin_type, bb_state_string(bb_ptr->state),
bb_state_string(state),
bb_ptr->user_id, bb_ptr->job_id, bb_ptr->name);
bb_ptr->state = state;
bb_ptr->state_time = time(NULL);
if (bb_ptr->state == BB_STATE_STAGED_OUT) {
if (bb_ptr->size != 0) {
//FIXME: VESTIGIAL: Use bb_limit_rem
// bb_remove_user_load(bb_ptr, &bb_state);
bb_ptr->size = 0;
}
}
if (bb_ptr->state == BB_STATE_STAGED_IN)
queue_job_scheduler();
}
if ((bb_ptr->state != BB_STATE_STAGED_OUT) && (bb_ptr->size != size)) {
//FIXME: VESTIGIAL: Use bb_limit_rem
// bb_remove_user_load(bb_ptr, &bb_state);
if (size != 0) {
error("%s: Size mismatch (%"PRIu64" != %"PRIu64"). "
"BB UserID=%u JobID=%u Name=%s",
plugin_type, bb_ptr->size, size,
bb_ptr->user_id, bb_ptr->job_id, bb_ptr->name);
}
bb_ptr->size = MAX(bb_ptr->size, size);
//FIXME: VESTIGIAL: Use bb_limit_add
// bb_add_user_load(bb_ptr, &bb_state);
}
return SLURM_SUCCESS;
}
extern int parse_blockreq(void **dest, slurm_parser_enum_t type,
const char *key, const char *value,
const char *line, char **leftover)
{
s_p_options_t block_options[] = {
{"Type", S_P_STRING},
{"32CNBlocks", S_P_UINT16},
{"128CNBlocks", S_P_UINT16},
#ifdef HAVE_BGL
{"Nodecards", S_P_UINT16},
{"Quarters", S_P_UINT16},
{"BlrtsImage", S_P_STRING},
{"LinuxImage", S_P_STRING},
{"RamDiskImage", S_P_STRING},
#else
{"16CNBlocks", S_P_UINT16},
{"64CNBlocks", S_P_UINT16},
{"256CNBlocks", S_P_UINT16},
{"CnloadImage", S_P_STRING},
{"IoloadImage", S_P_STRING},
#endif
{"MloaderImage", S_P_STRING},
{NULL}
};
s_p_hashtbl_t *tbl;
char *tmp = NULL;
select_ba_request_t *n = NULL;
hostlist_t hl = NULL;
tbl = s_p_hashtbl_create(block_options);
s_p_parse_line(tbl, *leftover, leftover);
if (!value) {
return 0;
}
n = xmalloc(sizeof(select_ba_request_t));
hl = hostlist_create(value);
n->save_name = hostlist_ranged_string_xmalloc(hl);
hostlist_destroy(hl);
#ifdef HAVE_BGL
s_p_get_string(&n->blrtsimage, "BlrtsImage", tbl);
s_p_get_string(&n->linuximage, "LinuxImage", tbl);
s_p_get_string(&n->ramdiskimage, "RamDiskImage", tbl);
#else
s_p_get_string(&n->linuximage, "CnloadImage", tbl);
s_p_get_string(&n->ramdiskimage, "IoloadImage", tbl);
#endif
s_p_get_string(&n->mloaderimage, "MloaderImage", tbl);
s_p_get_string(&tmp, "Type", tbl);
if (!tmp || !strcasecmp(tmp,"TORUS"))
n->conn_type[0] = SELECT_TORUS;
else if (!strcasecmp(tmp,"MESH"))
n->conn_type[0] = SELECT_MESH;
else
n->conn_type[0] = SELECT_SMALL;
xfree(tmp);
if (!s_p_get_uint16(&n->small32, "32CNBlocks", tbl)) {
#ifdef HAVE_BGL
s_p_get_uint16(&n->small32, "Nodecards", tbl);
#else
;
#endif
}
if (!s_p_get_uint16(&n->small128, "128CNBlocks", tbl)) {
#ifdef HAVE_BGL
s_p_get_uint16(&n->small128, "Quarters", tbl);
#else
;
#endif
}
#ifndef HAVE_BGL
s_p_get_uint16(&n->small16, "16CNBlocks", tbl);
