osd: compensate for replicas with tail > last_complete

Normally we shouldn't ever have a last_complete < log.tail (&& !backlog).
But maybe we do (old bugs, whatever; see #590).  In that case, the primary
can compensate by sending more log info to the replica.

Signed-off-by: Sage Weil <sage@newdream.net>

diff --git a/src/osd/PG.cc b/src/osd/PG.cc
index 0a61d128cb32a912b8ad2a611aec5908dac155b6..7175e13b7f175513fb5d94d918e3e816028d37a3 100644 (file)
--- a/src/osd/PG.cc
+++ b/src/osd/PG.cc
@@ -1855,13 +1855,24 @@ void PG::activate(ObjectStore::Transaction& t, list<Context*>& tfin,
       } 
       else {
        m = new MOSDPGLog(osd->osdmap->get_epoch(), info);
-       if (pi.last_update < log.tail) {
-         // summary/backlog
+       if (pi.log_tail > pi.last_complete && !pi.log_backlog) {
+         // the replica's tail is after it's last_complete and it has no backlog.
+         // ick, this shouldn't normally happen.  but we can compensate!
+         dout(10) << "activate peer osd" << peer << " has last_complete < log tail and no backlog, compensating" << dendl;
+         if (log.tail >= pi.last_complete) {
+           // _our_ log is sufficient, phew!
+           m->log.copy_after(log, pi.last_complete);
+         } else {
+           assert(log.backlog);
+           m->log = log;
+         }
+       } else if (log.tail > pi.last_update) {
+         // our tail is too new; send the full backlog.
          assert(log.backlog);
          m->log = log;
        } else {
-         // incremental log
-         assert(pi.last_update < info.last_update);
+         // send new stuff to append to replicas log
+         assert(info.last_update > pi.last_update);
          m->log.copy_after(log, pi.last_update);
        }
       }