mds: send resolve messages after all MDS reach resolve stage

Current code sends resolve messages when resolving MDS set changes.
There is no need to send resolve messages when some MDS leave the
resolve stage. Sending message while some MDS are replaying is also
not very useful.

Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com>

diff --git a/src/mds/MDCache.cc b/src/mds/MDCache.cc
index c6166521952abc77f0ee565664417752c4066139..cd24fc2f3a2108276193d0e4dacc489522132c7d 100644 (file)
--- a/src/mds/MDCache.cc
+++ b/src/mds/MDCache.cc
@@ -2740,6 +2740,10 @@ void MDCache::handle_resolve(MMDSResolve *m)
   int from = m->get_source().num();
 
   if (mds->get_state() < MDSMap::STATE_RESOLVE) {
+    if (mds->get_want_state() == CEPH_MDS_STATE_RESOLVE) {
+      mds->wait_for_resolve(new C_MDS_RetryMessage(mds, m));
+      return;
+    }
     // wait until we reach the resolve stage!
     m->put();
     return;

diff --git a/src/mds/MDS.cc b/src/mds/MDS.cc
index bd4bed91585c85b5d8102dd181fe055d63b408b6..4153417ad11bfa777b27d446eec9f50aeeb93250 100644 (file)
--- a/src/mds/MDS.cc
+++ b/src/mds/MDS.cc
@@ -974,14 +974,12 @@ void MDS::handle_mds_map(MMDSMap *m)
   // RESOLVE
   // is someone else newly resolving?
   if (is_resolve() || is_rejoin() || is_clientreplay() || is_active() || is_stopping()) {
-    set<int> oldresolve, resolve;
-    oldmap->get_mds_set(oldresolve, MDSMap::STATE_RESOLVE);
-    mdsmap->get_mds_set(resolve, MDSMap::STATE_RESOLVE);
-    if (oldresolve != resolve) {
-      dout(10) << " resolve set is " << resolve << ", was " << oldresolve << dendl;
+    if (!oldmap->is_resolving() && mdsmap->is_resolving()) {
+      set<int> oldresolve, resolve;
+      mdsmap->get_mds_set(resolve, MDSMap::STATE_RESOLVE);
+      dout(10) << " resolve set is " << resolve << dendl;
       calc_recovery_set();
-      if (!mdsmap->is_any_failed())
-       mdcache->send_resolves();
+      mdcache->send_resolves();
     }
   }
   
@@ -1410,6 +1408,7 @@ void MDS::resolve_start()
   reopen_log();
 
   mdcache->resolve_start();
+  finish_contexts(g_ceph_context, waiting_for_resolve);
 }
 void MDS::resolve_done()
 {

diff --git a/src/mds/MDS.h b/src/mds/MDS.h
index a90587e3aa224eb56375345281abb67c948132de..f61ad8ddac55fa4406231cc8279ade32a82538a7 100644 (file)
--- a/src/mds/MDS.h
+++ b/src/mds/MDS.h
@@ -196,7 +196,7 @@ class MDS : public Dispatcher {
   int state;         // my confirmed state
   int want_state;    // the state i want
 
-  list<Context*> waiting_for_active, waiting_for_replay, waiting_for_reconnect;
+  list<Context*> waiting_for_active, waiting_for_replay, waiting_for_reconnect, waiting_for_resolve;
   list<Context*> replay_queue;
   map<int, list<Context*> > waiting_for_active_peer;
   list<Message*> waiting_for_nolaggy;
@@ -219,6 +219,9 @@ class MDS : public Dispatcher {
   void wait_for_reconnect(Context *c) {
     waiting_for_reconnect.push_back(c);
   }
+  void wait_for_resolve(Context *c) {
+    waiting_for_resolve.push_back(c);
+  }
   void wait_for_mdsmap(epoch_t e, Context *c) {
     waiting_for_mdsmap[e].push_back(c);
   }

diff --git a/src/mds/MDSMap.h b/src/mds/MDSMap.h
index 3a83ed88b08e34abbba00e709aa894fbae7a35b2..47c2c52d23d38c375d158c0cc8d3c4d6fc6eecd8 100644 (file)
--- a/src/mds/MDSMap.h
+++ b/src/mds/MDSMap.h
@@ -449,6 +449,12 @@ public:
   bool is_any_failed() {
     return failed.size();
   }
+  bool is_resolving() {
+    return
+      get_num_mds(STATE_RESOLVE) > 0 &&
+      get_num_mds(STATE_REPLAY) == 0 &&
+      failed.empty();
+  }
   bool is_rejoining() {  
     // nodes are rejoining cache state
     return