s_p_get_uint16(&n->small64, "64CNBlocks", tbl);
s_p_get_uint16(&n->small256, "256CNBlocks", tbl);
#endif
s_p_hashtbl_destroy(tbl);
*dest = (void *)n;
return 1;
}
extern int parse_blockreq(void **dest, slurm_parser_enum_t type,
const char *key, const char *value,
const char *line, char **leftover)
{
s_p_options_t block_options[] = {
{"Type", S_P_STRING},
{"32CNBlocks", S_P_UINT16},
{"128CNBlocks", S_P_UINT16},
#ifdef HAVE_BGL
{"Nodecards", S_P_UINT16},
{"Quarters", S_P_UINT16},
{"BlrtsImage", S_P_STRING},
{"LinuxImage", S_P_STRING},
{"RamDiskImage", S_P_STRING},
#else
#ifdef HAVE_BGP
{"16CNBlocks", S_P_UINT16},
{"CnloadImage", S_P_STRING},
{"IoloadImage", S_P_STRING},
#endif
{"64CNBlocks", S_P_UINT16},
{"256CNBlocks", S_P_UINT16},
#endif
{"MloaderImage", S_P_STRING},
{NULL}
};
s_p_hashtbl_t *tbl;
char *tmp = NULL;
select_ba_request_t *n = NULL;
hostlist_t hl = NULL;
tbl = s_p_hashtbl_create(block_options);
s_p_parse_line(tbl, *leftover, leftover);
if (!value) {
return 0;
}
n = xmalloc(sizeof(select_ba_request_t));
hl = hostlist_create(value);
n->save_name = hostlist_ranged_string_xmalloc(hl);
hostlist_destroy(hl);
#ifdef HAVE_BGL
s_p_get_string(&n->blrtsimage, "BlrtsImage", tbl);
s_p_get_string(&n->linuximage, "LinuxImage", tbl);
s_p_get_string(&n->ramdiskimage, "RamDiskImage", tbl);
#elif defined HAVE_BGP
s_p_get_string(&n->linuximage, "CnloadImage", tbl);
s_p_get_string(&n->ramdiskimage, "IoloadImage", tbl);
#endif
s_p_get_string(&n->mloaderimage, "MloaderImage", tbl);
s_p_get_string(&tmp, "Type", tbl);
if (tmp) {
verify_conn_type(tmp, n->conn_type);
xfree(tmp);
}
if (!s_p_get_uint16(&n->small32, "32CNBlocks", tbl)) {
#ifdef HAVE_BGL
s_p_get_uint16(&n->small32, "Nodecards", tbl);
#else
;
#endif
}
if (!s_p_get_uint16(&n->small128, "128CNBlocks", tbl)) {
#ifdef HAVE_BGL
s_p_get_uint16(&n->small128, "Quarters", tbl);
#else
;
#endif
}
#ifndef HAVE_BGL
#ifdef HAVE_BGP
s_p_get_uint16(&n->small16, "16CNBlocks", tbl);
#endif
s_p_get_uint16(&n->small64, "64CNBlocks", tbl);
s_p_get_uint16(&n->small256, "256CNBlocks", tbl);
#endif
if (n->small16 || n->small32 || n->small64
|| n->small128 || n->small256) {
if (n->conn_type[0] < SELECT_SMALL) {
error("Block def on midplane(s) %s is "
"asking for small blocks but given "
"TYPE=%s, setting it to Small",
n->save_name, conn_type_string(n->conn_type[0]));
n->conn_type[0] = SELECT_SMALL;
}
} else {
if (n->conn_type[0] == (uint16_t)NO_VAL) {
n->conn_type[0] = bg_conf->default_conn_type[0];
} else if (n->conn_type[0] >= SELECT_SMALL) {
error("Block def on midplane(s) %s is given "
"TYPE=%s but isn't asking for any small "
"blocks. Giving it %s.",
n->save_name, conn_type_string(n->conn_type[0]),
conn_type_string(
bg_conf->default_conn_type[0]));
n->conn_type[0] = bg_conf->default_conn_type[0];
}
