osd/ReplicatedPG: do not evict blocked objects

If the object is blocked, that means we are either still promoting it, or
we are in the process of doing a blocking flush.  In both cases, do not
evict it just yet.  For the promotion case in particular this can cause a
very long op delay because we'll have to restart the promotion, and this
can potentially loop indefinitely.

For the flush case, flushes are generally only blocking when the user
explicitly requests it.  In that case, we are not particularly concerned
that we may delay the objects eventual eviction as the user is probably
controlling that anyway (and even if they aren't, the agent will still
try this object again later).

Fixes: #9285
Reported-by: Wang, Zhiqiang <zhiqiang.wang@intel.com>
Signed-off-by: Sage Weil <sage@redhat.com>

diff --git a/src/osd/ReplicatedPG.cc b/src/osd/ReplicatedPG.cc
index 27ee3f3dbdf3468041aa0b15c9ba92b4b97dd34a..937108236ed11d68d32c4b1e62c1e02c928fb174 100644 (file)
--- a/src/osd/ReplicatedPG.cc
+++ b/src/osd/ReplicatedPG.cc
@@ -11581,6 +11581,10 @@ bool ReplicatedPG::agent_maybe_evict(ObjectContextRef& obc)
     dout(20) << __func__ << " skip (watchers) " << obc->obs.oi << dendl;
     return false;
   }
+  if (obc->is_blocked()) {
+    dout(20) << __func__ << " skip (blocked) " << obc->obs.oi << dendl;
+    return false;
+  }
 
   if (soid.snap == CEPH_NOSNAP) {
     int result = _verify_no_head_clones(soid, obc->ssc->snapset);