[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

[Cluster-devel] cluster/rgmanager include/resgroup.h src/daemo ...



CVSROOT:	/cvs/cluster
Module name:	cluster
Branch: 	RHEL4
Changes by:	lhh sourceware org	2007-11-14 19:03:37

Modified files:
	rgmanager/include: resgroup.h 
	rgmanager/src/daemons: groups.c rg_state.c 

Log message:
	Fix #360401 - hang forever during shutdown due to previous service boot problem

Patches:
http://sourceware.org/cgi-bin/cvsweb.cgi/cluster/rgmanager/include/resgroup.h.diff?cvsroot=cluster&only_with_tag=RHEL4&r1=1.3.2.9&r2=1.3.2.10
http://sourceware.org/cgi-bin/cvsweb.cgi/cluster/rgmanager/src/daemons/groups.c.diff?cvsroot=cluster&only_with_tag=RHEL4&r1=1.8.2.21&r2=1.8.2.22
http://sourceware.org/cgi-bin/cvsweb.cgi/cluster/rgmanager/src/daemons/rg_state.c.diff?cvsroot=cluster&only_with_tag=RHEL4&r1=1.4.2.21&r2=1.4.2.22

--- cluster/rgmanager/include/resgroup.h	2007/01/03 21:08:17	1.3.2.9
+++ cluster/rgmanager/include/resgroup.h	2007/11/14 19:03:37	1.3.2.10
@@ -156,6 +156,7 @@
 cluster_member_list_t *member_list(void);
 uint64_t my_id(void);
 
+#define RG_ERELO	-9 /* Operation cannot complete here */
 #define RG_ENODEDEATH	-8 /* Processing node died */
 #define RG_ERUN		-7 /* Service is running already */
 #define RG_EAGAIN	-6 /* Try again */
--- cluster/rgmanager/src/daemons/groups.c	2007/09/28 15:14:52	1.8.2.21
+++ cluster/rgmanager/src/daemons/groups.c	2007/11/14 19:03:37	1.8.2.22
@@ -192,7 +192,8 @@
 		}
 
 		if (st.rs_state != RG_STATE_STARTED &&
-		     st.rs_state != RG_STATE_STARTING)
+		     st.rs_state != RG_STATE_STARTING &&
+		     st.rs_state != RG_STATE_STOPPING)
 			continue;
 
 		if (mp->cm_id != st.rs_owner)
--- cluster/rgmanager/src/daemons/rg_state.c	2007/06/28 11:54:50	1.4.2.21
+++ cluster/rgmanager/src/daemons/rg_state.c	2007/11/14 19:03:37	1.4.2.22
@@ -41,10 +41,13 @@
 int set_rg_state(char *servicename, rg_state_t *svcblk);
 int get_rg_state(char *servicename, rg_state_t *svcblk);
 void get_recovery_policy(char *rg_name, char *buf, size_t buflen);
-int have_exclusive_resources();
+int have_exclusive_resources(void);
 int check_exclusive_resources(cluster_member_list_t *membership, char *svcName);
 
 
+pthread_mutex_t exclusive_mutex = PTHREAD_MUTEX_INITIALIZER;
+
+
 uint64_t
 next_node_id(cluster_member_list_t *membership, uint64_t me)
 {
@@ -446,6 +449,7 @@
 		break;
 	
 	case RG_STATE_DISABLED:
+		ret = 2;
 	case RG_STATE_UNINITIALIZED:
 		if (req == RG_DISABLE) {
 			clulog(LOG_NOTICE,
@@ -536,7 +540,7 @@
 			}
 
 			clulog(LOG_NOTICE,
-			       "Starting stopped service%s\n",
+			       "Starting stopped service %s\n",
 			       svcName);
 			ret = 1;
 			break;
@@ -557,7 +561,7 @@
 			snprintf(query,
 				sizeof(query),
 				"/cluster/clusternodes/clusternode[ nodeid=\"%d\"]/@name",
-				svcStatus->rs_owner);
+				(int)svcStatus->rs_owner);
 			ccs_get(fd, query, &nodename);
 			ccs_disconnect(fd);
 		}
@@ -650,42 +654,61 @@
 svc_start(char *svcName, int req)
 {
 	void *lockp = NULL;
-	int ret;
+	int ret, xret;
 	rg_state_t svcStatus;
+	int need_check = have_exclusive_resources();
+	cluster_member_list_t *membership;
+
+ 	if (need_check)
+		pthread_mutex_lock(&exclusive_mutex);
+
+	ret = RG_EFAIL;
 
 	if (rg_lock(svcName, &lockp) < 0) {
 		clulog(LOG_ERR, "#45: Unable to obtain cluster lock: %s\n",
 		       strerror(errno));
-		return FAIL;
+		goto out_nolock;
 	}
 
 	if (get_rg_state(svcName, &svcStatus) != 0) {
-		rg_unlock(svcName, lockp);
 		clulog(LOG_ERR, "#46: Failed getting status for RG %s\n",
 		       svcName);
-		return FAIL;
+		goto out_unlock;
+	}
+
+	if (need_check) {
+		membership = member_list();
+		xret = check_exclusive_resources(membership, svcName);
+		cml_free(membership);
+		if (xret != 0) {
+			if (xret > 0)
+				/* Exc. service running */
+				ret = RG_ERELO;
+			else
+				/* XXX */
+				ret = RG_ENOSERVICE;
+			goto out_unlock;
+		}
 	}
 
