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

[Cluster-devel] cluster/rgmanager ChangeLog include/reslist.h ...



CVSROOT:	/cvs/cluster
Module name:	cluster
Branch: 	RHEL5
Changes by:	lhh sourceware org	2007-08-02 14:46:52

Modified files:
	rgmanager      : ChangeLog 
	rgmanager/include: reslist.h 
	rgmanager/src/daemons: groups.c main.c nodeevent.c restree.c 
	                       rg_forward.c rg_state.c 
	rgmanager/src/resources: vm.sh 

Log message:
	Fix #248727, round 2

Patches:
http://sourceware.org/cgi-bin/cvsweb.cgi/cluster/rgmanager/ChangeLog.diff?cvsroot=cluster&only_with_tag=RHEL5&r1=1.31.2.23&r2=1.31.2.24
http://sourceware.org/cgi-bin/cvsweb.cgi/cluster/rgmanager/include/reslist.h.diff?cvsroot=cluster&only_with_tag=RHEL5&r1=1.15.2.5&r2=1.15.2.6
http://sourceware.org/cgi-bin/cvsweb.cgi/cluster/rgmanager/src/daemons/groups.c.diff?cvsroot=cluster&only_with_tag=RHEL5&r1=1.25.2.11&r2=1.25.2.12
http://sourceware.org/cgi-bin/cvsweb.cgi/cluster/rgmanager/src/daemons/main.c.diff?cvsroot=cluster&only_with_tag=RHEL5&r1=1.34.2.7&r2=1.34.2.8
http://sourceware.org/cgi-bin/cvsweb.cgi/cluster/rgmanager/src/daemons/nodeevent.c.diff?cvsroot=cluster&only_with_tag=RHEL5&r1=1.4.2.4&r2=1.4.2.5
http://sourceware.org/cgi-bin/cvsweb.cgi/cluster/rgmanager/src/daemons/restree.c.diff?cvsroot=cluster&only_with_tag=RHEL5&r1=1.23.2.9&r2=1.23.2.10
http://sourceware.org/cgi-bin/cvsweb.cgi/cluster/rgmanager/src/daemons/rg_forward.c.diff?cvsroot=cluster&only_with_tag=RHEL5&r1=1.8.2.2&r2=1.8.2.3
http://sourceware.org/cgi-bin/cvsweb.cgi/cluster/rgmanager/src/daemons/rg_state.c.diff?cvsroot=cluster&only_with_tag=RHEL5&r1=1.24.2.11&r2=1.24.2.12
http://sourceware.org/cgi-bin/cvsweb.cgi/cluster/rgmanager/src/resources/vm.sh.diff?cvsroot=cluster&only_with_tag=RHEL5&r1=1.1.2.6&r2=1.1.2.7

--- cluster/rgmanager/ChangeLog	2007/07/31 17:54:54	1.31.2.23
+++ cluster/rgmanager/ChangeLog	2007/08/02 14:46:51	1.31.2.24
@@ -1,3 +1,16 @@
+2007-08-02 Lon Hohberger <lhh at redhat.com>
+	* general: More fixes around #248727
+	* include/reslist.h, src/daemons/restree.c: Make last-value be
+	returned or resources which have been checked recently
+	* src/daemons/groups.c: Make VMs use migrate semantics instead of
+	relocate semantics when employing failover domain rules
+	* src/daemons/nodeevent.c: Fix VMs ending up on wrong nodes when
+	simultaneous boot occurs
+	* src/daemons/rg_forward.c: Fix erroneous timeout
+	* src/daemons/rg_state.c: Handle RG_STATE_MIGRATE in svc_advise_*
+	Handle certain migration failures.
+	* src/resources/vm.sh: Handle certain migration failures
+
 2007-07-31 Lon Hohberger <lhh at redhat.com>
 	* general: Make VMs not change state when added/removed from the
 	cluster config or bounce services/VMs when minor config changes
--- cluster/rgmanager/include/reslist.h	2007/07/31 17:54:54	1.15.2.5
+++ cluster/rgmanager/include/reslist.h	2007/08/02 14:46:51	1.15.2.6
@@ -128,6 +128,10 @@
 	resource_act_t	*rn_actions;
 	int	rn_state; /* State of this instance of rn_resource */
 	int	rn_flags;
+	int	rn_last_status;
+	int 	rn_last_depth;
+	int	rn_checked;
+	int	rn_pad;
 } resource_node_t;
 
 typedef struct _fod_node {
--- cluster/rgmanager/src/daemons/groups.c	2007/07/31 17:54:54	1.25.2.11
+++ cluster/rgmanager/src/daemons/groups.c	2007/08/02 14:46:51	1.25.2.12
@@ -500,13 +500,14 @@
 consider_relocate(char *svcName, rg_state_t *svcStatus, uint32_t nodeid,
 		  cluster_member_list_t *membership)
 {
-	int a, b;
+	int a, b, req = RG_RELOCATE;
 
