Gitweb: http://git.fedorahosted.org/git/cluster.git?p=cluster.git;a=commitdiff;h=e1e... Commit: e1e194a8337d0ced672e9b6fd08f4ca0d02fb52e Parent: 952090aca003968b8e9b47baa865cd6d603b061d Author: Lon Hohberger lhh@redhat.com AuthorDate: Mon Dec 21 17:54:37 2009 -0500 Committer: Lon Hohberger lhh@redhat.com CommitterDate: Tue Dec 22 10:16:04 2009 -0500
rgmanager: Fix event generation with central_processing
This patch fixes event generation and processing when a node dies. Effectively, what was happening is that when a node failed and was fenced, no events for the dead services on that host were generated. This led to dependent services not restarting correctly in many cases.
Resolves: rhbz#523999
Signed-off-by: Lon Hohberger lhh@redhat.com --- rgmanager/include/resgroup.h | 2 ++ rgmanager/src/daemons/main.c | 2 +- rgmanager/src/daemons/rg_forward.c | 4 ++-- rgmanager/src/daemons/rg_state.c | 9 ++++++--- rgmanager/src/daemons/service_op.c | 2 +- rgmanager/src/resources/default_event_script.sl | 8 ++++++++ 6 files changed, 20 insertions(+), 7 deletions(-)
diff --git a/rgmanager/include/resgroup.h b/rgmanager/include/resgroup.h index a4b53fe..ec3e67d 100644 --- a/rgmanager/include/resgroup.h +++ b/rgmanager/include/resgroup.h @@ -166,6 +166,8 @@ int get_rg_state_local(char *servicename, rg_state_t *svcblk); uint32_t best_target_node(cluster_member_list_t *allowed, uint32_t owner, char *rg_name, int lock);
+extern int cluster_timeout; + #ifdef DEBUG int _rg_lock_dbg(char *, struct dlm_lksb *, char *, int); #define rg_lock(name, p) _rg_lock_dbg(name, p, __FILE__, __LINE__) diff --git a/rgmanager/src/daemons/main.c b/rgmanager/src/daemons/main.c index 601e7d0..93ecff3 100644 --- a/rgmanager/src/daemons/main.c +++ b/rgmanager/src/daemons/main.c @@ -63,6 +63,7 @@ int32_t master_event_callback(char *key, uint64_t viewno, void *data, uint32_t d int node_has_fencing(int nodeid); int fence_domain_joined(void);
+int cluster_timeout = 10; int shutdown_pending = 0, running = 1, need_reconfigure = 0; char debug = 0; /* XXX* */ static int signalled = 0; @@ -1022,7 +1023,6 @@ main(int argc, char **argv) msgctx_t *local_ctx; pthread_t th; cman_handle_t clu = NULL; - int cluster_timeout = 10;
while ((rv = getopt(argc, argv, "wfdN")) != EOF) { switch (rv) { diff --git a/rgmanager/src/daemons/rg_forward.c b/rgmanager/src/daemons/rg_forward.c index c827c62..bf32f1d 100644 --- a/rgmanager/src/daemons/rg_forward.c +++ b/rgmanager/src/daemons/rg_forward.c @@ -103,7 +103,7 @@ forwarding_thread(void *arg) build_message(&msg, req->rr_request, req->rr_group, req->rr_target, req->rr_arg0, req->rr_arg1);
- if (msg_open(MSG_CLUSTER, rgs.rs_owner, RG_PORT, ctx, 10) < 0) { + if (msg_open(MSG_CLUSTER, rgs.rs_owner, RG_PORT, ctx, 2 * cluster_timeout) < 0) { clulog(LOG_DEBUG, "FW: Failed to open channel to %d CTX: %p\n", rgs.rs_owner, ctx); goto out_fail; @@ -202,7 +202,7 @@ forwarding_thread_v2(void *arg) strerror(errno)); goto out_fail; } - if (msg_open(MSG_CLUSTER, target, RG_PORT, ctx, 10) < 0) { + if (msg_open(MSG_CLUSTER, target, RG_PORT, ctx, 2 * cluster_timeout) < 0) { clulog(LOG_DEBUG, "FW: Failed to open channel to %d CTX: %p\n", target, ctx); goto out_fail; diff --git a/rgmanager/src/daemons/rg_state.c b/rgmanager/src/daemons/rg_state.c index d68ce9b..ce5189c 100644 --- a/rgmanager/src/daemons/rg_state.c +++ b/rgmanager/src/daemons/rg_state.c @@ -1031,7 +1031,7 @@ get_new_owner(char *svcName) continue;
if (msg_open(MSG_CLUSTER, membership->cml_members[x].cn_nodeid, - RG_PORT, &ctx, 2) < 0) { + RG_PORT, &ctx, 2 * cluster_timeout) < 0) { /* failed to open: better to claim false successful status rather than claim a failure and possibly end up with a service on >1 node */ @@ -1315,7 +1315,10 @@ _svc_stop(char *svcName, int req, int recover, uint32_t newstate) rg_unlock(&lockp); return RG_EFAIL; } - /* FALLTHROUGH */ + rg_unlock(&lockp); + broadcast_event(svcName, RG_STATE_STOPPED, + -1, svcStatus.rs_last_owner); + return RG_ESUCCESS; case 2: rg_unlock(&lockp); return RG_ESUCCESS; @@ -1615,7 +1618,7 @@ svc_start_remote(char *svcName, int request, uint32_t target) msg_relo.sm_data.d_svcOwner = target; /* Open a connection to the other node */
- if (msg_open(MSG_CLUSTER, target, RG_PORT, &ctx, 2)< 0) { + if (msg_open(MSG_CLUSTER, target, RG_PORT, &ctx, 2 * cluster_timeout)< 0) { clulog(LOG_ERR, "#58: Failed opening connection to member #%d\n", target); diff --git a/rgmanager/src/daemons/service_op.c b/rgmanager/src/daemons/service_op.c index 8206d50..8e1cc91 100644 --- a/rgmanager/src/daemons/service_op.c +++ b/rgmanager/src/daemons/service_op.c @@ -162,7 +162,7 @@ service_op_stop(char *svcName, int do_disable, int event_type) } }
- if (msg_open(MSG_CLUSTER, msgtarget, RG_PORT, &ctx, 2)< 0) { + if (msg_open(MSG_CLUSTER, msgtarget, RG_PORT, &ctx, 2 * cluster_timeout)< 0) { clulog(LOG_ERR, "#58: Failed opening connection to member #%d\n", my_id()); diff --git a/rgmanager/src/resources/default_event_script.sl b/rgmanager/src/resources/default_event_script.sl index 7809b20..25347eb 100644 --- a/rgmanager/src/resources/default_event_script.sl +++ b/rgmanager/src/resources/default_event_script.sl @@ -157,6 +157,14 @@ define move_or_start(service, node_list)
(,,, owner, state) = service_status(service); debug("Evaluating ", service, " state=", state, " owner=", owner); + if ((event_type == EVENT_NODE) and (node_id == owner) and + (node_state == NODE_OFFLINE)) { + info("Marking service ", service, " on down member ", + owner, " as stopped"); + if (service_stop(service) < 0) { + return ERR_ABORT; + } + }
len = length(node_list); if (len == 0) {
cluster-commits@lists.fedorahosted.org
-
Lon Hohberger