 	/* LOCK HELD */
 	switch (svc_advise_start(&svcStatus, svcName, req)) {
 	case 0: /* Don't start service, return FAIL */
-		rg_unlock(svcName, lockp);
-		return FAIL;
+		goto out_unlock;
 	case 2: /* Don't start service, return 0 */
-		rg_unlock(svcName, lockp);
-		return 0;
+		ret = 0;
+		goto out_unlock;
 	case 3:
-		rg_unlock(svcName, lockp);
-		return RG_EAGAIN;
+		ret = RG_EAGAIN;
+		goto out_unlock;
 	case 4:
-		rg_unlock(svcName, lockp);
-		return RG_ERUN;
+		ret = RG_ERUN;
+		goto out_unlock;
 	default:
 		break;
 	}
 
 	/* LOCK HELD if we get here */
-
 	svcStatus.rs_owner = my_id();
 	svcStatus.rs_state = RG_STATE_STARTING;
 	svcStatus.rs_transition = (uint64_t)time(NULL);
@@ -699,10 +722,17 @@
 		clulog(LOG_ERR,
 		       "#47: Failed changing service status\n");
 		rg_unlock(svcName, lockp);
-		return FAIL;
+		goto out_unlock;
 	}
 	
 	rg_unlock(svcName, lockp);
+	
+	/* release excl. mutex during start */
+	if (need_check) {
+		/* Also clear need_check so we don't double-unlock */
+		pthread_mutex_unlock(&exclusive_mutex);
+		need_check = 0;
+	}
 
 	ret = group_op(svcName, RG_START);
 	ret = !!ret; /* Either it worked or it didn't.  Ignore all the
@@ -711,17 +741,17 @@
 	if (rg_lock(svcName, &lockp) < 0) {
 		clulog(LOG_ERR, "#74: Unable to obtain cluster lock: %s\n",
 		       strerror(errno));
-		return FAIL;
+		ret = RG_EFAIL;
+		goto out_nolock;
 	}
 
 	svcStatus.rs_state = RG_STATE_STARTED;
 	if (set_rg_state(svcName, &svcStatus) != 0) {
 		clulog(LOG_ERR,
 		       "#75: Failed changing service status\n");
-		rg_unlock(svcName, lockp);
-		return FAIL;
+		ret = RG_EFAIL;
+		goto out_unlock;
 	}
-	rg_unlock(svcName, lockp);
        
 	if (ret == 0)
 		clulog(LOG_NOTICE,
@@ -732,6 +762,11 @@
 		       "#68: Failed to start %s; return value: %d\n",
 		       svcName, ret);
 
+out_unlock:
+	rg_unlock(svcName, lockp);
+out_nolock:
+	if (need_check)
+		pthread_mutex_unlock(&exclusive_mutex);
 	return ret;
 }
 
@@ -1115,7 +1150,7 @@
 {
 	cluster_member_list_t *allowed_nodes, *backup = NULL;
 	uint64_t target = preferred_target, me = my_id();
-	int ret, x;
+	int ret, x, tried = 0;
 	
 	/*
 	 * Stop the service - if we haven't already done so.
@@ -1181,6 +1216,7 @@
 		 	 * It's legal to start the service on the given
 		 	 * node.  Try to do so.
 		 	 */
+			++tried;
 			if (relocate_service(svcName, request, target) == 0) {
 				*new_owner = target;
 				/*
@@ -1211,9 +1247,12 @@
 		if (target == me)
 			goto exhausted;
 
+		++tried;
+
+		/* Each node gets one try */
+		memb_mark_down(allowed_nodes, target);
 		switch (relocate_service(svcName, request, target)) {
 		case RG_EFAIL:
-			memb_mark_down(allowed_nodes, target);
 			continue;
 		case RG_EABORT:
 			svc_report_failure(svcName);
@@ -1228,9 +1267,10 @@
 			       (uint32_t)(target&0xffffffff), request);
 			return 0;
 		case 0:
-			*new_owner = target;
 			clulog(LOG_NOTICE, "Service %s is now running "
 			       "on member %d\n", svcName, (int)target);
+		case RG_ERUN:
+			*new_owner = target;
 			cml_free(allowed_nodes);
 			return 0;
 		default:
@@ -1254,9 +1294,10 @@
 	 */
 exhausted:
 	if (!rg_locked()) {
-		clulog(LOG_WARNING,
-		       "#70: Attempting to restart service %s locally.\n",
-		       svcName);
+		if (tried)
+			clulog(LOG_WARNING,
+			       "#70: Attempting to restart service %s locally.\n",
+			       svcName);
 		if (svc_start(svcName, RG_START_RECOVER) == 0) {
 			*new_owner = me;
 			return FAIL;
@@ -1275,9 +1316,9 @@
 int
 handle_fd_start_req(char *svcName, int request, uint64_t *new_owner)
 {
-	cluster_member_list_t *allowed_nodes, *backup = NULL;
+	cluster_member_list_t *allowed_nodes;
 	uint64_t target, me = my_id();
-	int ret, x;
+	int ret;
 	
 	allowed_nodes = member_list();
 
