Attachment 312662 Details for Bug 435466 – Pass 1 patch

[patch] Pass 1 patch

rdomain-crash-rhel5.patch (text/plain), 10.16 KB, created by Lon Hohberger on 2008-07-25 17:40:12 UTC

(hide)

Description:

Filename:

MIME Type:

Creator: Lon Hohberger

Created: 2008-07-25 17:40:12 UTC

Size: 10.16 KB

patch

obsolete

>diff -ur rgmanager-current/include/members.h rgmanager/include/members.h
>--- rgmanager-current/include/members.h	2008-07-24 16:24:24.000000000 -0400
>+++ rgmanager/include/members.h	2008-07-24 16:45:49.000000000 -0400
>@@ -17,6 +17,7 @@
> void member_set_state(int nodeid, int state);
> int memb_count(cluster_member_list_t *ml);
> int member_online(int nodeid);
>+int member_online_set(int **nodes, int *nodecount);
> int memb_online(cluster_member_list_t *ml, int nodeid);
> int memb_online_name(cluster_member_list_t *ml, char *name);
> int memb_name_to_id(cluster_member_list_t *ml, char *name);
>diff -ur rgmanager-current/include/reslist.h rgmanager/include/reslist.h
>--- rgmanager-current/include/reslist.h	2008-07-24 16:24:46.000000000 -0400
>+++ rgmanager/include/reslist.h	2008-07-24 17:23:34.000000000 -0400
>@@ -202,7 +202,7 @@
> void print_domains(fod_t **domains);
> int node_should_start(int nodeid, cluster_member_list_t *membership,
> 		      char *rg_name, fod_t **domains);
>-int node_domain_set(fod_t *domain, int **ret, int *retlen);
>+int node_domain_set(fod_t **domains, char *name, int **ret, int *retlen, int *flags);
> int node_domain_set_safe(char *domainname, int **ret, int *retlen, int *flags);
> 
> 
>diff -ur rgmanager-current/src/clulib/members.c rgmanager/src/clulib/members.c
>--- rgmanager-current/src/clulib/members.c	2008-07-24 16:24:46.000000000 -0400
>+++ rgmanager/src/clulib/members.c	2008-07-25 10:41:18.000000000 -0400
>@@ -213,6 +213,35 @@
> }
> 
> 
>+int
>+member_online_set(int **nodes, int *nodecount)
>+{
>+	int ret = 1, i;
>+
>+	pthread_rwlock_rdlock(&memblock);
>+	if (!membership)
>+		goto out_unlock;
>+
>+	*nodes = malloc(sizeof(int) * membership->cml_count);
>+	if (!*nodes)
>+		goto out_unlock;
>+
>+	*nodecount = 0;
>+	for (i = 0; i < membership->cml_count; i++) {
>+		if (membership->cml_members[i].cn_member &&
>+		    membership->cml_members[i].cn_nodeid != 0) {
>+			(*nodes)[*nodecount] = membership->cml_members[i].cn_nodeid;
>+			++(*nodecount);
>+		}
>+	}
>+
>+	ret = 0;
>+out_unlock:
>+	pthread_rwlock_unlock(&memblock);
>+	return ret;
>+}
>+
>+
> void
> member_set_state(int nodeid, int state)
> {
>diff -ur rgmanager-current/src/daemons/fo_domain.c rgmanager/src/daemons/fo_domain.c
>--- rgmanager-current/src/daemons/fo_domain.c	2008-07-24 16:24:46.000000000 -0400
>+++ rgmanager/src/daemons/fo_domain.c	2008-07-24 17:24:33.000000000 -0400
>@@ -349,13 +349,24 @@
> 
> 
> int
>-node_domain_set(fod_t *domain, int **ret, int *retlen)
>+node_domain_set(fod_t **domains, char *name, int **ret, int *retlen, int *flags)
> {
> 	int x, i, j;
> 	int *tmpset;
> 	int ts_count;
>-
> 	fod_node_t *fodn;
>+	fod_t *domain;
>+	int rv = -1, found = 0;
>+
>+	list_for(domains, domain, x) {
>+		if (!strcasecmp(domain->fd_name, name)) {
>+			found = 1;
>+			break;
>+		}
>+	} // while (!list_done(&_domains, fod));
>+
>+	if (!found)
>+		return -1;
> 
> 	/* Count domain length */
> 	list_for(&domain->fd_nodes, fodn, x) { }
>@@ -368,6 +379,8 @@
> 	if (!(*tmpset))
> 		return -1;
> 
>+	*flags = domain->fd_flags;
>+
> 	if (domain->fd_flags & FOD_ORDERED) {
> 		for (i = 1; i <= 100; i++) {
> 			
>diff -ur rgmanager-current/src/daemons/groups.c rgmanager/src/daemons/groups.c
>--- rgmanager-current/src/daemons/groups.c	2008-07-24 16:24:46.000000000 -0400
>+++ rgmanager/src/daemons/groups.c	2008-07-25 10:34:43.000000000 -0400
>@@ -58,6 +58,7 @@
