/* Translate a job's feature specification to node boot options
* RET node boot options, must be xfreed */
extern char *node_features_g_job_xlate(char *job_features)
{
DEF_TIMERS;
char *node_features = NULL, *tmp_str;
int i;
START_TIMER;
(void) node_features_g_init();
slurm_mutex_lock(&g_context_lock);
for (i = 0; i < g_context_cnt; i++) {
tmp_str = (*(ops[i].job_xlate))(job_features);
if (tmp_str) {
if (node_features) {
xstrfmtcat(node_features, ",%s", tmp_str);
xfree(tmp_str);
} else {
node_features = tmp_str;
}
}
}
slurm_mutex_unlock(&g_context_lock);
END_TIMER2("node_features_g_job_xlate");
return node_features;
}
/* Translate a node's feature specification by replacing any features associated
* with this plugin in the original value with the new values, preserving any
* features that are not associated with this plugin
* RET node's new merged features, must be xfreed */
extern char *node_features_g_node_xlate(char *new_features, char *orig_features)
{
DEF_TIMERS;
char *new_value = NULL, *tmp_str;
int i;
START_TIMER;
(void) node_features_g_init();
slurm_mutex_lock(&g_context_lock);
for (i = 0; i < g_context_cnt; i++) {
if (new_value)
tmp_str = new_value;
else if (orig_features)
tmp_str = xstrdup(orig_features);
else
tmp_str = NULL;
new_value = (*(ops[i].node_xlate))(new_features, tmp_str);
xfree(tmp_str);
}
slurm_mutex_unlock(&g_context_lock);
END_TIMER2("node_features_g_node_xlate");
return new_value;
}
/*
* Execute the job_submit() function in each job submit plugin.
* If any plugin function returns anything other than SLURM_SUCCESS
* then stop and forward it's return value.
* IN job_desc - Job request specification
* IN submit_uid - User issuing job submit request
* OUT err_msg - Custom error message to the user, caller to xfree results
*/
extern int job_submit_plugin_submit(struct job_descriptor *job_desc,
uint32_t submit_uid, char **err_msg)
{
DEF_TIMERS;
int i, rc;
xassert(verify_lock(CONF_LOCK, READ_LOCK));
xassert(verify_lock(JOB_LOCK, READ_LOCK));
xassert(verify_lock(NODE_LOCK, READ_LOCK));
xassert(verify_lock(PART_LOCK, READ_LOCK));
START_TIMER;
rc = job_submit_plugin_init();
slurm_mutex_lock(&g_context_lock);
/* NOTE: On function entry read locks are set on config, job, node and
* partition structures. Do not attempt to unlock them and then
* lock again (say with a write lock) since doing so will trigger
* a deadlock with the g_context_lock above. */
for (i = 0; ((i < g_context_cnt) && (rc == SLURM_SUCCESS)); i++)
rc = (*(ops[i].submit))(job_desc, submit_uid, err_msg);
slurm_mutex_unlock(&g_context_lock);
END_TIMER2("job_submit_plugin_submit");
return rc;
}
/*
* load_part_uid_allow_list - reload the allow_uid list of partitions
* if required (updated group file or force set)
* IN force - if set then always reload the allow_uid list
*/
void load_part_uid_allow_list(int force)
{
static time_t last_update_time;
time_t temp_time;
ListIterator part_iterator;
struct part_record *part_ptr;
DEF_TIMERS;
START_TIMER;
temp_time = _get_group_tlm();
if ((force == 0) && (temp_time == last_update_time))
return;
debug("Updating partition uid access list");
last_update_time = temp_time;
last_part_update = time(NULL);
part_iterator = list_iterator_create(part_list);
while ((part_ptr = (struct part_record *) list_next(part_iterator))) {
xfree(part_ptr->allow_uids);
part_ptr->allow_uids =
_get_groups_members(part_ptr->allow_groups);
}
clear_group_cache();
list_iterator_destroy(part_iterator);
END_TIMER2("load_part_uid_allow_list");
}
/* Get node features plugin configuration */
extern List node_features_g_get_config(void)
{
DEF_TIMERS;
int i, rc;
List conf_list = NULL;
config_plugin_params_t *p;
START_TIMER;
rc = node_features_g_init();
if (g_context_cnt > 0)
conf_list = list_create(destroy_config_plugin_params);
slurm_mutex_lock(&g_context_lock);
for (i = 0; ((i < g_context_cnt) && (rc == SLURM_SUCCESS)); i++) {
p = xmalloc(sizeof(config_plugin_params_t));
p->key_pairs = list_create(destroy_config_key_pair);
(*(ops[i].get_config))(p);
if (!p->name)
destroy_config_plugin_params(p);
else
list_append(conf_list, p);
}
slurm_mutex_unlock(&g_context_lock);
END_TIMER2("node_features_g_get_config");
return conf_list;
}
/* Get this node's current and available MCDRAM and NUMA settings from BIOS.
