[Cluster-devel] [PATCH] rgmanager: Fix reference count handling (RHEL5)
Lon Hohberger
lhh at redhat.com
Tue Apr 26 15:42:30 UTC 2011
rgmanager feeds reference counts to resource-agents so they can track
when to actually clean up (i.e. when there are no other references).
A problem was found where rgmanager was incorrectly presenting a higher
reference count than expected while stopping, preventing multi-instance
resources (specifically clusterfs.sh) from cleaning up.
Resolves: rhbz#692771
Signed-off-by: Lon Hohberger <lhh at redhat.com>
---
rgmanager/src/daemons/restree.c | 12 ++++++++++--
1 files changed, 10 insertions(+), 2 deletions(-)
diff --git a/rgmanager/src/daemons/restree.c b/rgmanager/src/daemons/restree.c
index 9652c6e..3892ab9 100644
--- a/rgmanager/src/daemons/restree.c
+++ b/rgmanager/src/daemons/restree.c
@@ -330,6 +330,7 @@ res_exec(resource_node_t *node, int op, const char *arg, int depth)
int childpid, pid;
int ret = 0;
int act_index;
+ int inc = node->rn_resource->r_incarnations;
time_t sleeptime = 0, timeout = 0;
char **env = NULL;
resource_t *res = node->rn_resource;
@@ -353,8 +354,15 @@ res_exec(resource_node_t *node, int op, const char *arg, int depth)
if (!(node->rn_flags & RF_ENFORCE_TIMEOUTS))
timeout = node->rn_actions[act_index].ra_timeout;
+ /* rgmanager ref counts are designed to track *other* incarnations
+ on the host. So, if we're started/failed, the RA should not count
+ this incarnation */
+ if (inc && (node->rn_state == RES_STARTED ||
+ node->rn_state == RES_FAILED))
+ --inc;
+
#ifdef DEBUG
- env = build_env(node, depth, node->rn_resource->r_incarnations, (int)timeout);
+ env = build_env(node, depth, inc, (int)timeout);
if (!env)
return -errno;
#endif
@@ -380,7 +388,7 @@ res_exec(resource_node_t *node, int op, const char *arg, int depth)
#endif
#ifndef DEBUG
- env = build_env(node, depth, node->rn_resource->r_incarnations, (int)timeout);
+ env = build_env(node, depth, inc, (int)timeout);
#endif
if (!env)
--
1.7.3.4
More information about the Cluster-devel
mailing list