Elector: send an OP_NAK MMonElection to old peers who support it

Only new monitors support receiving OP_NAK from a peer without crashing, but
when we add new required features in the future, our monitors can accept
an OP_NAK message which tells them what features they're missing. Then they
will print out an error message and shut down.
(Unfortunately, doing a clean shutdown from here would require a lot of
infrastructure, so we just call exit(0).)

Signed-off-by: Greg Farnum <greg@inktank.com>

diff --git a/src/include/ceph_features.h b/src/include/ceph_features.h
index 988a8574bcdcd44c3fad73d58412b6acdad341e1..80699697d80e00a3cad2bcb013489e2958a0b8af 100644 (file)
--- a/src/include/ceph_features.h
+++ b/src/include/ceph_features.h
@@ -44,7 +44,9 @@
 #define CEPH_FEATURE_EXPORT_PEER   (1ULL<<37)
 #define CEPH_FEATURE_OSD_ERASURE_CODES (1ULL<<38)
 #define CEPH_FEATURE_OSD_TMAP2OMAP (1ULL<<38)   /* overlap with EC */
-#define CEPH_FEATURE_OSDMAP_ENC    (1ULL<<39) /* supports new-style OSDMap encoding */
+/* The process supports new-style OSDMap encoding. Monitors also use
+   this bit to determine if peers support NAK messages. */
+#define CEPH_FEATURE_OSDMAP_ENC    (1ULL<<39)
 
 /*
  * The introduction of CEPH_FEATURE_OSD_SNAPMAPPER caused the feature

diff --git a/src/messages/MMonElection.h b/src/messages/MMonElection.h
index ea55bbdb45345479cedb635bb133dabc49e7d544..f1ef1dd792a45771e3ed442a16fb3069f7851b2e 100644 (file)
--- a/src/messages/MMonElection.h
+++ b/src/messages/MMonElection.h
@@ -45,11 +45,11 @@ public:
   bufferlist monmap_bl;
   set<int> quorum;
   uint64_t quorum_features;
+  bufferlist sharing_bl;
   /* the following were both used in the next branch for a while
    * on user cluster, so we've left them in for compatibility. */
   version_t defunct_one;
   version_t defunct_two;
-  bufferlist commands;
   
   MMonElection() : Message(MSG_MON_ELECTION, HEAD_VERSION, COMPAT_VERSION),
     op(0), epoch(0), quorum_features(0), defunct_one(0),
@@ -91,7 +91,7 @@ public:
     ::encode(quorum_features, payload);
     ::encode(defunct_one, payload);
     ::encode(defunct_two, payload);
-    ::encode(commands, payload);
+    ::encode(sharing_bl, payload);
   }
   void decode_payload() {
     bufferlist::iterator p = payload.begin();
@@ -112,7 +112,7 @@ public:
       ::decode(defunct_two, p);
     }
     if (header.version >= 5)
-      ::decode(commands, p);
+      ::decode(sharing_bl, p);
   }
   
 };

diff --git a/src/mon/Elector.cc b/src/mon/Elector.cc
index fdab343c8a5c66b44c0118f0ba9cefc31bdd1ccb..511d71411d96fa2f2157e18114a64d3b861684fb 100644 (file)
--- a/src/mon/Elector.cc
+++ b/src/mon/Elector.cc
@@ -111,7 +111,7 @@ void Elector::defer(int who)
   leader_acked = who;
   ack_stamp = ceph_clock_now(g_ceph_context);
   MMonElection *m = new MMonElection(MMonElection::OP_ACK, epoch, mon->monmap);
-  m->commands = mon->get_supported_commands_bl();
+  m->sharing_bl = mon->get_supported_commands_bl();
   mon->messenger->send_message(m, mon->monmap->get_inst(who));
   
   // set a timer
@@ -196,7 +196,7 @@ void Elector::victory()
     MMonElection *m = new MMonElection(MMonElection::OP_VICTORY, epoch, mon->monmap);
     m->quorum = quorum;
     m->quorum_features = features;
-    m->commands = *cmds_bl;
+    m->sharing_bl = *cmds_bl;
     mon->messenger->send_message(m, mon->monmap->get_inst(*p));
   }
     
