PG: cleanup and fixes for async recovery

Signed-off-by: Neha Ojha <nojha@redhat.com>

diff --git a/src/osd/PG.cc b/src/osd/PG.cc
index 3aadda0640953d585959b8acabd83cf4fd093ae8..337063ddebbfd3749e158ea1c8417b69cb86ccb5 100644 (file)
--- a/src/osd/PG.cc
+++ b/src/osd/PG.cc
@@ -1478,14 +1478,13 @@ bool PG::recoverable_and_ge_min_size(const vector<int> &want) const
           pool.info.is_erasure() ? shard_id_t(i) : shard_id_t::NO_SHARD));
     }
   }
-
   // We go incomplete if below min_size for ec_pools since backfill
   // does not currently maintain rollbackability
   // Otherwise, we will go "peered", but not "active"
   if (num_want_acting < pool.info.min_size &&
       (pool.info.is_erasure() ||
        !cct->_conf->osd_allow_recovery_below_min_size)) {
-    dout(10) << __func__ << "failed, below min size" << dendl;
+    dout(10) << __func__ << " failed, below min size" << dendl;
     return false;
   }
 
@@ -1527,7 +1526,7 @@ void PG::choose_async_recovery_ec(const map<pg_shard_t, pg_info_t> &all_info,
     }
   }
 
-  dout(20) << __func__ << "candidates by cost are: " << candidates_by_cost
+  dout(20) << __func__ << " candidates by cost are: " << candidates_by_cost
            << dendl;
 
   // take out as many osds as we can for async recovery, in order of cost
@@ -1540,7 +1539,7 @@ void PG::choose_async_recovery_ec(const map<pg_shard_t, pg_info_t> &all_info,
       async_recovery->insert(cur_shard);
     }
   }
-  dout(20) << __func__ << "result want=" << *want
+  dout(20) << __func__ << " result want=" << *want
            << " async_recovery=" << *async_recovery << dendl;
 }
 
@@ -1568,7 +1567,7 @@ void PG::choose_async_recovery_replicated(const map<pg_shard_t, pg_info_t> &all_
     }
   }
 
-  dout(20) << __func__ << "candidates by cost are: " << candidates_by_cost
+  dout(20) << __func__ << " candidates by cost are: " << candidates_by_cost
            << dendl;
 
   // take out as many osds as we can for async recovery, in order of cost
@@ -1578,6 +1577,7 @@ void PG::choose_async_recovery_replicated(const map<pg_shard_t, pg_info_t> &all_
     for (auto it = candidate_want.begin(); it != candidate_want.end(); ++it) {
       if (*it == cur_shard.osd) {
         candidate_want.erase(it);
+       want->swap(candidate_want);
        async_recovery->insert(cur_shard);
         break;
       }
@@ -1586,7 +1586,7 @@ void PG::choose_async_recovery_replicated(const map<pg_shard_t, pg_info_t> &all_
       break;
     }
   }
-  dout(20) << __func__ << "result want=" << *want
+  dout(20) << __func__ << " result want=" << *want
            << " async_recovery=" << *async_recovery << dendl;
 }
 
@@ -1683,7 +1683,6 @@ bool PG::choose_acting(pg_shard_t &auth_log_shard_id,
       choose_async_recovery_replicated(all_info, auth_log_shard->second, &want, &want_async_recovery);
     }
   }
-
   if (want != acting) {
     dout(10) << __func__ << " want " << want << " != acting " << acting
             << ", requesting pg_temp change" << dendl;
@@ -1707,6 +1706,10 @@ bool PG::choose_acting(pg_shard_t &auth_log_shard_id,
     // Caller is GetInfo
     backfill_targets = want_backfill;
   }
+  assert(async_recovery_targets.empty() || async_recovery_targets == want_async_recovery);
+  if (async_recovery_targets.empty()) {
+    async_recovery_targets = want_async_recovery;
+  }
   // Will not change if already set because up would have had to change
   // Verify that nothing in backfill is in stray_set
   for (set<pg_shard_t>::iterator i = want_backfill.begin();
@@ -1714,8 +1717,9 @@ bool PG::choose_acting(pg_shard_t &auth_log_shard_id,
       ++i) {
     assert(stray_set.find(*i) == stray_set.end());
   }
-  dout(10) << __func__ << " want " << want << " (== acting) backfill_targets "
-          << want_backfill << dendl;
+  dout(10) << "choose_acting want=" << want << " backfill_targets="
+           << want_backfill << " async_recovery_targets="
+           << async_recovery_targets << dendl;
   return true;
 }
 
@@ -2733,6 +2737,7 @@ void PG::clear_recovery_state()
     finish_recovery_op(soid, true);
   }
 
+  async_recovery_targets.clear();
   backfill_targets.clear();
   backfill_info.clear();
   peer_backfill_info.clear();

diff --git a/src/osd/PG.h b/src/osd/PG.h
index ed4907daaa139826b4d6244220e8359b22ccae23..b07f2fc07092693da5787b7e905190169fdb0b42 100644 (file)
--- a/src/osd/PG.h
+++ b/src/osd/PG.h
@@ -1167,7 +1167,7 @@ protected:
   bool backfill_reserved;
   bool backfill_reserving;
 
-  set<pg_shard_t> backfill_targets;
+  set<pg_shard_t> backfill_targets,  async_recovery_targets;
 
   bool is_backfill_targets(pg_shard_t osd) {
     return backfill_targets.count(osd);

diff --git a/src/osd/PrimaryLogPG.cc b/src/osd/PrimaryLogPG.cc
index c9811373be4aae047deb5777fb688c811b0d9d5a..2fa81d49b65c8c3d62c1a6d4b155b6bbd8edfd0b 100644 (file)
--- a/src/osd/PrimaryLogPG.cc
+++ b/src/osd/PrimaryLogPG.cc
@@ -940,6 +940,14 @@ int PrimaryLogPG::do_command(
         f->dump_stream("shard") << *p;
       f->close_section();
     }
+    if (!async_recovery_targets.empty()) {
+      f->open_array_section("async_recovery_targets");
+      for (set<pg_shard_t>::iterator p = async_recovery_targets.begin();
+          p != async_recovery_targets.end();
+          ++p)
+        f->dump_stream("shard") << *p;
+      f->close_section();
+    }
     if (!acting_recovery_backfill.empty()) {
       f->open_array_section("acting_recovery_backfill");
       for (set<pg_shard_t>::iterator p = acting_recovery_backfill.begin();