* avail_modes IN/OUT - available modes, must be xfreed
* current_mode IN/OUT - current modes, must be xfreed */
extern void node_features_g_node_state(char **avail_modes, char **current_mode)
{
DEF_TIMERS;
int i;
START_TIMER;
(void) node_features_g_init();
slurm_mutex_lock(&g_context_lock);
for (i = 0; i < g_context_cnt; i++) {
(*(ops[i].node_state))(avail_modes, current_mode);
}
slurm_mutex_unlock(&g_context_lock);
END_TIMER2("node_features_g_node_state");
}
/* Perform set up for step launch
* mem_sort IN - Trigger sort of memory pages (KNL zonesort)
* numa_bitmap IN - NUMA nodes allocated to this job */
extern void node_features_g_step_config(bool mem_sort, bitstr_t *numa_bitmap)
{
DEF_TIMERS;
int i;
START_TIMER;
if (node_features_g_init() != SLURM_SUCCESS)
return;
slurm_mutex_lock(&g_context_lock);
for (i = 0; i < g_context_cnt; i++)
(*(ops[i].step_config))(mem_sort, numa_bitmap);
slurm_mutex_unlock(&g_context_lock);
END_TIMER2("node_features_g_step_config");
}
/* Test if a job's feature specification is valid */
extern int node_features_g_job_valid(char *job_features)
{
DEF_TIMERS;
int i, rc;
START_TIMER;
rc = node_features_g_init();
slurm_mutex_lock(&g_context_lock);
for (i = 0; ((i < g_context_cnt) && (rc == SLURM_SUCCESS)); i++)
rc = (*(ops[i].job_valid))(job_features);
slurm_mutex_unlock(&g_context_lock);
END_TIMER2("node_features_g_job_valid");
return rc;
}
/* Update active and available features on specified nodes, sets features on
* all nodes is node_list is NULL */
extern int node_features_g_get_node(char *node_list)
{
DEF_TIMERS;
int i, rc;
START_TIMER;
rc = node_features_g_init();
slurm_mutex_lock(&g_context_lock);
for (i = 0; ((i < g_context_cnt) && (rc == SLURM_SUCCESS)); i++)
rc = (*(ops[i].get_node))(node_list);
slurm_mutex_unlock(&g_context_lock);
END_TIMER2("node_features_g_get_node");
return rc;
}
/* Reset plugin configuration information */
extern int node_features_g_reconfig(void)
{
DEF_TIMERS;
int i, rc;
START_TIMER;
rc = node_features_g_init();
slurm_mutex_lock(&g_context_lock);
for (i = 0; ((i < g_context_cnt) && (rc == SLURM_SUCCESS)); i++)
rc = (*(ops[i].reconfig))();
slurm_mutex_unlock(&g_context_lock);
END_TIMER2("node_features_g_reconfig");
return rc;
}
/* Set's the node's active features based upon job constraints.
* NOTE: Executed by the slurmd daemon.
* IN active_features - New active features
* RET error code */
extern int node_features_g_node_set(char *active_features)
{
DEF_TIMERS;
int i, rc = SLURM_SUCCESS;
START_TIMER;
(void) node_features_g_init();
slurm_mutex_lock(&g_context_lock);
for (i = 0; ((i < g_context_cnt) && (rc == SLURM_SUCCESS)); i++) {
rc = (*(ops[i].node_set))(active_features);
}
slurm_mutex_unlock(&g_context_lock);
END_TIMER2("node_features_g_node_set");
return rc;
}
/* Return TRUE if this (one) feature name is under this plugin's control */
extern bool node_features_g_changeable_feature(char *feature)
{
DEF_TIMERS;
int i;
bool changeable = false;
START_TIMER;
(void) node_features_g_init();
slurm_mutex_lock(&g_context_lock);
for (i = 0; ((i < g_context_cnt) && !changeable); i++)
changeable = (*(ops[i].changeable_feature))(feature);
slurm_mutex_unlock(&g_context_lock);
END_TIMER2("node_features_g_reconfig");
return changeable;
}
/* Determine if the specified user can modify the currently available node
* features */
extern bool node_features_g_user_update(uid_t uid)
{
DEF_TIMERS;
bool result = true;
int i;
START_TIMER;
(void) node_features_g_init();
slurm_mutex_lock(&g_context_lock);
for (i = 0; ((i < g_context_cnt) && (result == true)); i++) {
result = (*(ops[i].user_update))(uid);
}
slurm_mutex_unlock(&g_context_lock);
END_TIMER2("node_features_g_user_update");
return result;
}
/* Return estimated reboot time, in seconds */
extern uint32_t node_features_g_boot_time(void)
{
DEF_TIMERS;
uint32_t boot_time = 0;
int i;
START_TIMER;
(void) node_features_g_init();
slurm_mutex_lock(&g_context_lock);
for (i = 0; i < g_context_cnt; i++) {
boot_time = MAX(boot_time, (*(ops[i].boot_time))());
}
slurm_mutex_unlock(&g_context_lock);
END_TIMER2("node_features_g_user_update");
return boot_time;
}
/*
* Execute the job_modify() function in each job submit plugin.
* If any plugin function returns anything other than SLURM_SUCCESS
* then stop and forward it's return value.
*/
extern int job_submit_plugin_modify(struct job_descriptor *job_desc,
struct job_record *job_ptr,
uint32_t submit_uid)
{
DEF_TIMERS;
int i, rc;
START_TIMER;
rc = job_submit_plugin_init();
slurm_mutex_lock(&g_context_lock);
for (i = 0; ((i < g_context_cnt) && (rc == SLURM_SUCCESS)); i++)
rc = (*(ops[i].modify))(job_desc, job_ptr, submit_uid);
slurm_mutex_unlock(&g_context_lock);
END_TIMER2("job_submit_plugin_modify");
return rc;
}
/* Return true if the plugin requires RebootProgram for booting nodes */
extern bool node_features_g_node_reboot(void)
{
DEF_TIMERS;
bool node_reboot = false;
int i;
START_TIMER;
(void) node_features_g_init();
slurm_mutex_lock(&g_context_lock);
for (i = 0; i < g_context_cnt; i++) {
node_reboot = (*(ops[i].node_reboot))();
if (node_reboot)
break;
}
slurm_mutex_unlock(&g_context_lock);
END_TIMER2("node_features_g_node_reboot");
return node_reboot;
}
/* Return bitmap of KNL nodes, NULL if none identified */
extern bitstr_t *node_features_g_get_node_bitmap(void)
{
DEF_TIMERS;
bitstr_t *node_bitmap = NULL;
int i;
START_TIMER;
(void) node_features_g_init();
slurm_mutex_lock(&g_context_lock);
for (i = 0; i < g_context_cnt; i++) {
node_bitmap = (*(ops[i].get_node_bitmap))();
if (node_bitmap)
break;
}
slurm_mutex_unlock(&g_context_lock);
END_TIMER2("node_features_g_get_node_bitmap");
return node_bitmap;
}
/*
* Return TRUE if the specified node update request is valid with respect
* to features changes (i.e. don't permit a non-KNL node to set KNL features).
