From: David Zafman <dzafman@redhat.com>
Date: Wed, 12 Aug 2020 21:32:32 +0000 (-0700)
Subject: osd: Feature to add might_have_unfound info to list_unfound requests
X-Git-Tag: v17.0.0~1099^2~4
X-Git-Url: http://git.apps.os.sepia.ceph.com/?a=commitdiff_plain;h=a787148ed20273b396f4a0e47bf64bd8ebf01765;p=ceph.git

osd: Feature to add might_have_unfound info to list_unfound requests

fixes: https://tracker.ceph.com/issues/39012

Signed-off-by: David Zafman <dzafman@redhat.com>
---

diff --git a/src/osd/PeeringState.cc b/src/osd/PeeringState.cc
index 891405e1c63f9..af51893e41614 100644
--- a/src/osd/PeeringState.cc
+++ b/src/osd/PeeringState.cc
@@ -318,6 +318,36 @@ void PeeringState::purge_strays()
   peer_missing_requested.clear();
 }
 
+void PeeringState::query_unfound(Formatter *f, string state)
+{
+  psdout(20) << "Enter PeeringState common QueryUnfound" << dendl;
+  {
+    f->dump_string("state", state);
+    f->dump_bool("available_might_have_unfound", true);
+    f->open_array_section("might_have_unfound");
+    for (auto p = might_have_unfound.begin();
+	 p != might_have_unfound.end();
+	 ++p) {
+      if (peer_missing.count(*p)) {
+	; // Ignore already probed OSDs
+      } else {
+        f->open_object_section("osd");
+        f->dump_stream("osd") << *p;
+	if (peer_missing_requested.count(*p)) {
+	  f->dump_string("status", "querying");
+        } else if (!get_osdmap()->is_up(p->osd)) {
+	  f->dump_string("status", "osd is down");
+        } else {
+	  f->dump_string("status", "not queried");
+        }
+        f->close_section();
+      }
+    }
+    f->close_section();
+  }
+  psdout(20) << "Exit PeeringState common QueryUnfound" << dendl;
+  return;
+}
 
 bool PeeringState::proc_replica_info(
   pg_shard_t from, const pg_info_t &oinfo, epoch_t send_epoch)
@@ -4324,6 +4354,13 @@ boost::statechart::result PeeringState::Started::react(const QueryState& q)
   return discard_event();
 }
 
+boost::statechart::result PeeringState::Started::react(const QueryUnfound& q)
+{
+  q.f->dump_string("state", "Started");
+  q.f->dump_bool("available_might_have_unfound", false);
+  return discard_event();
+}
+
 void PeeringState::Started::exit()
 {
   context< PeeringMachine >().log_exit(state_name, enter_time);
@@ -4412,6 +4449,13 @@ boost::statechart::result PeeringState::Reset::react(const QueryState& q)
   return discard_event();
 }
 
+boost::statechart::result PeeringState::Reset::react(const QueryUnfound& q)
+{
+  q.f->dump_string("state", "Reset");
+  q.f->dump_bool("available_might_have_unfound", false);
+  return discard_event();
+}
+
 void PeeringState::Reset::exit()
 {
   context< PeeringMachine >().log_exit(state_name, enter_time);
@@ -4625,6 +4669,13 @@ boost::statechart::result PeeringState::Peering::react(const QueryState& q)
   return forward_event();
 }
 
+boost::statechart::result PeeringState::Peering::react(const QueryUnfound& q)
+{
+  q.f->dump_string("state", "Peering");
+  q.f->dump_bool("available_might_have_unfound", false);
+  return discard_event();
+}
+
 void PeeringState::Peering::exit()
 {
 
@@ -4905,6 +4956,14 @@ PeeringState::NotBackfilling::NotBackfilling(my_context ctx)
   pl->publish_stats_to_osd();
 }
 
+boost::statechart::result PeeringState::NotBackfilling::react(const QueryUnfound& q)
+{
+  DECLARE_LOCALS;
+
+  ps->query_unfound(q.f, "NotBackfilling");
+  return discard_event();
+}
+
 boost::statechart::result
 PeeringState::NotBackfilling::react(const RemoteBackfillReserved &evt)
 {
@@ -4938,6 +4997,14 @@ PeeringState::NotRecovering::NotRecovering(my_context ctx)
   pl->publish_stats_to_osd();
 }
 
