Changeset: 3551b30d82fe for MonetDB
URL: http://dev.monetdb.org/hg/MonetDB?cmd=changeset;node=3551b30d82fe
Modified Files:
tools/merovingian/daemon/forkmserver.c
tools/merovingian/daemon/merovingian.c
Branch: Jun2016
Log Message:
Move terminateProcess to same file as formMserver, and use fork_lock.
fork_lock is held while a new server is being started so that multiple
starts of the same server don't interfere with each other. We should
then also hold the same lock while stopping a server.
Part of the ongoing quest to quash bug 4066.
diffs (285 lines):
diff --git a/tools/merovingian/daemon/forkmserver.c
b/tools/merovingian/daemon/forkmserver.c
--- a/tools/merovingian/daemon/forkmserver.c
+++ b/tools/merovingian/daemon/forkmserver.c
@@ -31,6 +31,145 @@
static pthread_mutex_t fork_lock = PTHREAD_MUTEX_INITIALIZER;
/**
+ * The terminateProcess function tries to let the given mserver process
+ * shut down gracefully within a given time-out. If that fails, it
+ * sends the deadly SIGKILL signal to the mserver process and returns.
+ */
+void *
+terminateProcess(void *p)
+{
+ dpair d = (dpair)p;
+ sabdb *stats;
+ char *er;
+ int i;
+ confkeyval *kv;
+ /* make local copies since d will disappear when killed */
+ pid_t pid = d->pid;
+ char *dbname = strdup(d->dbname);
+
+ pthread_mutex_lock(&fork_lock);
+
+ er = msab_getStatus(&stats, dbname);
+ if (er != NULL) {
+ pthread_mutex_unlock(&fork_lock);
+ Mfprintf(stderr, "cannot terminate process " LLFMT ": %s\n",
+ (long long int)pid, er);
+ free(er);
+ free(dbname);
+ return NULL;
+ }
+
+ if (stats == NULL) {
+ pthread_mutex_unlock(&fork_lock);
+ Mfprintf(stderr, "strange, process " LLFMT " serves database
'%s' "
+ "which does not exist\n", (long long int)pid,
dbname);
+ free(dbname);
+ return NULL;
+ }
+
+ switch (stats->state) {
+ case SABdbRunning:
+ /* ok, what we expect */
+ break;
+ case SABdbCrashed:
+ pthread_mutex_unlock(&fork_lock);
+ Mfprintf(stderr, "cannot shut down database '%s',
mserver "
+ "(pid " LLFMT ") has crashed\n",
+ dbname, (long long int)pid);
+ msab_freeStatus(&stats);
+ free(dbname);
+ return NULL;
+ case SABdbInactive:
+ pthread_mutex_unlock(&fork_lock);
+ Mfprintf(stdout, "database '%s' appears to have shut
down already\n",
+ dbname);
+ fflush(stdout);
+ msab_freeStatus(&stats);
+ free(dbname);
+ return NULL;
+ case SABdbStarting:
+ pthread_mutex_unlock(&fork_lock);
+ Mfprintf(stderr, "database '%s' appears to be starting
up\n",
+ dbname);
+ /* starting up, so we'll go to the shut down phase */
+ break;
+ default:
+ pthread_mutex_unlock(&fork_lock);
+ Mfprintf(stderr, "unknown state: %d\n",
(int)stats->state);
+ msab_freeStatus(&stats);
+ free(dbname);
+ return NULL;
+ }
+
+ if (d->type == MEROFUN) {
+ pthread_mutex_unlock(&fork_lock);
+ multiplexDestroy(dbname);
+ msab_freeStatus(&stats);
+ free(dbname);
+ return NULL;
+ } else if (d->type != MERODB) {
+ /* barf */
+ pthread_mutex_unlock(&fork_lock);
+ Mfprintf(stderr, "cannot stop merovingian process role: %s\n",
dbname);
+ msab_freeStatus(&stats);
+ free(dbname);
+ return NULL;
+ }
+
+ /* ok, once we get here, we'll be shutting down the server */
+ Mfprintf(stdout, "sending process " LLFMT " (database '%s') the "
+ "TERM signal\n", (long long int)pid, dbname);
+ kill(pid, SIGTERM);
+ kv = findConfKey(_mero_props, "exittimeout");
+ for (i = 0; i < atoi(kv->val) * 2; i++) {
+ if (stats != NULL)
+ msab_freeStatus(&stats);
+ sleep_ms(500);
+ er = msab_getStatus(&stats, dbname);
+ if (er != NULL) {
+ Mfprintf(stderr, "unexpected problem: %s\n", er);
+ free(er);
+ /* don't die, just continue, so we KILL in the end */
+ } else if (stats == NULL) {
+ Mfprintf(stderr, "hmmmm, database '%s' suddenly doesn't
exist "
+ "any more\n", dbname);
+ } else {
+ switch (stats->state) {
+ case SABdbRunning:
+ case SABdbStarting:
+ /* ok, try again */
+ break;
+ case SABdbCrashed:
+ pthread_mutex_unlock(&fork_lock);
+ Mfprintf (stderr, "database '%s'
crashed after SIGTERM\n",
+ dbname);
+ msab_freeStatus(&stats);
+ free(dbname);
+ return NULL;
+ case SABdbInactive:
+ pthread_mutex_unlock(&fork_lock);
+ Mfprintf(stdout, "database '%s' has
shut down\n", dbname);
+ fflush(stdout);
+ msab_freeStatus(&stats);
+ free(dbname);
+ return NULL;
+ default:
+ Mfprintf(stderr, "unknown state: %d\n",
(int)stats->state);
+ break;
+ }
+ }
+ }
+ Mfprintf(stderr, "timeout of %s seconds expired, sending process " LLFMT
+ " (database '%s') the KILL signal\n",
+ kv->val, (long long int)pid, dbname);
+ kill(pid, SIGKILL);
+ msab_freeStatus(&stats);
+ free(dbname);
+ pthread_mutex_unlock(&fork_lock);
+ return NULL;
+}
+
+/**
* Fork an mserver and detach. Before forking off, Sabaoth is consulted
* to see if forking makes sense, or whether it is necessary at all, or
* forbidden by restart policy, e.g. when in maintenance.