@@ -213,8 +213,9 @@ void Elector::handle_propose(MMonElection *m)
   assert(m->epoch % 2 == 1); // election
   if ((required_features ^ m->get_connection()->get_features()) &
       required_features) {
-    dout(5) << " ignoring propose from mon without required features" << dendl;
-    m->put();
+    dout(5) << " ignoring propose from mon" << from
+           << " without required features" << dendl;
+    nak_old_peer(m);
     return;
   } else if (m->epoch > epoch) {
     bump_epoch(m->epoch);
@@ -278,7 +279,7 @@ void Elector::handle_ack(MMonElection *m)
   if (electing_me) {
     // thanks
     acked_me[from] = m->get_connection()->get_features();
-    if (!m->commands.length())
+    if (!m->sharing_bl.length())
       classic_mons.insert(from);
     dout(5) << " so far i have " << acked_me << dendl;
     
@@ -324,10 +325,10 @@ void Elector::handle_victory(MMonElection *m)
   cancel_timer();
 
   // stash leader's commands
-  if (m->commands.length()) {
+  if (m->sharing_bl.length()) {
     MonCommand *new_cmds;
     int cmdsize;
-    bufferlist::iterator bi = m->commands.begin();
+    bufferlist::iterator bi = m->sharing_bl.begin();
     MonCommand::decode_array(&new_cmds, &cmdsize, bi);
     mon->set_leader_supported_commands(new_cmds, cmdsize);
   } else { // they are a legacy monitor; use known legacy command set
@@ -340,8 +341,41 @@ void Elector::handle_victory(MMonElection *m)
   m->put();
 }
 
+void Elector::nak_old_peer(MMonElection *m)
+{
+  uint64_t supported_features = m->get_connection()->get_features();
+
+  if (supported_features & CEPH_FEATURE_OSDMAP_ENC) {
+    uint64_t required_features = mon->apply_compatset_features_to_quorum_requirements();
+    dout(10) << "sending nak to peer " << m->get_source()
+            << " that only supports " << supported_features
+            << " of the required " << required_features << dendl;
+    
+    MMonElection *reply = new MMonElection(MMonElection::OP_NAK, m->epoch,
+                                          mon->monmap);
+    reply->quorum_features = required_features;
+    mon->features.encode(reply->sharing_bl);
+    mon->messenger->send_message(reply, m->get_connection());
+  }
+  m->put();
+}
 
+void Elector::handle_nak(MMonElection *m)
+{
+  dout(1) << "handle_nak from " << m->get_source()
+         << " quorum_features " << m->quorum_features << dendl;
 
+  CompatSet other;
+  bufferlist::iterator bi = m->sharing_bl.begin();
+  other.decode(bi);
+  CompatSet diff = Monitor::get_supported_features().unsupported(other);
+  
+  derr << "Shutting down because I do not support required monitor features: { "
+       << diff << " }" << dendl;
+  
+  exit(0);
+  // the end!
+}
 
 void Elector::dispatch(Message *m)
 {
@@ -422,6 +456,9 @@ void Elector::dispatch(Message *m)
       case MMonElection::OP_VICTORY:
        handle_victory(em);
        return;
+      case MMonElection::OP_NAK:
+       handle_nak(em);
+       return;
       default:
        assert(0);
       }

diff --git a/src/mon/Elector.h b/src/mon/Elector.h
index 0b2b893ba34e192d4e18162ca762fc2a207e5508..b88e8304aacc61c300d7bd15da327cbafdadb5d3 100644 (file)
--- a/src/mon/Elector.h
+++ b/src/mon/Elector.h
@@ -245,7 +245,7 @@ class Elector {
    * @post  We sent a message of type OP_VICTORY to each quorum member.
    */
   void victory();
-  
+
   /**
    * Handle a message from some other node proposing himself to become him
    * the Leader.
@@ -317,6 +317,31 @@ class Elector {
    * @param m A message with an operation type of OP_VICTORY
    */
   void handle_victory(class MMonElection *m);
+  /**
+   * Send a nak to a peer who's out of date, containing information about why.
+   *
+   * If we get a message from a peer who can't support the required quorum
+   * features, we have to ignore them. This function will at least send
+   * them a message about *why* they're being ignored -- if they're new
+   * enough to support such a message.
+   *
+   * @param m A message from a monitor not supporting required features. We
+   * take ownership of the reference.
+   */
+  void nak_old_peer(class MMonElection *m);
+  /**
+   * Handle a message from some other participant declaring
+   * we cannot join the quorum.
+   *
+   * Apparently the quorum requires some feature that we do not implement. Shut
+   * down gracefully.
+   *
+   * @pre Election is on-going.
+   * @post We've shut down.
+   *
+   * @param m A message with an operation type of OP_NAK
+   */
+  void handle_nak(class MMonElection *m);
   
  public:
   /**