+boost::statechart::result PeeringState::NotRecovering::react(const QueryUnfound& q)
+{
+  DECLARE_LOCALS;
+
+  ps->query_unfound(q.f, "NotRecovering");
+  return discard_event();
+}
+
 void PeeringState::NotRecovering::exit()
 {
   context< PeeringMachine >().log_exit(state_name, enter_time);
@@ -5802,6 +5869,14 @@ boost::statechart::result PeeringState::Active::react(const QueryState& q)
   return forward_event();
 }
 
+boost::statechart::result PeeringState::Active::react(const QueryUnfound& q)
+{
+  DECLARE_LOCALS;
+
+  ps->query_unfound(q.f, "Active");
+  return discard_event();
+}
+
 boost::statechart::result PeeringState::Active::react(
   const ActivateCommitted &evt)
 {
@@ -6101,6 +6176,13 @@ boost::statechart::result PeeringState::ReplicaActive::react(const QueryState& q
   return forward_event();
 }
 
+boost::statechart::result PeeringState::ReplicaActive::react(const QueryUnfound& q)
+{
+  q.f->dump_string("state", "ReplicaActive");
+  q.f->dump_bool("available_might_have_unfound", false);
+  return discard_event();
+}
+
 void PeeringState::ReplicaActive::exit()
 {
   context< PeeringMachine >().log_exit(state_name, enter_time);
@@ -6469,6 +6551,13 @@ boost::statechart::result PeeringState::GetInfo::react(const QueryState& q)
   return forward_event();
 }
 
+boost::statechart::result PeeringState::GetInfo::react(const QueryUnfound& q)
+{
+  q.f->dump_string("state", "GetInfo");
+  q.f->dump_bool("available_might_have_unfound", false);
+  return discard_event();
+}
+
 void PeeringState::GetInfo::exit()
 {
   context< PeeringMachine >().log_exit(state_name, enter_time);
@@ -6603,6 +6692,13 @@ boost::statechart::result PeeringState::GetLog::react(const QueryState& q)
   return forward_event();
 }
 
+boost::statechart::result PeeringState::GetLog::react(const QueryUnfound& q)
+{
+  q.f->dump_string("state", "GetLog");
+  q.f->dump_bool("available_might_have_unfound", false);
+  return discard_event();
+}
+
 void PeeringState::GetLog::exit()
 {
   context< PeeringMachine >().log_exit(state_name, enter_time);
@@ -6665,6 +6761,13 @@ boost::statechart::result PeeringState::WaitActingChange::react(const QueryState
   return forward_event();
 }
 
+boost::statechart::result PeeringState::WaitActingChange::react(const QueryUnfound& q)
+{
+  q.f->dump_string("state", "WaitActingChange");
+  q.f->dump_bool("available_might_have_unfound", false);
+  return discard_event();
+}
+
 void PeeringState::WaitActingChange::exit()
 {
   context< PeeringMachine >().log_exit(state_name, enter_time);
@@ -6714,6 +6817,13 @@ boost::statechart::result PeeringState::Down::react(const QueryState& q)
   return forward_event();
 }
 
+boost::statechart::result PeeringState::Down::react(const QueryUnfound& q)
+{
+  q.f->dump_string("state", "Down");
+  q.f->dump_bool("available_might_have_unfound", false);
+  return discard_event();
+}
+
 boost::statechart::result PeeringState::Down::react(const MNotifyRec& infoevt)
 {
   DECLARE_LOCALS;
@@ -6791,6 +6901,13 @@ boost::statechart::result PeeringState::Incomplete::react(
   return forward_event();
 }
 
+boost::statechart::result PeeringState::Incomplete::react(const QueryUnfound& q)
+{
+  q.f->dump_string("state", "Incomplete");
+  q.f->dump_bool("available_might_have_unfound", false);
+  return discard_event();
+}
+
 void PeeringState::Incomplete::exit()
 {
   context< PeeringMachine >().log_exit(state_name, enter_time);
@@ -6943,6 +7060,13 @@ boost::statechart::result PeeringState::GetMissing::react(const QueryState& q)
   return forward_event();
 }
 
