osd: improve up_thru request behavior

There is some epoch the OSD wants for up_thru, based on when the PG mapping
last changed.  However, once the monitor gets to the point where it must
update the map, it should set up_thru to the most recent epoch the OSD has
seen (i.e. the epoch it is known to be "up thru"!).  This will hopefully/
frequently avoid any subsequent up_thru requests.

MOSDAlive already has a separate field (in PaxosServiceMessage) to hold the
latest epoch; just fix the constructor to set it properly, and make the
monitor use it.  No protocol change, yay!

Signed-off-by: Sage Weil <sage.weil@dreamhost.com>

diff --git a/src/messages/MOSDAlive.h b/src/messages/MOSDAlive.h
index 9dcae67b34750f3430c782680afd91a09ac35238..cac0a30ca12d50fc1d801e74be76ba1622186996 100644 (file)
--- a/src/messages/MOSDAlive.h
+++ b/src/messages/MOSDAlive.h
@@ -21,9 +21,9 @@
 
 class MOSDAlive : public PaxosServiceMessage {
  public:
-  epoch_t map_epoch;
+  epoch_t want;
 
-  MOSDAlive(epoch_t e) : PaxosServiceMessage(MSG_OSD_ALIVE, e), map_epoch(e) { }
+  MOSDAlive(epoch_t h, epoch_t w) : PaxosServiceMessage(MSG_OSD_ALIVE, h), want(w) { }
   MOSDAlive() : PaxosServiceMessage(MSG_OSD_ALIVE, 0) {}
 private:
   ~MOSDAlive() {}
@@ -31,17 +31,17 @@ private:
 public:
   void encode_payload() {
     paxos_encode();
-    ::encode(map_epoch, payload);
+    ::encode(want, payload);
   }
   void decode_payload() {
     bufferlist::iterator p = payload.begin();
     paxos_decode(p);
-    ::decode(map_epoch, p);
+    ::decode(want, p);
   }
 
   const char *get_type_name() { return "osd_alive"; }
   void print(ostream &out) {
-    out << "osd_alive(" << map_epoch << " v" << version << ")";
+    out << "osd_alive(want up_thru " << want << " have " << version << ")";
   }
   
 };

diff --git a/src/mon/OSDMonitor.cc b/src/mon/OSDMonitor.cc
index 11bffad36900e472879660627963e8b5dcba3bf3..44b8e9213d7730ae3470c85799fde88b20a5956a 100644 (file)
--- a/src/mon/OSDMonitor.cc
+++ b/src/mon/OSDMonitor.cc
@@ -666,14 +666,14 @@ bool OSDMonitor::preprocess_alive(MOSDAlive *m)
 
   if (osdmap.is_up(from) &&
       osdmap.get_inst(from) == m->get_orig_source_inst() &&
-      osdmap.get_up_thru(from) >= m->map_epoch) {
+      osdmap.get_up_thru(from) >= m->want) {
     // yup.
-    dout(7) << "preprocess_alive e" << m->map_epoch << " dup from " << m->get_orig_source_inst() << dendl;
-    _reply_map(m, m->map_epoch);
+    dout(7) << "preprocess_alive want up_thru " << m->want << " dup from " << m->get_orig_source_inst() << dendl;
+    _reply_map(m, m->version);
     return true;
   }
   
-  dout(10) << "preprocess_alive e" << m->map_epoch
+  dout(10) << "preprocess_alive want up_thru " << m->want
           << " from " << m->get_orig_source_inst() << dendl;
   return false;
 
@@ -690,9 +690,10 @@ bool OSDMonitor::prepare_alive(MOSDAlive *m)
     mon->clog.debug() << m->get_orig_source_inst() << " alive\n";
   }
 
-  dout(7) << "prepare_alive e" << m->map_epoch << " from " << m->get_orig_source_inst() << dendl;
-  pending_inc.new_up_thru[from] = m->map_epoch;
-  paxos->wait_for_commit(new C_ReplyMap(this, m, m->map_epoch));
+  dout(7) << "prepare_alive want up_thru " << m->want << " have " << m->version
+         << " from " << m->get_orig_source_inst() << dendl;
+  pending_inc.new_up_thru[from] = m->version;  // set to the latest map the OSD has
+  paxos->wait_for_commit(new C_ReplyMap(this, m, m->version));
   return true;
 }
 

diff --git a/src/osd/OSD.cc b/src/osd/OSD.cc
index 5b8555ab96658cbe19ab9b48fa8e19ed5fa7a25a..44df6581849fdb47192ad92bcd0197fb5d7b691c 100644 (file)
--- a/src/osd/OSD.cc
+++ b/src/osd/OSD.cc
@@ -1901,7 +1901,7 @@ void OSD::send_alive()
   if (up_thru_wanted > up_thru) {
     up_thru_pending = up_thru_wanted;
     dout(10) << "send_alive want " << up_thru_wanted << dendl;
-    monc->send_mon_message(new MOSDAlive(up_thru_wanted));
+    monc->send_mon_message(new MOSDAlive(osdmap->get_epoch(), up_thru_wanted));
   }
 }