Merge pull request ceph#59615 from xxhdx1985126/wip-63844-67916

crimson/osd/recovery_backend: always set the primary osd's BackfillInterval::version to the last_update before the backfill scan

Reviewed-by: Samuel Just <[email protected]>

Author: Matan-B

src/crimson/osd/backfill_facades.h

@@ -28,8 +28,8 @@ struct PeeringFacade final : BackfillState::PeeringFacade {
     return peering_state.get_peer_info(peer).last_backfill;
   }
 
-  const eversion_t& get_last_update() const override {
-    return peering_state.get_info().last_update;
+  eversion_t get_pg_committed_to() const override {
+    return peering_state.get_pg_committed_to();
   }
 
   const eversion_t& get_log_tail() const override {

src/crimson/osd/backfill_state.cc

@@ -406,7 +406,7 @@ BackfillState::Enqueuing::Enqueuing(my_context ctx)
 BackfillState::PrimaryScanning::PrimaryScanning(my_context ctx)
   : my_base(ctx)
 {
-  backfill_state().backfill_info.version = peering_state().get_last_update();
+  backfill_state().backfill_info.version = peering_state().get_pg_committed_to();
   backfill_listener().request_primary_scan(
     backfill_state().backfill_info.begin);
 }
@@ -416,6 +416,7 @@ BackfillState::PrimaryScanning::react(PrimaryScanned evt)
 {
   LOG_PREFIX(BackfillState::PrimaryScanning::react::PrimaryScanned);
   DEBUGDPP("", pg());
+  evt.result.version = backfill_state().backfill_info.version;
   backfill_state().backfill_info = std::move(evt.result);
   if (!backfill_state().is_suspended()) {
     return transit<Enqueuing>();

src/crimson/osd/backfill_state.h

@@ -387,7 +387,7 @@ struct BackfillState::PeeringFacade {
   virtual const std::set<pg_shard_t>& get_backfill_targets() const = 0;
   virtual const hobject_t& get_peer_last_backfill(pg_shard_t peer) const = 0;
   virtual const PGLog& get_pg_log() const = 0;
-  virtual const eversion_t& get_last_update() const = 0;
+  virtual eversion_t get_pg_committed_to() const = 0;
   virtual const eversion_t& get_log_tail() const = 0;
 
   // the performance impact of `std::function` has not been considered yet.

src/crimson/osd/pg_recovery.cc

@@ -531,10 +531,10 @@ void PGRecovery::enqueue_push(
 {
   logger().info("{}: obj={} v={} peers={}", __func__, obj, v, peers);
   auto &peering_state = pg->get_peering_state();
-  peering_state.prepare_backfill_for_missing(obj, v, peers);
   auto [recovering, added] = pg->get_recovery_backend()->add_recovering(obj);
   if (!added)
     return;
+  peering_state.prepare_backfill_for_missing(obj, v, peers);
   std::ignore = pg->get_recovery_backend()->recover_object(obj, v).\
   handle_exception_interruptible([] (auto) {
     ceph_abort_msg("got exception on backfill's push");

src/crimson/osd/recovery_backend.cc

@@ -265,11 +265,11 @@ RecoveryBackend::scan_for_backfill(
 	  }, PGBackend::load_metadata_ertr::assert_all{});
 	}
       });
-    }).then_interruptible([FNAME, this, version_map, start=std::move(start), next=std::move(next)] {
+    }).then_interruptible([FNAME, version_map, start=std::move(start),
+			  next=std::move(next), this] {
       BackfillInterval bi;
       bi.begin = std::move(start);
       bi.end = std::move(next);
-      bi.version = pg.get_info().last_update;
       bi.objects = std::move(*version_map);
       DEBUGDPP("{} BackfillInterval filled, leaving, {}",
 	       "scan_for_backfill",

src/test/crimson/test_backfill.cc

@@ -233,7 +233,7 @@ struct BackfillFixture::PeeringFacade
   const hobject_t& get_peer_last_backfill(pg_shard_t peer) const override {
     return backfill_targets.at(peer).last_backfill;
   }
-  const eversion_t& get_last_update() const override {
+  eversion_t get_pg_committed_to() const override {
     return backfill_source.last_update;
   }
   const eversion_t& get_log_tail() const override {