*
* node_ptr IN - Pointer to struct node_record record
* update_node_msg IN - Pointer to update request
*/
extern bool node_features_g_node_update_valid(void *node_ptr,
update_node_msg_t *update_node_msg)
{
DEF_TIMERS;
bool update_valid = true;
int i;
START_TIMER;
(void) node_features_g_init();
slurm_mutex_lock(&g_context_lock);
for (i = 0; i < g_context_cnt; i++) {
update_valid = (*(ops[i].node_update_valid))(node_ptr,
update_node_msg);
if (!update_valid)
break;
}
slurm_mutex_unlock(&g_context_lock);
END_TIMER2("node_features_g_node_update_valid");
return update_valid;
}
/*
* event_notify - Notify Moab of some event
* event_code IN - message code to send Moab
* 1234 - job state change
* 1235 - partition state change
* desc IN - event description
* RET 0 on success, -1 on failure
*/
extern int event_notify(int event_code, char *desc)
{
time_t now = time(NULL);
int rc = 0, retry = 2;
char *event_msg;
DEF_TIMERS;
START_TIMER;
if (e_port == 0) {
/* Event notification disabled */
return 0;
}
if (event_code == 1234) { /* job change */
if (job_aggregation_time
&& (difftime(now, last_notify_time) < job_aggregation_time)) {
debug("wiki event notification already sent recently");
return 0;
}
event_msg = "1234";
} else if (event_code == 1235) { /* configuration change */
event_msg = "1235";
} else {
error("event_notify: invalid event code: %d", event_code);
return -1;
}
slurm_mutex_lock(&event_mutex);
while (retry) {
if ((event_fd == -1) && ((rc = _open_fd(now)) == -1)) {
/* Can't even open socket.
* Don't retry again for a while (2 mins)
* to avoid long delays from ETIMEDOUT */
last_notify_time = now + 120;
break;
}
if (write(event_fd, event_msg, (strlen(event_msg) + 1)) > 0) {
verbose("wiki event_notification sent: %s", desc);
last_notify_time = now;
rc = 0;
/* Dave Jackson says to leave the connection
* open, but Moab isn't. Without the _close_fd()
* here, the next write() generates a broken pipe
* error. Just remove the _close_fd() and this
* comment when Moab maintains the connection. */
_close_fd();
break; /* success */
}
error("wiki event notification failure: %m");
rc = -1;
retry--;
if ((errno == EAGAIN) || (errno == EINTR))
continue;
_close_fd();
if (errno == EPIPE) {
/* If Moab closed the socket we get an EPIPE,
* retry once */
continue;
} else {
break;
}
}
slurm_mutex_unlock(&event_mutex);
END_TIMER2("event_notify");
return rc;
}
/* rollup usage for one cluster */
static int
_cluster_rollup_usage(pgsql_conn_t *pg_conn, char *cluster,
time_t sent_start, time_t sent_end,
uint16_t archive_data)
{
DEF_VARS;
int rc = SLURM_SUCCESS;
time_t last_hour = sent_start;
time_t last_day = sent_start;
time_t last_month = sent_start;
time_t start_time = 0;
time_t end_time = 0;
time_t my_time = sent_end;
struct tm start_tm, end_tm;
DEF_TIMERS;
char *ru_fields = "hourly_rollup, daily_rollup, monthly_rollup";
enum {
F_HOUR,
F_DAY,
F_MONTH,
F_COUNT
};
if (!sent_start) {
query = xstrdup_printf("SELECT %s FROM %s.%s LIMIT 1",
ru_fields, cluster, last_ran_table);
result = DEF_QUERY_RET;
if (!result)
return SLURM_ERROR;
if (PQntuples(result)) {
last_hour = atoi(PG_VAL(F_HOUR));
last_day = atoi(PG_VAL(F_DAY));
last_month = atoi(PG_VAL(F_MONTH));
PQclear(result);
} else {
time_t now = time(NULL);
PQclear(result);
query = xstrdup_printf("SELECT %s.init_last_ran(%ld);",
cluster, now);
result = DEF_QUERY_RET;
if (!result)
return SLURM_ERROR;
last_hour = last_day = last_month =
atoi(PG_VAL(0));
PQclear(result);
if (last_hour < 0) {
debug("cluster %s not registered, "
"not doing rollup", cluster);
return SLURM_SUCCESS;
}
}
}
if (!my_time)
my_time = time(NULL);
if (!localtime_r(&last_hour, &start_tm)) {
error("Couldn't get localtime from hour start %ld", last_hour);
return SLURM_ERROR;
}
if (!localtime_r(&my_time, &end_tm)) {
error("Couldn't get localtime from hour end %ld", my_time);
return SLURM_ERROR;
}
/* below and anywhere in a rollup plugin when dealing with
* epoch times we need to set the tm_isdst = -1 so we don't
* have to worry about the time changes. Not setting it to -1
* will cause problems in the day and month with the date change.