+boost::statechart::result PeeringState::GetMissing::react(const QueryUnfound& q)
+{
+  q.f->dump_string("state", "GetMising");
+  q.f->dump_bool("available_might_have_unfound", false);
+  return discard_event();
+}
+
 void PeeringState::GetMissing::exit()
 {
   context< PeeringMachine >().log_exit(state_name, enter_time);
@@ -6989,6 +7113,13 @@ boost::statechart::result PeeringState::WaitUpThru::react(const QueryState& q)
   return forward_event();
 }
 
+boost::statechart::result PeeringState::WaitUpThru::react(const QueryUnfound& q)
+{
+  q.f->dump_string("state", "WaitUpThru");
+  q.f->dump_bool("available_might_have_unfound", false);
+  return discard_event();
+}
+
 void PeeringState::WaitUpThru::exit()
 {
   context< PeeringMachine >().log_exit(state_name, enter_time);
diff --git a/src/osd/PeeringState.h b/src/osd/PeeringState.h
index 7583a583b5fe1..e0fe398833660 100644
--- a/src/osd/PeeringState.h
+++ b/src/osd/PeeringState.h
@@ -438,6 +438,14 @@ public:
     }
   };
 
+  struct QueryUnfound : boost::statechart::event< QueryUnfound > {
+    ceph::Formatter *f;
+    explicit QueryUnfound(ceph::Formatter *f) : f(f) {}
+    void print(std::ostream *out) const {
+      *out << "QueryUnfound";
+    }
+  };
+
   struct AdvMap : boost::statechart::event< AdvMap > {
     OSDMapRef osdmap;
     OSDMapRef lastmap;
@@ -680,6 +688,7 @@ public:
 
     typedef boost::mpl::list <
       boost::statechart::custom_reaction< QueryState >,
+      boost::statechart::custom_reaction< QueryUnfound >,
       boost::statechart::custom_reaction< AdvMap >,
       boost::statechart::custom_reaction< ActMap >,
       boost::statechart::custom_reaction< NullEvt >,
@@ -687,6 +696,7 @@ public:
       boost::statechart::transition< boost::statechart::event_base, Crashed >
       > reactions;
     boost::statechart::result react(const QueryState& q);
+    boost::statechart::result react(const QueryUnfound& q);
     boost::statechart::result react(const AdvMap&);
     boost::statechart::result react(const ActMap&);
     boost::statechart::result react(const IntervalFlush&);
@@ -703,6 +713,7 @@ public:
 
     typedef boost::mpl::list <
       boost::statechart::custom_reaction< QueryState >,
+      boost::statechart::custom_reaction< QueryUnfound >,
       boost::statechart::custom_reaction< AdvMap >,
       boost::statechart::custom_reaction< IntervalFlush >,
       // ignored
@@ -717,6 +728,7 @@ public:
       boost::statechart::transition< boost::statechart::event_base, Crashed >
       > reactions;
     boost::statechart::result react(const QueryState& q);
+    boost::statechart::result react(const QueryUnfound& q);
     boost::statechart::result react(const AdvMap&);
     boost::statechart::result react(const IntervalFlush&);
     boost::statechart::result react(const boost::statechart::event_base&) {
@@ -768,6 +780,7 @@ public:
 		      NamedState {
     typedef boost::mpl::list <
       boost::statechart::custom_reaction< QueryState >,
+      boost::statechart::custom_reaction< QueryUnfound >,
       boost::statechart::custom_reaction< AdvMap >,
       boost::statechart::custom_reaction< MLogRec >,
       boost::statechart::custom_reaction< MInfoRec >,
@@ -775,6 +788,7 @@ public:
       > reactions;
     explicit WaitActingChange(my_context ctx);
     boost::statechart::result react(const QueryState& q);
+    boost::statechart::result react(const QueryUnfound& q);
     boost::statechart::result react(const AdvMap&);
     boost::statechart::result react(const MLogRec&);
     boost::statechart::result react(const MInfoRec&);
@@ -794,10 +808,12 @@ public:
 
     typedef boost::mpl::list <
       boost::statechart::custom_reaction< QueryState >,
+      boost::statechart::custom_reaction< QueryUnfound >,
       boost::statechart::transition< Activate, Active >,
       boost::statechart::custom_reaction< AdvMap >
       > reactions;
     boost::statechart::result react(const QueryState& q);
+    boost::statechart::result react(const QueryUnfound& q);
     boost::statechart::result react(const AdvMap &advmap);
   };
 