diff --git a/tools/merovingian/daemon/merovingian.c
b/tools/merovingian/daemon/merovingian.c
--- a/tools/merovingian/daemon/merovingian.c
+++ b/tools/merovingian/daemon/merovingian.c
@@ -233,132 +233,6 @@ logListener(void *x)
}
/**
- * The terminateProcess function tries to let the given mserver process
- * shut down gracefully within a given time-out. If that fails, it
- * sends the deadly SIGKILL signal to the mserver process and returns.
- */
-void *
-terminateProcess(void *p)
-{
- dpair d = (dpair)p;
- sabdb *stats;
- char *er;
- int i;
- confkeyval *kv;
- /* make local copies since d will disappear when killed */
- pid_t pid = d->pid;
- char *dbname = strdup(d->dbname);
-
- er = msab_getStatus(&stats, dbname);
- if (er != NULL) {
- Mfprintf(stderr, "cannot terminate process " LLFMT ": %s\n",
- (long long int)pid, er);
- free(er);
- free(dbname);
- return NULL;
- }
-
- if (stats == NULL) {
- Mfprintf(stderr, "strange, process " LLFMT " serves database
'%s' "
- "which does not exist\n", (long long int)pid,
dbname);
- free(dbname);
- return NULL;
- }
-
- switch (stats->state) {
- case SABdbRunning:
- /* ok, what we expect */
- break;
- case SABdbCrashed:
- Mfprintf(stderr, "cannot shut down database '%s',
mserver "
- "(pid " LLFMT ") has crashed\n",
- dbname, (long long int)pid);
- msab_freeStatus(&stats);
- free(dbname);
- return NULL;
- case SABdbInactive:
- Mfprintf(stdout, "database '%s' appears to have shut
down already\n",
- dbname);
- fflush(stdout);
- msab_freeStatus(&stats);
- free(dbname);
- return NULL;
- case SABdbStarting:
- Mfprintf(stderr, "database '%s' appears to be starting
up\n",
- dbname);
- /* starting up, so we'll go to the shut down phase */
- break;
- default:
- Mfprintf(stderr, "unknown state: %d\n",
(int)stats->state);
- msab_freeStatus(&stats);
- free(dbname);
- return NULL;
- }
-
- if (d->type == MEROFUN) {
- multiplexDestroy(dbname);
- msab_freeStatus(&stats);
- free(dbname);
- return NULL;
- } else if (d->type != MERODB) {
- /* barf */
- Mfprintf(stderr, "cannot stop merovingian process role: %s\n",
dbname);
- msab_freeStatus(&stats);
- free(dbname);
- return NULL;
- }
-
- /* ok, once we get here, we'll be shutting down the server */
- Mfprintf(stdout, "sending process " LLFMT " (database '%s') the "
- "TERM signal\n", (long long int)pid, dbname);
- kill(pid, SIGTERM);
- kv = findConfKey(_mero_props, "exittimeout");
- for (i = 0; i < atoi(kv->val) * 2; i++) {
- if (stats != NULL)
- msab_freeStatus(&stats);
- sleep_ms(500);
- er = msab_getStatus(&stats, dbname);
- if (er != NULL) {
- Mfprintf(stderr, "unexpected problem: %s\n", er);
- free(er);
- /* don't die, just continue, so we KILL in the end */
- } else if (stats == NULL) {
- Mfprintf(stderr, "hmmmm, database '%s' suddenly doesn't
exist "
- "any more\n", dbname);
- } else {
- switch (stats->state) {
- case SABdbRunning:
- case SABdbStarting:
- /* ok, try again */
- break;
- case SABdbCrashed:
- Mfprintf (stderr, "database '%s'
crashed after SIGTERM\n",
- dbname);
- msab_freeStatus(&stats);
- free(dbname);
- return NULL;
- case SABdbInactive:
- Mfprintf(stdout, "database '%s' has
shut down\n", dbname);
- fflush(stdout);
- msab_freeStatus(&stats);
- free(dbname);
- return NULL;
- default:
- Mfprintf(stderr, "unknown state: %d\n",
(int)stats->state);
- break;
- }
- }
- }
- Mfprintf(stderr, "timeout of %s seconds expired, sending process " LLFMT
- " (database '%s') the KILL signal\n",
- kv->val, (long long int)pid, dbname);
- kill(pid, SIGKILL);
- msab_freeStatus(&stats);
- free(dbname);
- return NULL;
-}
-
-/**
* Creates a new error, allocated with malloc. The error should be
* freed using freeErr().
*/
_______________________________________________
checkin-list mailing list
checkin-list@monetdb.org
https://www.monetdb.org/mailman/listinfo/checkin-list