> 
> void res_build_name(char *, size_t, resource_t *);
> int group_migratory(char *groupname, int lock);
>+int _group_property(char *groupname, char *property, char *ret, size_t len);
> 
> 
> struct status_arg {
>@@ -88,23 +89,9 @@
> int
> node_domain_set_safe(char *domainname, int **ret, int *retlen, int *flags)
> {
>-	fod_t *fod;
>-	int rv = -1, found = 0, x = 0;
>-
>+	int rv = 0;
> 	pthread_rwlock_rdlock(&resource_lock);
>-
>-	list_for(&_domains, fod, x) {
>-		if (!strcasecmp(fod->fd_name, domainname)) {
>-			found = 1;
>-			break;
>-		}
>-	} // while (!list_done(&_domains, fod));
>-
>-	if (found) {
>-		rv = node_domain_set(fod, ret, retlen);
>-		*flags = fod->fd_flags;
>-	}
>-
>+	rv = node_domain_set(&_domains, domainname, ret, retlen, flags);
> 	pthread_rwlock_unlock(&resource_lock);
> 
> 	return rv;
>@@ -438,6 +425,47 @@
> }
> 
> 
>+int
>+check_rdomain_crash(char *svcName)
>+{
>+	int *nodes = NULL, nodecount;
>+	int *fd_nodes = NULL, fd_nodecount, fl;
>+	int *isect = NULL, icount;
>+	char fd_name[256];
>+
>+	if (_group_property(svcName, "domain", fd_name, sizeof(fd_name)) != 0)
>+		goto out_free;
>+
>+	if (node_domain_set(_domains, fd_name, &fd_nodes,
>+			    &fd_nodecount, &fl) != 0)
>+		goto out_free;
>+
>+	if (!(fl & FOD_RESTRICTED))
>+		goto out_free;
>+	
>+	if (s_intersection(fd_nodes, fd_nodecount, nodes, nodecount, 
>+		    &isect, &icount) < 0)
>+		goto out_free;
>+
>+	if (icount == 0) {
>+		clulog(LOG_NOTICE, "Marking %s as stopped: "
>+		       "Restricted domain unavailable\n", svcName);
>+		rt_enqueue_request(svcName, RG_STOP, NULL, 0, 0,
>+				   0, 0);
>+	}
>+
>+out_free:
>+	if (fd_nodes)
>+		free(fd_nodes);
>+	if (nodes)
>+		free(nodes);
>+	if (isect)
>+		free(isect);
>+
>+	return 0;
>+}
>+
>+
> /**
>   Start or failback a resource group: if it's not running, start it.
>   If it is running and we're a better member to run it, then ask for
>@@ -451,6 +479,7 @@
> 	cman_node_t *mp;
> 	int autostart, exclusive;
> 	struct dlm_lksb lockp;
>+	int fod_ret;
> 
> 	mp = memb_id_to_p(membership, my_id());
> 	assert(mp);
>@@ -542,10 +571,13 @@
> 	 * Start any stopped services, or started services
> 	 * that are owned by a down node.
> 	 */
>-	if (node_should_start(mp->cn_nodeid, membership, svcName, &_domains) ==
>-	    FOD_BEST)
>+	fod_ret = node_should_start(mp->cn_nodeid, membership,
>+				    svcName, &_domains);
>+	if (fod_ret == FOD_BEST)
> 		rt_enqueue_request(svcName, RG_START, NULL, 0, mp->cn_nodeid,
> 				   0, 0);
>+	else if (fod_ret == FOD_ILLEGAL)
>+		check_rdomain_crash(svcName);
> }
> 
> 
>@@ -1060,15 +1092,13 @@
>    @return		0 on success, -1 on failure.
>  */
> int
>-group_property(char *groupname, char *property, char *ret, size_t len)
>+_group_property(char *groupname, char *property, char *ret, size_t len)
> {
> 	resource_t *res = NULL;
> 	int x = 0;
> 
>-	pthread_rwlock_rdlock(&resource_lock);
> 	res = find_root_by_ref(&_resources, groupname);
> 	if (!res) {
>-		pthread_rwlock_unlock(&resource_lock);
> 		return -1;
> 	}
> 
>@@ -1076,15 +1106,24 @@
> 		if (strcasecmp(res->r_attrs[x].ra_name, property))
> 			continue;
> 		strncpy(ret, res->r_attrs[x].ra_value, len);
>-		pthread_rwlock_unlock(&resource_lock);
> 		return 0;
> 	}
>-	pthread_rwlock_unlock(&resource_lock);
> 
> 	return -1;
> }
> 
> 
>+int
>+group_property(char *groupname, char *property, char *ret_val, size_t len)
>+{
>+	int ret = -1;
>+	pthread_rwlock_rdlock(&resource_lock);
>+	ret = _group_property(groupname, property, ret_val, len);
>+	pthread_rwlock_unlock(&resource_lock);
>+	return ret;
>+}
>+	
>+
> /**
>   Send the state of a resource group to a given file descriptor.
> 
>diff -ur rgmanager-current/src/daemons/rg_state.c rgmanager/src/daemons/rg_state.c