#ifndef HAVE_BG_L_P
int i;
for (i=1; i<SYSTEM_DIMENSIONS; i++) {
if (n->conn_type[i] == (uint16_t)NO_VAL)
n->conn_type[i] = bg_conf->default_conn_type[i];
else if (n->conn_type[i] >= SELECT_SMALL) {
error("Block def on midplane(s) %s dim %d "
"is given TYPE=%s but isn't asking "
"for any small blocks. Giving it %s.",
n->save_name, i,
conn_type_string(n->conn_type[i]),
conn_type_string(
bg_conf->default_conn_type[i]));
n->conn_type[i] = bg_conf->default_conn_type[i];
}
}
#endif
}
s_p_hashtbl_destroy(tbl);
*dest = (void *)n;
return 1;
}
static int _parse_job_info(void **dest, slurm_parser_enum_t type,
const char *key, const char *value,
const char *line, char **leftover)
{
s_p_hashtbl_t *job_tbl;
char *name = NULL, *tmp = NULL, tmp_name[64];
uint32_t job_id = 0, size = 0, user_id = 0;
uint16_t state = 0;
bb_alloc_t *bb_ptr;
struct job_record *job_ptr = NULL;
static s_p_options_t _job_options[] = {
{"JobID",S_P_STRING},
{"Name", S_P_STRING},
{"Size", S_P_STRING},
{"State", S_P_STRING},
{NULL}
};
*dest = NULL;
user_id = atoi(value);
job_tbl = s_p_hashtbl_create(_job_options);
s_p_parse_line(job_tbl, *leftover, leftover);
if (s_p_get_string(&tmp, "JobID", job_tbl))
job_id = atoi(tmp);
s_p_get_string(&name, "Name", job_tbl);
if (s_p_get_string(&tmp, "Size", job_tbl))
size = _get_size_num(tmp);
if (s_p_get_string(&tmp, "State", job_tbl))
state = bb_state_num(tmp);
#if 1
info("%s: JobID:%u Name:%s Size:%u State:%u UserID:%u",
__func__, job_id, name, size, state, user_id);
#endif
if (job_id) {
job_ptr = find_job_record(job_id);
if (!job_ptr) {
error("%s: Vestigial buffer for job ID %u. "
"Clear manually",
plugin_type, job_id);
}
snprintf(tmp_name, sizeof(tmp_name), "VestigialJob%u", job_id);
job_id = 0;
name = tmp_name;
}
if (job_ptr) {
if ((bb_ptr = _find_bb_job_rec(job_ptr)) == NULL) {
bb_ptr = _alloc_bb_job_rec(job_ptr);
bb_ptr->state = state;
}
} else {
if ((bb_ptr = _find_bb_name_rec(name, user_id)) == NULL) {
bb_ptr = _alloc_bb_name_rec(name, user_id);
bb_ptr->size = size;
bb_ptr->state = state;
return SLURM_SUCCESS;
}
}
if (bb_ptr->user_id != user_id) {
error("%s: User ID mismatch (%u != %u). "
"BB UserID=%u JobID=%u Name=%s",
plugin_type, bb_ptr->user_id, user_id,
bb_ptr->user_id, bb_ptr->job_id, bb_ptr->name);
}
if (bb_ptr->size != size) {
error("%s: Size mismatch (%u != %u). "
"BB UserID=%u JobID=%u Name=%s",
plugin_type, bb_ptr->size, size,
bb_ptr->user_id, bb_ptr->job_id, bb_ptr->name);
bb_ptr->size = MAX(bb_ptr->size, size);
}
if (bb_ptr->state != state) {
/* State is subject to real-time changes */
debug("%s: State mismatch (%s != %s). "
"BB UserID=%u JobID=%u Name=%s",
plugin_type, bb_state_string(bb_ptr->state),
bb_state_string(state),
bb_ptr->user_id, bb_ptr->job_id, bb_ptr->name);
}
return SLURM_SUCCESS;
}