*/
/* align to hour boundary */
start_tm.tm_sec = 0;
start_tm.tm_min = 0;
start_tm.tm_isdst = -1;
start_time = mktime(&start_tm);
end_tm.tm_sec = 0;
end_tm.tm_min = 0;
end_tm.tm_isdst = -1;
end_time = mktime(&end_tm);
/* info("hour start %s", ctime(&start_time)); */
/* info("hour end %s", ctime(&end_time)); */
/* info("diff is %d", end_time-start_time); */
//slurm_mutex_lock(&rollup_lock);
global_last_rollup = end_time;
//slurm_mutex_unlock(&rollup_lock);
if (end_time-start_time > 0) {
START_TIMER;
if ((rc = pgsql_hourly_rollup(pg_conn, cluster, start_time, end_time))
!= SLURM_SUCCESS)
return rc;
END_TIMER3("hourly_rollup", 5000000);
/* If we have a sent_end do not update the last_run_table */
if (!sent_end)
query = xstrdup_printf(
"UPDATE %s.%s SET hourly_rollup=%ld",
cluster, last_ran_table, end_time);
} else {
debug2("no need to run this hour %ld <= %ld",
end_time, start_time);
}
if (!localtime_r(&last_day, &start_tm)) {
error("Couldn't get localtime from day %ld", last_day);
return SLURM_ERROR;
}
/* align to day boundary */
start_tm.tm_sec = 0;
start_tm.tm_min = 0;
start_tm.tm_hour = 0;
start_tm.tm_isdst = -1;
start_time = mktime(&start_tm);
end_tm.tm_hour = 0;
end_tm.tm_isdst = -1;
end_time = mktime(&end_tm);
/* info("day start %s", ctime(&start_time)); */
/* info("day end %s", ctime(&end_time)); */
/* info("diff is %d", end_time-start_time); */
if (end_time-start_time > 0) {
START_TIMER;
if ((rc = pgsql_daily_rollup(pg_conn, cluster, start_time,
end_time, archive_data))
!= SLURM_SUCCESS)
return rc;
END_TIMER2("daily_rollup");
if (query && !sent_end)
xstrfmtcat(query, ", daily_rollup=%ld", (long)end_time);
else if (!sent_end)
query = xstrdup_printf(
"UPDATE %s.%s SET daily_rollup=%ld",
cluster, last_ran_table, (long)end_time);
} else {
debug2("no need to run this day %ld <= %ld",
(long)end_time, (long)start_time);
}
if (!localtime_r(&last_month, &start_tm)) {
error("Couldn't get localtime from month %ld", last_month);
return SLURM_ERROR;
}
/* align to month boundary */
start_tm.tm_sec = 0;
start_tm.tm_min = 0;
start_tm.tm_hour = 0;
start_tm.tm_mday = 1;
start_tm.tm_isdst = -1;
start_time = mktime(&start_tm);
end_time = mktime(&end_tm);
end_tm.tm_sec = 0;
end_tm.tm_min = 0;
end_tm.tm_hour = 0;
end_tm.tm_mday = 1;
end_tm.tm_isdst = -1;
end_time = mktime(&end_tm);
/* info("month start %s", ctime(&start_time)); */
/* info("month end %s", ctime(&end_time)); */
/* info("diff is %d", end_time-start_time); */
if (end_time-start_time > 0) {
START_TIMER;
if ((rc = pgsql_monthly_rollup(pg_conn, cluster, start_time,
end_time, archive_data))
!= SLURM_SUCCESS)
return rc;
END_TIMER2("monthly_rollup");
if (query && !sent_end)
xstrfmtcat(query, ", monthly_rollup=%ld",
(long)end_time);
else if (!sent_end)
query = xstrdup_printf(
"UPDATE %s.%s SET monthly_rollup=%ld",
cluster, last_ran_table, (long)end_time);
} else {
debug2("no need to run this month %ld <= %ld",
(long)end_time, (long)start_time);
}
if (query) {
rc = DEF_QUERY_RET_RC;
}
return rc;
}
/*****************************************************************************\
* Parse, process and respond to a request
\*****************************************************************************/
static void _proc_msg(slurm_fd_t new_fd, char *msg)
{
DEF_TIMERS;
char *req, *cmd_ptr, *msg_type = NULL;
char response[128];
if (new_fd < 0)
return;
START_TIMER;
if (!msg) {
err_code = -300;
err_msg = "NULL request message";
error("wiki: NULL request message");
goto resp_msg;
}
if (_parse_msg(msg, &req) != 0)
goto resp_msg;
cmd_ptr = strstr(req, "CMD=");
if (cmd_ptr == NULL) {
err_code = -300;
err_msg = "request lacks CMD";
error("wiki: request lacks CMD");
goto resp_msg;
}
cmd_ptr +=4;
err_code = 0;
if (strncmp(cmd_ptr, "GETJOBS", 7) == 0) {
msg_type = "wiki:GETJOBS";
if (!get_jobs(cmd_ptr, &err_code, &err_msg))
goto free_resp_msg;
} else if (strncmp(cmd_ptr, "GETNODES", 8) == 0) {
msg_type = "wiki:GETNODES";
if (!get_nodes(cmd_ptr, &err_code, &err_msg))
goto free_resp_msg;
} else if (strncmp(cmd_ptr, "STARTJOB", 8) == 0) {
msg_type = "wiki:STARTJOB";
start_job(cmd_ptr, &err_code, &err_msg);
} else if (strncmp(cmd_ptr, "CANCELJOB", 9) == 0) {
msg_type = "wiki:CANCELJOB";
cancel_job(cmd_ptr, &err_code, &err_msg);
} else if (strncmp(cmd_ptr, "SUSPENDJOB", 10) == 0) {
msg_type = "wiki:SUSPENDJOB";
suspend_job(cmd_ptr, &err_code, &err_msg);
} else if (strncmp(cmd_ptr, "RESUMEJOB", 9) == 0) {
msg_type = "wiki:RESUMEJOB";
resume_job(cmd_ptr, &err_code, &err_msg);
} else if (strncmp(cmd_ptr, "MODIFYJOB", 9) == 0) {
msg_type = "wiki:MODIFYJOB";
job_modify_wiki(cmd_ptr, &err_code, &err_msg);
} else {
err_code = -300;
err_msg = "unsupported request type";
error("wiki: unrecognized request type: %s", req);
}
END_TIMER2(msg_type);
resp_msg:
snprintf(response, sizeof(response),
"SC=%d RESPONSE=%s", err_code, err_msg);
_send_reply(new_fd, response);
return;
free_resp_msg:
/* Message is pre-formatted by get_jobs and get_nodes
* ONLY if no error. Send message and xfree the buffer. */
_send_reply(new_fd, err_msg);
xfree(err_msg);
return;
