From dc1db5518b47ae0d3cc4657716a97f0a0f37972f Mon Sep 17 00:00:00 2001
From: sageweil <sageweil@29311d96-e01e-0410-9327-a35deaab8ce9>
Date: Mon, 4 Jun 2007 21:09:42 +0000
Subject: [PATCH] * mds boot osdmap weirdness fixup

git-svn-id: https://ceph.svn.sf.net/svnroot/ceph@1392 29311d96-e01e-0410-9327-a35deaab8ce9
---
 branches/sage/cephmds2/TODO       |  3 +--
 branches/sage/cephmds2/mds/MDS.cc | 43 ++++++++++++++++++++-----------
 branches/sage/cephmds2/mds/MDS.h  |  1 +
 3 files changed, 30 insertions(+), 17 deletions(-)

diff --git a/branches/sage/cephmds2/TODO b/branches/sage/cephmds2/TODO
index 72bb37d21098d..9150c21fcfdb7 100644
--- a/branches/sage/cephmds2/TODO
+++ b/branches/sage/cephmds2/TODO
@@ -20,10 +20,9 @@ mds
 - more testing of failures + thrashing.
   - is export prep dir open deadlock properly fixed by forge_replica_dir()?
 
-- locker vs node failure
 - osd needs a set_floor_and_read op for safe failover/STOGITH-like semantics.
 - failures during recovery stages (resolve, rejoin)... make sure rejoin still works!
-- fix mds initial osdmap weirdness (which will currently screw up on standby -> almost anything)
+
 - incremental mdsmaps?
 - client failure
 
diff --git a/branches/sage/cephmds2/mds/MDS.cc b/branches/sage/cephmds2/mds/MDS.cc
index fc2fd5ae49b7f..7a1bf6d69d88e 100644
--- a/branches/sage/cephmds2/mds/MDS.cc
+++ b/branches/sage/cephmds2/mds/MDS.cc
@@ -440,6 +440,7 @@ void MDS::beacon_kill(utime_t lab)
 
 void MDS::handle_mds_map(MMDSMap *m)
 {
+  version_t hadepoch = mdsmap->get_epoch();
   version_t epoch = m->get_epoch();
   dout(5) << "handle_mds_map epoch " << epoch << " from " << m->get_source() << endl;
 
@@ -670,6 +671,12 @@ void MDS::handle_mds_map(MMDSMap *m)
   }
   */
 
+  // just got mdsmap+osdmap?
+  if (hadepoch == 0 && 
+      mdsmap->get_epoch() > 0 &&
+      osdmap->get_epoch() > 0)
+    boot();
+
   delete m;
 }
 
@@ -690,24 +697,30 @@ void MDS::bcast_mds_map()
 
 void MDS::handle_osd_map(MOSDMap *m)
 {
-  version_t had = osdmap->get_epoch();
+  version_t hadepoch = osdmap->get_epoch();
+  dout(10) << "handle_osd_map had " << hadepoch << endl;
   
-  dout(10) << "handle_osd_map had " << had << endl;
-
-  // process locally
+  // process
   objecter->handle_osd_map(m);
 
-  if (had == 0 && osdmap->get_epoch() > 0) {
-    if (is_creating()) 
-      boot_create();    // new tables, journal
-    else if (is_starting())
-      boot_start();     // old tables, empty journal
-    else if (is_replay()) 
-      boot_replay();    // replay, join
-    else 
-      assert(is_standby());
-  }  
-  
+  // just got mdsmap+osdmap?
+  if (hadepoch == 0 && 
+      osdmap->get_epoch() > 0 &&
+      mdsmap->get_epoch() > 0) 
+    boot();
+}
+
+
+void MDS::boot()
+{   
+  if (is_creating()) 
+    boot_create();    // new tables, journal
+  else if (is_starting())
+    boot_start();     // old tables, empty journal
+  else if (is_replay()) 
+    boot_replay();    // replay, join
+  else 
+    assert(0);
 }
 
 
diff --git a/branches/sage/cephmds2/mds/MDS.h b/branches/sage/cephmds2/mds/MDS.h
index d7645c322a518..a030e53b62768 100644
--- a/branches/sage/cephmds2/mds/MDS.h
+++ b/branches/sage/cephmds2/mds/MDS.h
@@ -188,6 +188,7 @@ class MDS : public Dispatcher {
   int init(bool standby=false);
   void reopen_logger();
 
+  void boot();
   void boot_create();             // i am new mds.
   void boot_start();              // i am old but empty (was down:out) mds.
   void boot_replay(int step=0);   // i am recovering existing (down:failed) mds.
-- 
2.39.5