 	/*
 	   Service must be running locally in order to consider for
 	   a relocate
 	 */
-	if (svcStatus->rs_state != RG_STATE_STARTED ||
+	if ((svcStatus->rs_state != RG_STATE_STARTING &&
+	    svcStatus->rs_state != RG_STATE_STARTED) ||
 	    svcStatus->rs_owner != my_id())
 		return;
 
@@ -526,11 +527,16 @@
 	if (a <= b)
 		return;
 
-	clulog(LOG_DEBUG, "Relocating group %s to better node %s\n",
+	if (group_migratory(svcName, 1)) {
+		req = RG_MIGRATE;
+	}
+
+	clulog(LOG_NOTICE, "%s %s to better node %s\n",
+	       req==RG_MIGRATE ? "Migrating":"Relocating",
 	       svcName,
 	       memb_id_to_name(membership, nodeid));
 
-	rt_enqueue_request(svcName, RG_RELOCATE, NULL, 0, nodeid, 0, 0);
+	rt_enqueue_request(svcName, req, NULL, 0, nodeid, 0, 0);
 }
 
 
--- cluster/rgmanager/src/daemons/main.c	2007/07/24 13:53:08	1.34.2.7
+++ cluster/rgmanager/src/daemons/main.c	2007/08/02 14:46:51	1.34.2.8
@@ -43,7 +43,7 @@
 #ifdef WRAP_THREADS
 void dump_thread_states(FILE *);
 #endif
-int configure_logging(int ccsfd, int debug);
+int configure_rgmanager(int ccsfd, int debug);
 
 void node_event(int, int, int, int);
 void node_event_q(int, int, int, int);
@@ -730,7 +730,7 @@
 
 	if (need_reconfigure || check_config_update()) {
 		need_reconfigure = 0;
-		configure_logging(-1, 0);
+		configure_rgmanager(-1, 0);
 		init_resource_groups(1);
 		return 0;
 	}
@@ -789,7 +789,7 @@
  * Configure logging based on data in cluster.conf
  */
 int
-configure_logging(int ccsfd, int dbg)
+configure_rgmanager(int ccsfd, int dbg)
 {
 	char *v;
 	char internal = 0;
@@ -812,6 +812,12 @@
 		free(v);
 	}
 
+	if (ccs_get(ccsfd, "/cluster/rm/@transition_throttling", &v) == 0) {
+		if (!dbg)
+			set_transition_throttling(atoi(v));
+		free(v);
+	}
+
 	if (internal)
 		ccs_disconnect(ccsfd);
 
@@ -956,7 +962,7 @@
 	   We know we're quorate.  At this point, we need to
 	   read the resource group trees from ccsd.
 	 */
-	configure_logging(-1, debug);
+	configure_rgmanager(-1, debug);
 	clulog(LOG_NOTICE, "Resource Group Manager Starting\n");
 
 	if (init_resource_groups(0) != 0) {
--- cluster/rgmanager/src/daemons/nodeevent.c	2007/07/24 13:53:08	1.4.2.4
+++ cluster/rgmanager/src/daemons/nodeevent.c	2007/08/02 14:46:51	1.4.2.5
@@ -42,6 +42,7 @@
 #endif
 static nevent_t *event_queue = NULL;
 static pthread_t ne_thread = 0;
+static int transition_throttling = 5;
 int ne_queue_request(int local, int nodeid, int state);
 
 void hard_exit(void);
@@ -53,6 +54,15 @@
 extern int shutdown_pending;
 
 
+void
+set_transition_throttling(int nsecs)
+{
+	if (nsecs < 0)
+		nsecs = 0;
+	transition_throttling = nsecs;
+}
+
+
 /**
   Called to handle the transition of a cluster member from up->down or
   down->up.  This handles initializing services (in the local node-up case),
@@ -88,11 +98,16 @@
 		if (shutdown_pending) {
 			clulog(LOG_NOTICE, "Processing delayed exit signal\n");
 			running = 0;
+			return;
 		}
 		setup_signal(SIGINT, flag_shutdown);
 		setup_signal(SIGTERM, flag_shutdown);
 		setup_signal(SIGHUP, flag_reconfigure);
 
