Gitweb: http://git.fedorahosted.org/git/?p=cluster.git;a=commitdiff;h=03e2215bd27... Commit: 03e2215bd277fd79b8a6ee70a49de711e0f343ad Parent: e080a6249f25a3bc18a6c9028c9da06bf4ef14c6 Author: Ryan McCabe <rmccabe(a)redhat.com&gt; AuthorDate: Mon Jul 16 10:57:28 2012 -0400 Committer: Ryan McCabe <rmccabe(a)redhat.com&gt; CommitterDate: Mon Jul 16 10:57:28 2012 -0400 rgmanager: Fix for services stuck in recovery Patch from John Ruemker <jruemker(a)redhat.com&gt;: "When starting rgmanager throughout the cluster around the same time, multiple nodes may end up acting as the "root" for a particular service. If that service happens to fail on startup, you can end up with each of those nodes sending remote-start requests around the cluster. Eventually the service will get stuck in a recovering state, and cannot be modified in any way with clusvcadm. The only remedy we've found is to kill rgmanager and start it back up. Acked-by: Lon Hohberger <lhh(a)redhat.com&gt; Signed-off-by: Ryan McCabe <rmccabe(a)redhat.com&gt; --- rgmanager/src/daemons/groups.c | 6 ++++-- 1 files changed, 4 insertions(+), 2 deletions(-) diff --git a/rgmanager/src/daemons/groups.c b/rgmanager/src/daemons/groups.c index 4a72cb7..4537cc8 100644 --- a/rgmanager/src/daemons/groups.c +++ b/rgmanager/src/daemons/groups.c @@ -731,7 +731,8 @@ eval_groups(int local, uint32_t nodeid, int nodeStatus) (svcStatus.rs_state == RG_STATE_STARTED || svcStatus.rs_state == RG_STATE_RECOVER || svcStatus.rs_state == RG_STATE_STARTING || - svcStatus.rs_state == RG_STATE_STOPPING )) { + svcStatus.rs_state == RG_STATE_STOPPING || + svcStatus.rs_state == RG_STATE_ERROR)) { logt_print(LOG_DEBUG, "Marking %s on down member %d as stopped", @@ -773,7 +774,8 @@ eval_groups(int local, uint32_t nodeid, int nodeStatus) /* Disabled/failed/in recovery? Do nothing */ if ((svcStatus.rs_state == RG_STATE_DISABLED) || (svcStatus.rs_state == RG_STATE_FAILED) || - (svcStatus.rs_state == RG_STATE_RECOVER)) { + (svcStatus.rs_state == RG_STATE_RECOVER) || + (svcStatus.rs_state == RG_STATE_ERROR)) { continue; }