}
/* NOTE: Insure that mysql_conn->lock is NOT set on function entry */
static int _mysql_make_table_current(mysql_conn_t *mysql_conn, char *table_name,
storage_field_t *fields, char *ending)
{
char *query = NULL;
char *correct_query = NULL;
MYSQL_RES *result = NULL;
MYSQL_ROW row;
int i = 0;
List columns = NULL;
ListIterator itr = NULL;
char *col = NULL;
int adding = 0;
int run_update = 0;
char *primary_key = NULL;
char *unique_index = NULL;
int old_primary = 0;
char *old_index = NULL;
char *temp = NULL, *temp2 = NULL;
List keys_list = NULL;
db_key_t *db_key = NULL;
DEF_TIMERS;
/* figure out the unique keys in the table */
query = xstrdup_printf("show index from %s where non_unique=0",
table_name);
if (!(result = mysql_db_query_ret(mysql_conn, query, 0))) {
xfree(query);
return SLURM_ERROR;
}
xfree(query);
while ((row = mysql_fetch_row(result))) {
// row[2] is the key name
if (!strcasecmp(row[2], "PRIMARY"))
old_primary = 1;
else if (!old_index)
old_index = xstrdup(row[2]);
}
mysql_free_result(result);
/* figure out the non-unique keys in the table */
query = xstrdup_printf("show index from %s where non_unique=1",
table_name);
if (!(result = mysql_db_query_ret(mysql_conn, query, 0))) {
xfree(query);
return SLURM_ERROR;
}
xfree(query);
itr = NULL;
keys_list = list_create(_destroy_db_key);
while ((row = mysql_fetch_row(result))) {
if (!itr)
itr = list_iterator_create(keys_list);
else
list_iterator_reset(itr);
while ((db_key = list_next(itr))) {
if (!strcmp(db_key->name, row[2]))
break;
}
if (db_key) {
xstrfmtcat(db_key->columns, ", %s", row[4]);
} else {
db_key = xmalloc(sizeof(db_key_t));
db_key->name = xstrdup(row[2]); // name
db_key->columns = xstrdup(row[4]); // column name
list_append(keys_list, db_key); // don't use list_push
}
}
mysql_free_result(result);
if (itr) {
list_iterator_destroy(itr);
itr = NULL;
}
/* figure out the existing columns in the table */
query = xstrdup_printf("show columns from %s", table_name);
if (!(result = mysql_db_query_ret(mysql_conn, query, 0))) {
xfree(query);
xfree(old_index);
FREE_NULL_LIST(keys_list);
return SLURM_ERROR;
}
xfree(query);
columns = list_create(slurm_destroy_char);
while ((row = mysql_fetch_row(result))) {
col = xstrdup(row[0]); //Field
list_append(columns, col);
}
mysql_free_result(result);
itr = list_iterator_create(columns);
query = xstrdup_printf("alter ignore table %s", table_name);
correct_query = xstrdup_printf("alter ignore table %s", table_name);
START_TIMER;
while (fields[i].name) {
int found = 0;
list_iterator_reset(itr);
while ((col = list_next(itr))) {
if (!strcmp(col, fields[i].name)) {
xstrfmtcat(query, " modify `%s` %s,",
fields[i].name,
fields[i].options);
xstrfmtcat(correct_query, " modify `%s` %s,",
fields[i].name,
fields[i].options);
list_delete_item(itr);
found = 1;
break;
}
}
if (!found) {
if (i) {
info("adding column %s after %s in table %s",
fields[i].name,
fields[i-1].name,
table_name);
xstrfmtcat(query, " add `%s` %s after %s,",
fields[i].name,
fields[i].options,
fields[i-1].name);
xstrfmtcat(correct_query, " modify `%s` %s,",
fields[i].name,
fields[i].options);
} else {
info("adding column %s at the beginning "
"of table %s",
fields[i].name,
table_name);
xstrfmtcat(query, " add `%s` %s first,",
fields[i].name,
fields[i].options);
xstrfmtcat(correct_query, " modify `%s` %s,",
fields[i].name,
fields[i].options);
}
adding = 1;
}
i++;
}
list_iterator_reset(itr);
while ((col = list_next(itr))) {
adding = 1;
info("dropping column %s from table %s", col, table_name);
xstrfmtcat(query, " drop %s,", col);
}
list_iterator_destroy(itr);
list_destroy(columns);
if ((temp = strstr(ending, "primary key ("))) {
int open = 0, close =0;
int end = 0;
while (temp[end++]) {
if (temp[end] == '(')
open++;
else if (temp[end] == ')')
close++;
else
continue;
if (open == close)
break;
}
if (temp[end]) {
end++;
primary_key = xstrndup(temp, end);
if (old_primary) {
xstrcat(query, " drop primary key,");
xstrcat(correct_query, " drop primary key,");
}
xstrfmtcat(query, " add %s,", primary_key);
xstrfmtcat(correct_query, " add %s,", primary_key);
xfree(primary_key);
}
}
if ((temp = strstr(ending, "unique index ("))) {
int open = 0, close =0;
int end = 0;
while (temp[end++]) {
if (temp[end] == '(')
open++;
else if (temp[end] == ')')
close++;
else
continue;
if (open == close)
break;
}
if (temp[end]) {
end++;
unique_index = xstrndup(temp, end);
if (old_index) {
xstrfmtcat(query, " drop index %s,",
old_index);
xstrfmtcat(correct_query, " drop index %s,",
old_index);
}
xstrfmtcat(query, " add %s,", unique_index);
xstrfmtcat(correct_query, " add %s,", unique_index);
xfree(unique_index);
}
}
xfree(old_index);
temp2 = ending;
itr = list_iterator_create(keys_list);
while ((temp = strstr(temp2, ", key "))) {
int open = 0, close = 0, name_end = 0;
int end = 5;
char *new_key_name = NULL, *new_key = NULL;
while (temp[end++]) {
if (!name_end && (temp[end] == ' ')) {
name_end = end;
continue;
} else if (temp[end] == '(') {
open++;
if (!name_end)
name_end = end;
} else if (temp[end] == ')')
close++;
else
continue;