>--- rgmanager-current/src/daemons/rg_state.c	2008-07-24 16:24:46.000000000 -0400
>+++ rgmanager/src/daemons/rg_state.c	2008-07-24 17:12:44.000000000 -0400
>@@ -470,6 +470,8 @@
>  *			2 = DO NOT stop service, return 0 (success)
>  *                      3 = DO NOT stop service, return RG_EFORWARD
>  *			4 = DO NOT stop service, return RG_EAGAIN
>+ *			5 = DO NOT stop service, mark stopped and return
>+ *			    RG_SUCCESS (0)
>  */
> int
> svc_advise_stop(rg_state_t *svcStatus, char *svcName, int req)
>@@ -529,9 +531,10 @@
> 
> 		/*
> 		   Service is marked as running but node is down.
>-		   Doesn't make much sense to stop it.
>+		   Doesn't make much sense to stop it - but we need
>+		   to mark it stopped
> 		 */
>-		ret = 2;
>+		ret = 5;
> 		break;
> 
> 	case RG_STATE_ERROR:
>@@ -1278,6 +1281,16 @@
> 		clulog(LOG_DEBUG, "Unable to stop RG %s in %s state\n",
> 		       svcName, rg_state_str(svcStatus.rs_state));
> 		return RG_EFAIL;
>+	case 5:
>+		/* Mark stopped, but do not do anything */
>+		svcStatus.rs_last_owner = svcStatus.rs_owner;
>+		svcStatus.rs_owner = 0;
>+		svcStatus.rs_state = RG_STATE_STOPPED;
>+		if (set_rg_state(svcName, &svcStatus) != 0) {
>+			rg_unlock(&lockp);
>+			return RG_EFAIL;
>+		}
>+		/* FALLTHROUGH */
> 	case 2:
> 		rg_unlock(&lockp);
> 		return RG_ESUCCESS;
>diff -ur rgmanager-current/src/daemons/service_op.c rgmanager/src/daemons/service_op.c
>--- rgmanager-current/src/daemons/service_op.c	2008-07-24 16:24:46.000000000 -0400
>+++ rgmanager/src/daemons/service_op.c	2008-07-24 17:13:02.000000000 -0400
>@@ -150,8 +150,17 @@
> 
> 	if (get_service_state_internal(svcName, &svcStatus) < 0)
> 		return RG_EFAIL;
>-	if (svcStatus.rs_owner > 0)
>-		msgtarget = svcStatus.rs_owner;
>+	if (svcStatus.rs_owner > 0) {
>+		if (member_online(svcStatus.rs_owner)) {
>+			msgtarget = svcStatus.rs_owner;
>+		} else {
>+			/* If the owner is not online, 
>+			   mark the service as 'stopped' but
>+			   otherwise, do nothing.
>+			 */
>+			return svc_stop(svcName, RG_STOP);
>+		}
>+	}
> 
> 	if (msg_open(MSG_CLUSTER, msgtarget, RG_PORT, &ctx, 2)< 0) {
> 		clulog(LOG_ERR,
>diff -ur rgmanager-current/src/daemons/slang_event.c rgmanager/src/daemons/slang_event.c
>--- rgmanager-current/src/daemons/slang_event.c	2008-07-24 16:24:46.000000000 -0400
>+++ rgmanager/src/daemons/slang_event.c	2008-07-24 16:45:14.000000000 -0400
>@@ -595,24 +595,12 @@
> void
> sl_nodes_online(void)
> {
>-	int i, *nodes, nodecount = 0;
>+	int x, *nodes = NULL, nodecount = 0;
> 
>-	cluster_member_list_t *membership = member_list();
>-	if (!membership)
>-		return;
>-	nodes = malloc(sizeof(int) * membership->cml_count);
>-	if (!nodes)
>+	x = member_online_set(&nodes, &nodecount);
>+	if (x < 0 || !nodes || !nodecount)
> 		return;
> 
>-	nodecount = 0;
>-	for (i = 0; i < membership->cml_count; i++) {
>-		if (membership->cml_members[i].cn_member &&
>-		    membership->cml_members[i].cn_nodeid != 0) {
>-			nodes[nodecount] = membership->cml_members[i].cn_nodeid;
>-			++nodecount;
>-		}
>-	}
>-	free_member_list(membership);
> 	push_int_array(nodes, nodecount);
> 	free(nodes);
> }
>diff -ur rgmanager-current/src/resources/default_event_script.sl rgmanager/src/resources/default_event_script.sl
>--- rgmanager-current/src/resources/default_event_script.sl	2008-07-24 16:24:46.000000000 -0400
>+++ rgmanager/src/resources/default_event_script.sl	2008-07-25 10:36:42.000000000 -0400
>@@ -31,7 +31,8 @@
> 
> 	len = length(node_list);
> 	if (len == 0) {
>-		debug(service, " is not runnable");
>+		notice(service, " is not runnable - restricted domain offline");
>+		()=service_stop(service);
> 		return ERR_DOMAIN;
> 	}
>

Actions: View | Diff

Attachments on bug 435466: 312662