@@ -813,6 +829,7 @@ public:
 
     typedef boost::mpl::list <
       boost::statechart::custom_reaction< QueryState >,
+      boost::statechart::custom_reaction< QueryUnfound >,
       boost::statechart::custom_reaction< ActMap >,
       boost::statechart::custom_reaction< AdvMap >,
       boost::statechart::custom_reaction< MInfoRec >,
@@ -834,6 +851,7 @@ public:
       boost::statechart::custom_reaction< CheckReadable>
       > reactions;
     boost::statechart::result react(const QueryState& q);
+    boost::statechart::result react(const QueryUnfound& q);
     boost::statechart::result react(const ActMap&);
     boost::statechart::result react(const AdvMap&);
     boost::statechart::result react(const MInfoRec& infoevt);
@@ -955,23 +973,27 @@ public:
 
   struct NotBackfilling : boost::statechart::state< NotBackfilling, Active>, NamedState {
     typedef boost::mpl::list<
+      boost::statechart::custom_reaction< QueryUnfound >,
       boost::statechart::transition< RequestBackfill, WaitLocalBackfillReserved>,
       boost::statechart::custom_reaction< RemoteBackfillReserved >,
       boost::statechart::custom_reaction< RemoteReservationRejectedTooFull >
       > reactions;
     explicit NotBackfilling(my_context ctx);
     void exit();
+    boost::statechart::result react(const QueryUnfound& q);
     boost::statechart::result react(const RemoteBackfillReserved& evt);
     boost::statechart::result react(const RemoteReservationRejectedTooFull& evt);
   };
 
   struct NotRecovering : boost::statechart::state< NotRecovering, Active>, NamedState {
     typedef boost::mpl::list<
+      boost::statechart::custom_reaction< QueryUnfound >,
       boost::statechart::transition< DoRecovery, WaitLocalRecoveryReserved >,
       boost::statechart::custom_reaction< DeferRecovery >,
       boost::statechart::custom_reaction< UnfoundRecovery >
       > reactions;
     explicit NotRecovering(my_context ctx);
+    boost::statechart::result react(const QueryUnfound& q);
     boost::statechart::result react(const DeferRecovery& evt) {
       /* no-op */
       return discard_event();
@@ -991,6 +1013,7 @@ public:
 
     typedef boost::mpl::list <
       boost::statechart::custom_reaction< QueryState >,
+      boost::statechart::custom_reaction< QueryUnfound >,
       boost::statechart::custom_reaction< ActMap >,
       boost::statechart::custom_reaction< MQuery >,
       boost::statechart::custom_reaction< MInfoRec >,
@@ -1009,6 +1032,7 @@ public:
       boost::statechart::custom_reaction< MLease >
       > reactions;
     boost::statechart::result react(const QueryState& q);
+    boost::statechart::result react(const QueryUnfound& q);
     boost::statechart::result react(const MInfoRec& infoevt);
     boost::statechart::result react(const MLogRec& logevt);
     boost::statechart::result react(const MTrim& trimevt);
@@ -1239,11 +1263,13 @@ public:
 
     typedef boost::mpl::list <
       boost::statechart::custom_reaction< QueryState >,
+      boost::statechart::custom_reaction< QueryUnfound >,
       boost::statechart::transition< GotInfo, GetLog >,
       boost::statechart::custom_reaction< MNotifyRec >,
       boost::statechart::transition< IsDown, Down >
       > reactions;
     boost::statechart::result react(const QueryState& q);
+    boost::statechart::result react(const QueryUnfound& q);
     boost::statechart::result react(const MNotifyRec& infoevt);
   };
 