if (open == close)
break;
}
if (temp[end]) {
end++;
new_key_name = xstrndup(temp+6, name_end-6);
new_key = xstrndup(temp+2, end-2); // skip ', '
while ((db_key = list_next(itr))) {
if (!strcmp(db_key->name, new_key_name)) {
list_remove(itr);
break;
}
}
list_iterator_reset(itr);
if (db_key) {
xstrfmtcat(query,
" drop key %s,", db_key->name);
xstrfmtcat(correct_query,
" drop key %s,", db_key->name);
_destroy_db_key(db_key);
} else
info("adding %s to table %s",
new_key, table_name);
xstrfmtcat(query, " add %s,", new_key);
xstrfmtcat(correct_query, " add %s,", new_key);
xfree(new_key);
xfree(new_key_name);
}
temp2 = temp + end;
}
/* flush extra (old) keys */
while ((db_key = list_next(itr))) {
info("dropping key %s from table %s", db_key->name, table_name);
xstrfmtcat(query, " drop key %s,", db_key->name);
}
list_iterator_destroy(itr);
list_destroy(keys_list);
query[strlen(query)-1] = ';';
correct_query[strlen(correct_query)-1] = ';';
//info("%d query\n%s", __LINE__, query);
/* see if we have already done this definition */
if (!adding) {
char *quoted = slurm_add_slash_to_quotes(query);
char *query2 = xstrdup_printf("select table_name from "
"%s where definition='%s'",
table_defs_table, quoted);
MYSQL_RES *result = NULL;
MYSQL_ROW row;
xfree(quoted);
run_update = 1;
if ((result = mysql_db_query_ret(mysql_conn, query2, 0))) {
if ((row = mysql_fetch_row(result)))
run_update = 0;
mysql_free_result(result);
}
xfree(query2);
if (run_update) {
run_update = 2;
query2 = xstrdup_printf("select table_name from "
"%s where table_name='%s'",
table_defs_table, table_name);
if ((result = mysql_db_query_ret(
mysql_conn, query2, 0))) {
if ((row = mysql_fetch_row(result)))
run_update = 1;
mysql_free_result(result);
}
xfree(query2);
}
}
/* if something has changed run the alter line */
if (run_update || adding) {
time_t now = time(NULL);
char *query2 = NULL;
char *quoted = NULL;
if (run_update == 2)
debug4("Table %s doesn't exist, adding", table_name);
else
debug("Table %s has changed. Updating...", table_name);
if (mysql_db_query(mysql_conn, query)) {
xfree(query);
return SLURM_ERROR;
}
quoted = slurm_add_slash_to_quotes(correct_query);
query2 = xstrdup_printf("insert into %s (creation_time, "
"mod_time, table_name, definition) "
"values (%ld, %ld, '%s', '%s') "
"on duplicate key update "
"definition='%s', mod_time=%ld;",
table_defs_table, now, now,
table_name, quoted,
quoted, now);
xfree(quoted);
if (mysql_db_query(mysql_conn, query2)) {
xfree(query2);
return SLURM_ERROR;
}
xfree(query2);
}
xfree(query);
xfree(correct_query);
query = xstrdup_printf("make table current %s", table_name);
END_TIMER2(query);
xfree(query);
return SLURM_SUCCESS;
}
/*****************************************************************************\
* Parse, process and respond to a request
\*****************************************************************************/
static void _proc_msg(slurm_fd_t new_fd, char *msg)
{
DEF_TIMERS;
char *req, *cmd_ptr, *msg_type = NULL;
char response[128];
if (new_fd < 0)
return;
START_TIMER;
if (!msg) {
err_code = -300;
err_msg = "NULL request message";
error("wiki: NULL request message");
goto resp_msg;
}
if (_parse_msg(msg, &req) != 0)
goto resp_msg;
cmd_ptr = strstr(req, "CMD=");
if (cmd_ptr == NULL) {
err_code = -300;
err_msg = "request lacks CMD";
error("wiki: request lacks CMD");
goto resp_msg;
}
cmd_ptr +=4;
err_code = 0;
if (strncmp(cmd_ptr, "GETJOBS", 7) == 0) {
msg_type = "wiki:GETJOBS";
if (!get_jobs(cmd_ptr, &err_code, &err_msg))
goto free_resp_msg;
} else if (strncmp(cmd_ptr, "GETNODES", 8) == 0) {
msg_type = "wiki:GETNODES";
if (!get_nodes(cmd_ptr, &err_code, &err_msg))
goto free_resp_msg;
} else if (strncmp(cmd_ptr, "STARTJOB", 8) == 0) {
msg_type = "wiki:STARTJOB";
start_job(cmd_ptr, &err_code, &err_msg);
} else if (strncmp(cmd_ptr, "CANCELJOB", 9) == 0) {
msg_type = "wiki:CANCELJOB";
cancel_job(cmd_ptr, &err_code, &err_msg);
} else if (strncmp(cmd_ptr, "REQUEUEJOB", 10) == 0) {
msg_type = "wiki:REQUEUEJOB";
job_requeue_wiki(cmd_ptr, &err_code, &err_msg);
} else if (strncmp(cmd_ptr, "SUSPENDJOB", 10) == 0) {
msg_type = "wiki:SUSPENDJOB";
suspend_job(cmd_ptr, &err_code, &err_msg);
} else if (strncmp(cmd_ptr, "RESUMEJOB", 9) == 0) {
msg_type = "wiki:RESUMEJOB";
resume_job(cmd_ptr, &err_code, &err_msg);
} else if (strncmp(cmd_ptr, "JOBADDTASK", 10) == 0) {
msg_type = "wiki:JOBADDTASK";
job_add_task(cmd_ptr, &err_code, &err_msg);
} else if (strncmp(cmd_ptr, "JOBRELEASETASK", 14) == 0) {
msg_type = "wiki:JOBRELEASETASK";
job_release_task(cmd_ptr, &err_code, &err_msg);
} else if (strncmp(cmd_ptr, "JOBWILLRUN", 10) == 0) {
msg_type = "wiki:JOBWILLRUN";
if (strstr(cmd_ptr, "NODES=")) {
/* Updated format input and output */
if (!job_will_run2(cmd_ptr, &err_code, &err_msg))
goto free_resp_msg;
} else {
if (!job_will_run(cmd_ptr, &err_code, &err_msg))
goto free_resp_msg;
}
} else if (strncmp(cmd_ptr, "MODIFYJOB", 9) == 0) {
msg_type = "wiki:MODIFYJOB";
job_modify_wiki(cmd_ptr, &err_code, &err_msg);
} else if (strncmp(cmd_ptr, "NOTIFYJOB", 9) == 0) {
msg_type = "wiki:NOTIFYJOB";