@@ -1326,7 +1367,6 @@
 }
 
 
-pthread_mutex_t exclusive_mutex = PTHREAD_MUTEX_INITIALIZER;
 /**
  * handle_start_req - Handle a generic start request from a user or during
  * service manager boot.
@@ -1342,7 +1382,6 @@
 {
 	int ret, tolerance = FOD_BEST;
 	cluster_member_list_t *membership = member_list();
-	int need_check = have_exclusive_resources();
 
 	/*
 	 * When a service request is from a user application (eg, clusvcadm),
@@ -1358,18 +1397,6 @@
 		cml_free(membership);
 		return FAIL;
 	}
-	if (need_check) {
-		pthread_mutex_lock(&exclusive_mutex);
-		ret = check_exclusive_resources(membership, svcName);
-		if (ret != 0) {
-			cml_free(membership);
-			pthread_mutex_unlock(&exclusive_mutex);
-			if (ret > 0)
-				goto relocate;
-			else
-				return FAIL;
-		}
-	}
 	cml_free(membership);
 	
 	/*
@@ -1377,25 +1404,22 @@
 	 * mask here - so that we can try all nodes if necessary.
 	 */
 	ret = svc_start(svcName, req);
-	if (need_check)
-		pthread_mutex_unlock(&exclusive_mutex);
-
-	/* 
-	   If services are locked, return the error 
-	  */
-	if (ret == RG_EAGAIN || ret == RG_ERUN)
+	switch(ret) {
+	case RG_ERELO:
+		goto relocate;
+
+	case RG_EAGAIN:
+		/* If services are locked, return the error */
+	case RG_ENOSERVICE:
+		/* service doesn't exist? */
+	case RG_ERUN:
+		/* If service is already running, return that value */
 		return ret;
 
-	/*
-	 * If we succeeded, then we're done.
-	 */
-	if (ret == SUCCESS) {
+	case SUCCESS:
+		/* If we succeeded, then we're done.  */
 		*new_owner = my_id();
-		return SUCCESS;
-	}
-
-	/* Already running? */
-	if (ret == NO) {
+	case NO: 
 		return SUCCESS;
 	}
 	
@@ -1418,13 +1442,13 @@
 		return RG_EABORT;
 	}
 	
-relocate:
 	/*
 	 * OK, it failed to start - but succeeded to stop.  Now,
 	 * we should relocate the service.
 	 */
 	clulog(LOG_WARNING, "#71: Relocating failed service %s\n",
 	       svcName);
+relocate:
 	ret = handle_relocate_req(svcName, RG_START_RECOVER, -1, new_owner);
 
 	/* If we leave the service stopped, instead of disabled, someone
@@ -1456,7 +1480,6 @@
 	int x;
 	uint64_t me = my_id();
 	cluster_member_list_t *membership = member_list();
-	int need_check = have_exclusive_resources();
 
 	/* XXX ok, so we need to say "should I start this if I was the
 	   only cluster member online */
@@ -1477,23 +1500,29 @@
 		cml_free(membership);
 		return FAIL;
 	}
- 	if (need_check) {
- 		pthread_mutex_lock(&exclusive_mutex);
- 		if (check_exclusive_resources(membership, svcName) != 0) {
- 			pthread_mutex_unlock(&exclusive_mutex);
- 			cml_free(membership);
- 			return FAIL;
- 		}
- 	}
 	cml_free(membership);
 
 	x = svc_start(svcName, req);
-	if (need_check)
- 		pthread_mutex_unlock(&exclusive_mutex);
-	if (x == 0)
-		return 0;
-	if (x == RG_ERUN)
-		return RG_ERUN;
+	switch(x) {
+	case RG_ERELO:
+		/* Don't relocate from here; it was a remote start */
+		/* Return fail so the other node can go ahead and 
+		   try the other nodes in the cluster */
+	case NO: 
+		return RG_EFAIL;
+
+	case RG_EAGAIN:
+		/* If services are locked, return the error */
+	case RG_ENOSERVICE:
+		/* service doesn't exist? */
+	case RG_ERUN:
+		/* If service is already running, return that value */
+		return x;
+
+	case SUCCESS:
+		/* If we succeeded, then we're done.  */
+		return SUCCESS;
+	}
 
 	if (svc_stop(svcName, RG_STOP_RECOVER) == 0)
 		return RG_EFAIL;


[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]