+		/* Let things settle if we're booting multiple */
+		if (transition_throttling)
+			sleep(transition_throttling);
+
 		eval_groups(1, nodeID, 1);
 		return;
 	}
--- cluster/rgmanager/src/daemons/restree.c	2007/07/31 17:54:54	1.23.2.9
+++ cluster/rgmanager/src/daemons/restree.c	2007/08/02 14:46:51	1.23.2.10
@@ -665,8 +665,10 @@
 			}
 		}
 		/* No resource rule matching the child?  Press on... */
-		if (!flags)
+		if (!flags) {
+			free(ref);
 			continue;
+		}
 
 		flags = 0;
 		/* Don't descend on anything we should have already picked
@@ -686,11 +688,9 @@
 			break;
 		}
 
-		if (flags == 2) {
-			free(ref);
-			continue;
-		}
 		free(ref);
+		if (flags == 2)
+			continue;
 
 		x = 1;
 		switch(do_load_resource(ccsfd, tok, childrule, tree,
@@ -1035,12 +1035,21 @@
 	}
 
 	/* No check levels ready at the moment. */
-	if (idx == -1)
+	if (idx == -1) {
+		if (node->rn_checked)
+			return node->rn_last_status;
 		return 0;
+	}
 
-       node->rn_actions[idx].ra_last = now;
-	if ((x = res_exec(node, RS_STATUS, NULL,
-                         node->rn_actions[idx].ra_depth)) == 0)
+
+	node->rn_actions[idx].ra_last = now;
+	x = res_exec(node, RS_STATUS, NULL, node->rn_actions[idx].ra_depth);
+
+	node->rn_last_status = x;
+	node->rn_last_depth = node->rn_actions[idx].ra_depth;
+	node->rn_checked = 1;
+
+	if (x == 0)
 		return 0;
 
 	if (!has_recover)
@@ -1101,14 +1110,18 @@
 
 	now = res->r_started;
 
-       for (; node->rn_actions[x].ra_name; x++) {
+	for (; node->rn_actions[x].ra_name; x++) {
 
-               if (strcmp(node->rn_actions[x].ra_name, "monitor") &&
-                   strcmp(node->rn_actions[x].ra_name, "status"))
+		if (strcmp(node->rn_actions[x].ra_name, "monitor") &&
+		    strcmp(node->rn_actions[x].ra_name, "status"))
 			continue;
 
-               node->rn_actions[x].ra_last = now;
+		node->rn_actions[x].ra_last = now;
 	}
+
+	node->rn_checked = 0;
+	node->rn_last_status = 0;
+	node->rn_last_depth = 0;
 }
 
 
--- cluster/rgmanager/src/daemons/rg_forward.c	2007/07/24 13:53:08	1.8.2.2
+++ cluster/rgmanager/src/daemons/rg_forward.c	2007/08/02 14:46:51	1.8.2.3
@@ -122,10 +122,12 @@
 				m = NULL;
 				continue;
 			}
-			goto out_fail;
+
+			if (ret == 0)
+				continue;
 		}
 		break;
-	} while(++retries < 60); /* old 60 second rule */
+	} while(++retries < 60); /* old 600 second rule */
 
 	swab_SmMessageSt(&msg);
 
--- cluster/rgmanager/src/daemons/rg_state.c	2007/07/24 13:53:08	1.24.2.11
+++ cluster/rgmanager/src/daemons/rg_state.c	2007/08/02 14:46:51	1.24.2.12
@@ -35,6 +35,7 @@
 #include <ccs.h>
 #include <rg_queue.h>
 #include <msgsimple.h>
+#include <res-ocf.h>
 
 /* XXX - copied :( */
 #define cn_svccount cn_address.cna_address[0] /* Theses are uint8_t size */
@@ -467,6 +468,7 @@
 	case RG_STATE_CHECK:
 	case RG_STATE_STARTING:
 	case RG_STATE_RECOVER:
+	case RG_STATE_MIGRATE:
 		if ((svcStatus->rs_owner != my_id()) &&
 		    memb_online(membership, svcStatus->rs_owner)) {
 			/*
@@ -583,6 +585,10 @@
 		       "#43: Service %s has failed; can not start.\n",
 		       svcName);
 		break;
+
+	case RG_STATE_MIGRATE:
+		ret = 4;
+		break;
 		
 	case RG_STATE_STOPPING:
 	case RG_STATE_STARTED:
@@ -892,16 +898,60 @@
        
 	ret = group_migrate(svcName, target);
 