job_notify_wiki(cmd_ptr, &err_code, &err_msg);
} else if (strncmp(cmd_ptr, "SIGNALJOB", 9) == 0) {
msg_type = "wiki:SIGNALJOB";
job_signal_wiki(cmd_ptr, &err_code, &err_msg);
} else if (strncmp(cmd_ptr, "INITIALIZE", 10) == 0) {
msg_type = "wiki:INITIALIZE";
initialize_wiki(cmd_ptr, &err_code, &err_msg);
} else {
err_code = -300;
err_msg = "unsupported request type";
error("wiki: unrecognized request type: %s", req);
}
END_TIMER2(msg_type);
resp_msg:
snprintf(response, sizeof(response),
"SC=%d RESPONSE=%s", err_code, err_msg);
_send_reply(new_fd, response);
return;
free_resp_msg:
/* Message is pre-formatted by get_jobs and get_nodes
* ONLY if no error. Send message and xfree the buffer. */
_send_reply(new_fd, err_msg);
xfree(err_msg);
return;
}
/* dump_all_front_end_state - save the state of all front_end nodes to file */
extern int dump_all_front_end_state(void)
{
#ifdef HAVE_FRONT_END
/* Save high-water mark to avoid buffer growth with copies */
static int high_buffer_size = (1024 * 1024);
int error_code = 0, i, log_fd;
char *old_file, *new_file, *reg_file;
front_end_record_t *front_end_ptr;
/* Locks: Read config and node */
slurmctld_lock_t node_read_lock = { READ_LOCK, NO_LOCK, READ_LOCK,
NO_LOCK };
Buf buffer = init_buf(high_buffer_size);
DEF_TIMERS;
START_TIMER;
/* write header: version, time */
packstr(FRONT_END_STATE_VERSION, buffer);
pack_time(time(NULL), buffer);
/* write node records to buffer */
lock_slurmctld (node_read_lock);
for (i = 0, front_end_ptr = front_end_nodes;
i < front_end_node_cnt; i++, front_end_ptr++) {
xassert(front_end_ptr->magic == FRONT_END_MAGIC);
_dump_front_end_state(front_end_ptr, buffer);
}
old_file = xstrdup (slurmctld_conf.state_save_location);
xstrcat (old_file, "/front_end_state.old");
reg_file = xstrdup (slurmctld_conf.state_save_location);
xstrcat (reg_file, "/front_end_state");
new_file = xstrdup (slurmctld_conf.state_save_location);
xstrcat (new_file, "/front_end_state.new");
unlock_slurmctld (node_read_lock);
/* write the buffer to file */
lock_state_files();
log_fd = creat (new_file, 0600);
if (log_fd < 0) {
error ("Can't save state, error creating file %s %m", new_file);
error_code = errno;
} else {
int pos = 0, nwrite = get_buf_offset(buffer), amount, rc;
char *data = (char *)get_buf_data(buffer);
high_buffer_size = MAX(nwrite, high_buffer_size);
while (nwrite > 0) {
amount = write(log_fd, &data[pos], nwrite);
if ((amount < 0) && (errno != EINTR)) {
error("Error writing file %s, %m", new_file);
error_code = errno;
break;
}
nwrite -= amount;
pos += amount;
}
rc = fsync_and_close(log_fd, "front_end");
if (rc && !error_code)
error_code = rc;
}
if (error_code)
(void) unlink (new_file);
else { /* file shuffle */
(void) unlink (old_file);
if (link(reg_file, old_file))
debug4("unable to create link for %s -> %s: %m",
reg_file, old_file);
(void) unlink (reg_file);
if (link(new_file, reg_file))
debug4("unable to create link for %s -> %s: %m",
new_file, reg_file);
(void) unlink (new_file);
}
xfree (old_file);
xfree (reg_file);
xfree (new_file);
unlock_state_files ();
free_buf (buffer);
END_TIMER2("dump_all_front_end_state");
return error_code;
#else
return SLURM_SUCCESS;
#endif
}
/* dump_all_part_state - save the state of all partitions to file */
int dump_all_part_state(void)
{
/* Save high-water mark to avoid buffer growth with copies */
static int high_buffer_size = BUF_SIZE;
ListIterator part_iterator;
struct part_record *part_ptr;
int error_code = 0, log_fd;
char *old_file, *new_file, *reg_file;
/* Locks: Read partition */
slurmctld_lock_t part_read_lock =
{ READ_LOCK, NO_LOCK, NO_LOCK, READ_LOCK };
Buf buffer = init_buf(high_buffer_size);
DEF_TIMERS;
START_TIMER;
/* write header: time */
packstr(PART_STATE_VERSION, buffer);
pack_time(time(NULL), buffer);
/* write partition records to buffer */
lock_slurmctld(part_read_lock);
part_iterator = list_iterator_create(part_list);
while ((part_ptr = (struct part_record *) list_next(part_iterator))) {
xassert (part_ptr->magic == PART_MAGIC);
_dump_part_state(part_ptr, buffer);
}
list_iterator_destroy(part_iterator);
old_file = xstrdup(slurmctld_conf.state_save_location);
xstrcat(old_file, "/part_state.old");
reg_file = xstrdup(slurmctld_conf.state_save_location);
xstrcat(reg_file, "/part_state");
new_file = xstrdup(slurmctld_conf.state_save_location);
xstrcat(new_file, "/part_state.new");
unlock_slurmctld(part_read_lock);
/* write the buffer to file */
lock_state_files();
log_fd = creat(new_file, 0600);
if (log_fd < 0) {
error("Can't save state, error creating file %s, %m",
new_file);
error_code = errno;
} else {
int pos = 0, nwrite = get_buf_offset(buffer), amount, rc;
char *data = (char *)get_buf_data(buffer);
high_buffer_size = MAX(nwrite, high_buffer_size);
while (nwrite > 0) {
amount = write(log_fd, &data[pos], nwrite);
if ((amount < 0) && (errno != EINTR)) {
error("Error writing file %s, %m", new_file);
error_code = errno;
break;
}
nwrite -= amount;
pos += amount;
}
rc = fsync_and_close(log_fd, "partition");
if (rc && !error_code)
error_code = rc;
}
