osd: Fix the problem by rechecking backfill status each time

a RemoteReservationRevoked message is received, so we can
restart the backfill process in a more safe and less-likely raced way.

Fixes: http://tracker.ceph.com/issues/38034
Signed-off-by: xie xingguo <xie.xingguo@zte.com.cn>
(cherry picked from commit 9111ac7177c48fb0e4a0017e5734477a92198d2a)

diff --git a/src/osd/PG.cc b/src/osd/PG.cc
index b06852859efc29633916aa7e765872f48c5ca1ff..dcc545f82a212725fae2e4acf1c6670d57ef07d1 100644 (file)
--- a/src/osd/PG.cc
+++ b/src/osd/PG.cc
@@ -7152,7 +7152,12 @@ PG::RecoveryState::Backfilling::react(const RemoteReservationRevoked &)
   PG *pg = context< RecoveryMachine >().pg;
   pg->state_set(PG_STATE_BACKFILL_WAIT);
   cancel_backfill();
-  return transit<WaitLocalBackfillReserved>();
+  if (pg->needs_backfill()) {
+    return transit<WaitLocalBackfillReserved>();
+  } else {
+    // raced with MOSDPGBackfill::OP_BACKFILL_FINISH, ignore
+    return discard_event();
+  }
 }
 
 void PG::RecoveryState::Backfilling::exit()