osd: improve build_prior logic

If, during some interval since the pg last went active, we may have gone
rw, but none of the osds survived, then we include all of those osds
in the prior_set (even tho they're down), because they may have written data
that we want.

The prior logic appears to have been broken.  It was only looking at the
primary osd.

diff --git a/src/osd/PG.cc b/src/osd/PG.cc
index f05309df4330f738a1e853dc4feeea2499889bc6..86bfc6f0b862c182637e991512b156c2d40f6e50 100644 (file)
--- a/src/osd/PG.cc
+++ b/src/osd/PG.cc
@@ -693,45 +693,53 @@ void PG::build_prior()
             << dendl;
     
     int num_still_up_or_clean = 0;
+    bool any_survived = false;
     for (unsigned i=0; i<acting.size(); i++) {
       if (osd->osdmap->is_up(acting[i])) {  // is up now
-       if (first_epoch <= info.history.last_epoch_started &&  // FIXME this is sloppy i think?
+
+       // FIXME this is sloppy i think?
+       if (first_epoch <= info.history.last_epoch_started && 
            last_epoch >= info.history.last_epoch_started)
          any_up = true;
 
+       any_survived = true;
+
        if (acting[i] != osd->whoami)       // and is not me
          prior_set.insert(acting[i]);
-
+       
        // has it been up this whole time?
        if (osd->osdmap->get_up_from(acting[i]) <= first_epoch)
          num_still_up_or_clean++;
       } else {
-       dout(10) << "build_prior  prior osd" << acting[i] << " is down, must notify mon" << dendl;
+       dout(10) << "build_prior  prior osd" << acting[i]
+                << " is down, must notify mon" << dendl;
        must_notify_mon = true;
-
-       // include osd in set anyway
        prior_set_down.insert(acting[i]);
+      }
+    }
 
-       // fixme: how do we identify a "clean" shutdown anyway?
-
-       if (i == 0) {
-         if (maybe_went_rw) {
-           dout(10) << "build_prior  pg possibly went active+rw in epoch " 
-                    << (lastmap->get_up_thru(acting[i]) + 1) << dendl;
-           some_down = true;
-           prior_set.insert(acting[i]);
-
-           // take note that we care about this osd's up_thru.  if it
-           // changes later, it will affect our prior_set, and we'll want
-           // to rebuild the prior set!
-           prior_set_up_thru[acting[0]] = lastmap->get_up_thru(acting[0]);
-         }
+    // if nobody survived this interval, and we may have gone rw,
+    // then we need to wait for one of those osds to recover to
+    // ensure that we haven't lost any information.
+    if (!any_survived && maybe_went_rw) {
+      // fixme: how do we identify a "clean" shutdown anyway?
+      dout(10) << "build_prior  possibly went active+rw, no survivors, including" << dendl;
+      for (unsigned i=0; i<acting.size(); i++)
+       if (osd->osdmap->is_down(acting[i])) {
+         prior_set.insert(acting[i]);
+         prior_set_down.erase(acting[i]);
        }
-      }
+      some_down = true;
+      
+      // take note that we care about the primary's up_thru.  if it
+      // changes later, it will affect our prior_set, and we'll want
+      // to rebuild it!
+      prior_set_up_thru[acting[0]] = lastmap->get_up_thru(acting[0]);
     }
 
     if (num_still_up_or_clean == 0) {
-      dout(10) << "build_prior  none of " << acting << " still up or cleanly shutdown, pg crashed" << dendl;
+      dout(10) << "build_prior  none of " << acting 
+              << " still up or cleanly shutdown, pg crashed" << dendl;
       state_set(PG_STATE_CRASHED);
     }
   }