if (error_code)
(void) unlink(new_file);
else { /* file shuffle */
(void) unlink(old_file);
if (link(reg_file, old_file)) {
debug4("unable to create link for %s -> %s: %m",
reg_file, old_file);
}
(void) unlink(reg_file);
if (link(new_file, reg_file)) {
debug4("unable to create link for %s -> %s: %m",
new_file, reg_file);
}
(void) unlink(new_file);
}
xfree(old_file);
xfree(reg_file);
xfree(new_file);
unlock_state_files();
free_buf(buffer);
END_TIMER2("dump_all_part_state");
return 0;
}
extern int fed_mgr_state_save(char *state_save_location)
{
int error_code = 0, log_fd;
char *old_file = NULL, *new_file = NULL, *reg_file = NULL;
slurmctld_lock_t fed_read_lock = {
NO_LOCK, NO_LOCK, NO_LOCK, NO_LOCK, READ_LOCK };
Buf buffer = init_buf(0);
DEF_TIMERS;
START_TIMER;
/* write header: version, time */
pack16(SLURM_PROTOCOL_VERSION, buffer);
pack_time(time(NULL), buffer);
lock_slurmctld(fed_read_lock);
slurmdb_pack_federation_rec(fed_mgr_fed_rec, SLURM_PROTOCOL_VERSION,
buffer);
unlock_slurmctld(fed_read_lock);
/* write the buffer to file */
reg_file = xstrdup_printf("%s/%s", state_save_location,
FED_MGR_STATE_FILE);
old_file = xstrdup_printf("%s.old", reg_file);
new_file = xstrdup_printf("%s.new", reg_file);
log_fd = creat(new_file, 0600);
if (log_fd < 0) {
error("Can't save state, create file %s error %m", new_file);
error_code = errno;
} else {
int pos = 0, nwrite = get_buf_offset(buffer), amount;
char *data = (char *)get_buf_data(buffer);
while (nwrite > 0) {
amount = write(log_fd, &data[pos], nwrite);
if ((amount < 0) && (errno != EINTR)) {
error("Error writing file %s, %m", new_file);
error_code = errno;
break;
}
nwrite -= amount;
pos += amount;
}
fsync(log_fd);
close(log_fd);
}
if (error_code)
(void) unlink(new_file);
else { /* file shuffle */
(void) unlink(old_file);
if (link(reg_file, old_file))
debug4("unable to create link for %s -> %s: %m",
reg_file, old_file);
(void) unlink(reg_file);
if (link(new_file, reg_file))
debug4("unable to create link for %s -> %s: %m",
new_file, reg_file);
(void) unlink(new_file);
}
xfree(old_file);
xfree(reg_file);
xfree(new_file);
free_buf(buffer);
END_TIMER2("fed_mgr_state_save");
return error_code;
}
extern int
pgsql_db_make_table_current(PGconn *pgsql_db, char *schema, char *table_name,
storage_field_t *fields)
{
char *query = NULL, *opt_part = NULL, *temp_char = NULL;
char *type = NULL;
int not_null = 0;
char *default_str = NULL;
char* original_ptr = NULL;
int i = 0;
PGresult *result = NULL;
List columns = NULL;
ListIterator itr = NULL;
char *col = NULL;
DEF_TIMERS;
query = xstrdup_printf("select column_name from "
"information_schema.columns where "
"table_name='%s' and table_schema='%s' ",
table_name, schema);
if (!(result = pgsql_db_query_ret(pgsql_db, query))) {
xfree(query);
return SLURM_ERROR;
}
xfree(query);
columns = list_create(slurm_destroy_char);
for (i = 0; i < PQntuples(result); i++) {
col = xstrdup(PQgetvalue(result, i, 0)); //column_name
list_append(columns, col);
}
PQclear(result);
itr = list_iterator_create(columns);
query = xstrdup_printf("alter table %s.%s", schema, table_name);
START_TIMER;
i=0;
while (fields[i].name) {
int found = 0;
not_null = 0;
if (!strcasecmp("serial", fields[i].options)) {
i++;
continue;
}
opt_part = xstrdup(fields[i].options);
original_ptr = opt_part;
opt_part = strtok_r(opt_part, " ", &temp_char);
if (opt_part) {
/* XXX: only one identifier supported */
type = xstrdup(opt_part);
opt_part = strtok_r(NULL, " ", &temp_char);
while (opt_part) {
if (!strcasecmp("not", opt_part)) {
opt_part = strtok_r(NULL, " ",
&temp_char);
if (!strcasecmp("null", opt_part)) {
not_null = 1;
}
} else if (!strcasecmp("default", opt_part)){
opt_part = strtok_r(NULL,
" ", &temp_char);
default_str = xstrdup(opt_part);
}
opt_part = strtok_r(NULL,
" ", &temp_char);
}
} else {
type = xstrdup(fields[i].options);
}
xfree(original_ptr);
list_iterator_reset(itr);
while ((col = list_next(itr))) {
if (!strcmp(col, fields[i].name)) {
list_delete_item(itr);
found = 1;
break;
}
}
temp_char = NULL;
if (!found) {
info("adding column %s", fields[i].name);
if (default_str)
xstrfmtcat(temp_char,
" default %s", default_str);
if (not_null)
xstrcat(temp_char, " not null");
xstrfmtcat(query,
" add %s %s",
fields[i].name, type);
if (temp_char)
xstrcat(query, temp_char);
xstrcat(query, ",");
} else {
if (default_str)
xstrfmtcat(temp_char,
" alter %s set default %s,",
fields[i].name, default_str);
else
xstrfmtcat(temp_char,
" alter %s drop default,",
fields[i].name);
if (not_null)
xstrfmtcat(temp_char,
" alter %s set not null,",
fields[i].name);
else
xstrfmtcat(temp_char,
" alter %s drop not null,",
fields[i].name);
xstrfmtcat(query, " alter %s type %s,%s",
fields[i].name, type, temp_char);
}
xfree(temp_char);
xfree(default_str);
xfree(type);
i++;
}
list_iterator_destroy(itr);
list_destroy(columns);
query[strlen(query)-1] = ';';
//debug4("pgsql db create/alter table:\n %s", query);
if (pgsql_db_query(pgsql_db, query)) {
xfree(query);
return SLURM_ERROR;
}
xfree(query);
END_TIMER2("make table current");
return SLURM_SUCCESS;
}