@@ -1260,6 +1286,7 @@ public:
 
     typedef boost::mpl::list <
       boost::statechart::custom_reaction< QueryState >,
+      boost::statechart::custom_reaction< QueryUnfound >,
       boost::statechart::custom_reaction< MLogRec >,
       boost::statechart::custom_reaction< GotLog >,
       boost::statechart::custom_reaction< AdvMap >,
@@ -1268,6 +1295,7 @@ public:
       > reactions;
     boost::statechart::result react(const AdvMap&);
     boost::statechart::result react(const QueryState& q);
+    boost::statechart::result react(const QueryUnfound& q);
     boost::statechart::result react(const MLogRec& logevt);
     boost::statechart::result react(const GotLog&);
   };
@@ -1282,10 +1310,12 @@ public:
 
     typedef boost::mpl::list <
       boost::statechart::custom_reaction< QueryState >,
+      boost::statechart::custom_reaction< QueryUnfound >,
       boost::statechart::custom_reaction< MLogRec >,
       boost::statechart::transition< NeedUpThru, WaitUpThru >
       > reactions;
     boost::statechart::result react(const QueryState& q);
+    boost::statechart::result react(const QueryUnfound& q);
     boost::statechart::result react(const MLogRec& logevt);
   };
 
@@ -1295,10 +1325,12 @@ public:
 
     typedef boost::mpl::list <
       boost::statechart::custom_reaction< QueryState >,
+      boost::statechart::custom_reaction< QueryUnfound >,
       boost::statechart::custom_reaction< ActMap >,
       boost::statechart::custom_reaction< MLogRec >
       > reactions;
     boost::statechart::result react(const QueryState& q);
+    boost::statechart::result react(const QueryUnfound& q);
     boost::statechart::result react(const ActMap& am);
     boost::statechart::result react(const MLogRec& logrec);
   };
@@ -1307,9 +1339,11 @@ public:
     explicit Down(my_context ctx);
     typedef boost::mpl::list <
       boost::statechart::custom_reaction< QueryState >,
+      boost::statechart::custom_reaction< QueryUnfound >,
       boost::statechart::custom_reaction< MNotifyRec >
       > reactions;
     boost::statechart::result react(const QueryState& q);
+    boost::statechart::result react(const QueryUnfound& q);
     boost::statechart::result react(const MNotifyRec& infoevt);
     void exit();
   };
@@ -1318,11 +1352,13 @@ public:
     typedef boost::mpl::list <
       boost::statechart::custom_reaction< AdvMap >,
       boost::statechart::custom_reaction< MNotifyRec >,
+      boost::statechart::custom_reaction< QueryUnfound >,
       boost::statechart::custom_reaction< QueryState >
       > reactions;
     explicit Incomplete(my_context ctx);
     boost::statechart::result react(const AdvMap &advmap);
     boost::statechart::result react(const MNotifyRec& infoevt);
+    boost::statechart::result react(const QueryUnfound& q);
     boost::statechart::result react(const QueryState& q);
     void exit();
   };
@@ -1478,6 +1514,7 @@ public:
   }
 
   void update_heartbeat_peers();
+  void query_unfound(Formatter *f, string state);
   bool proc_replica_info(
     pg_shard_t from, const pg_info_t &oinfo, epoch_t send_epoch);
   void remove_down_peer_info(const OSDMapRef &osdmap);
diff --git a/src/osd/PrimaryLogPG.cc b/src/osd/PrimaryLogPG.cc
index fbcb64c72e84f..3b7f6338e2edb 100644
--- a/src/osd/PrimaryLogPG.cc
+++ b/src/osd/PrimaryLogPG.cc
@@ -1119,6 +1119,9 @@ void PrimaryLogPG::do_command(
       }
       f->close_section();
     }
+    // Get possible locations of missing objects from pg information
+    PeeringState::QueryUnfound q(f.get());
+    recovery_state.handle_event(q, 0);
     f->dump_bool("more", p != needs_recovery_map.end());
     f->close_section();
   }