-	if (ret == -1 || ret > 0) {
+	switch(ret) {
+	default:
+	case -1:
+	case OCF_RA_ERROR:
+		svc_fail(svcName);
 		/* XXX run svc_status again here to see if it's still
 		   healthy; if it is, don't FAIL it; it could be that
 		   the target node simply died; in this case, set status
 		   back to started */
-		/* if ret > 0 { svc_status... */
-		svc_fail(svcName);
+		return RG_EFAIL;
+		break;
+	case OCF_RA_NOT_RUNNING:
+		/* For these two, the VM was either not running or 
+		   migration is simply impossible. */
+		/* Don't mark the service as failed; since it's either
+		   recoverable or still running. */
 		ret = RG_EFAIL;
+		break;
+	case OCF_RA_NOT_CONFIGURED:
+		ret = RG_EINVAL;
+		break;
+	case 0:
+		return 0;
 	}
 
+	/* Ok, we've hit a recoverable condition.  Since VMs and migratory
+	   services are ... well, migratable, we can just flip the state
+	   back to 'started' and error checking will fix it later. */
+	if (rg_lock(svcName, &lockp) < 0) {
+		clulog(LOG_ERR, "#45: Unable to obtain cluster lock: %s\n",
+		       strerror(errno));
+		return ret;
+	}
+
+	if (get_rg_state(svcName, &svcStatus) != 0) {
+		rg_unlock(&lockp);
+		clulog(LOG_ERR, "#46: Failed getting status for RG %s\n",
+		       svcName);
+		return ret;
+	}
+
+	if (svcStatus.rs_last_owner != my_id() ||
+	    svcStatus.rs_owner != target ||
+	    svcStatus.rs_state != RG_STATE_MIGRATE) {
+		rg_unlock(&lockp);
+		return ret;
+	}
+
+	svcStatus.rs_owner = my_id();
+	svcStatus.rs_state = RG_STATE_STARTED;
+
+	set_rg_state(svcName, &svcStatus);
+	rg_unlock(&lockp);
+
 	return ret;
 }
 
@@ -954,7 +1004,8 @@
 		}
 
 		msg_send(&ctx, &msgp, sizeof(msgp));
-		msg_receive(&ctx, &response, sizeof (response), 5);
+		if (msg_receive(&ctx, &response, sizeof (response), 5) != sizeof(response))
+			goto cont;;
 
 		swab_SmMessageSt(&response);
 		if (response.sm_data.d_ret == RG_SUCCESS)
@@ -962,6 +1013,7 @@
 		else
 			ret = -1;
 
+cont:
 		msg_close(&ctx);
 	}
 
@@ -1937,7 +1989,7 @@
 	allowed_nodes = member_list();
 
 	while (memb_count(allowed_nodes)) {
-		target = best_target_node(allowed_nodes, -1,
+		target = best_target_node(allowed_nodes, 0,
 		    			  svcName, 1);
 		if (target == me) {
 		      	ret = handle_start_remote_req(svcName, request);
@@ -1947,7 +1999,7 @@
 			ret = RG_EFAIL;
 			goto out;
 		} else {
-			ret = relocate_service(svcName, request, target);
+			ret = relocate_service(svcName, RG_START_REMOTE, target);
 		}
 
 		switch(ret) {
--- cluster/rgmanager/src/resources/vm.sh	2007/07/31 17:54:55	1.1.2.6
+++ cluster/rgmanager/src/resources/vm.sh	2007/08/02 14:46:52	1.1.2.7
@@ -22,6 +22,8 @@
 
 export PATH
 
+. $(dirname $0)/ocf-shellfuncs
+
 #
 # Virtual Machine start/stop script (requires the xm command)
 #
@@ -375,8 +377,22 @@
 migrate()
 {
 	declare target=$1
+	declare errstr rv
+	
+	err=$(xm migrate $OCF_RESKEY_name $target 2>&1 | head -1)
+	rv=$?
+
+	if [ $rv -ne 0 ]; then
+		if [ "$err" != "${err/does not exist/}" ]; then
+			ocf_log warn "Trying to migrate '$OCF_RESKEY_name' - domain does not exist"
+			return $OCF_NOT_RUNNING
+		fi
+		if [ "$err" != "${err/Connection refused/}" ]; then
+			ocf_log warn "Trying to migrate '$OCF_RESKEY_name' - connect refused"
+			return $OCF_ERR_CONFIGURED
+		fi
+	fi
 
-	xm migrate $OCF_RESKEY_name $target
 	return $?
 }
 


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