crimson/osd: decouple backfill from PG::start_recovery_ops().

Signed-off-by: Radoslaw Zarzynski <rzarzyns@redhat.com>

diff --git a/src/crimson/osd/pg.h b/src/crimson/osd/pg.h
index 1768c0a52eb85a090748344c1f63ae36e4a0b249..ccf874fae306e9e3d844674eab11efe8589d46b4 100644 (file)
--- a/src/crimson/osd/pg.h
+++ b/src/crimson/osd/pg.h
@@ -344,8 +344,7 @@ public:
   }
 
   void on_backfill_reserved() final {
-    recovery_handler->start_background_recovery(
-      crimson::osd::scheduler::scheduler_class_t::background_best_effort);
+    ceph_assert(0 == "Not implemented");
   }
   void on_backfill_canceled() final {
     ceph_assert(0 == "Not implemented");

diff --git a/src/crimson/osd/pg_recovery.cc b/src/crimson/osd/pg_recovery.cc
index f954d4553ecff7821585121e06122cd89a89a6fc..545b2a87be3cd40fb708f50641eefa39bf74a8eb 100644 (file)
--- a/src/crimson/osd/pg_recovery.cc
+++ b/src/crimson/osd/pg_recovery.cc
@@ -20,6 +20,12 @@
 #include "osd/osd_types.h"
 #include "osd/PeeringState.h"
 
+namespace {
+  seastar::logger& logger() {
+    return crimson::get_logger(ceph_subsys_osd);
+  }
+}
+
 void PGRecovery::start_background_recovery(
   crimson::osd::scheduler::scheduler_class_t klass)
 {
@@ -36,21 +42,22 @@ PGRecovery::start_recovery_ops(size_t max_to_start)
 {
   assert(pg->is_primary());
   assert(pg->is_peered());
+  assert(pg->is_recovering());
+  // in ceph-osd the do_recovery() path handles both the pg log-based
+  // recovery and the backfill, albeit they are separated at the layer
+  // of PeeringState. In crimson-osd backfill has been cut from it, so
+  // and do_recovery() is actually solely for pg log-based recovery.
+  // At the time of writing it's considered to move it to FSM and fix
+  // the naming as well.
+  assert(!pg->is_backfilling());
   assert(!pg->get_peering_state().is_deleting());
 
-  if (!pg->is_recovering() && !pg->is_backfilling()) {
-    return crimson::osd::make_ready_blocking_future<bool>(false);
-  }
-
   std::vector<crimson::osd::blocking_future<>> started;
   started.reserve(max_to_start);
   max_to_start -= start_primary_recovery_ops(max_to_start, &started);
   if (max_to_start > 0) {
     max_to_start -= start_replica_recovery_ops(max_to_start, &started);
   }
-  if (max_to_start > 0) {
-    max_to_start -= start_backfill_ops(max_to_start, &started);
-  }
   return crimson::osd::join_blocking_futures(std::move(started)).then(
     [this] {
     bool done = !pg->get_peering_state().needs_recovery();
@@ -58,14 +65,29 @@ PGRecovery::start_recovery_ops(size_t max_to_start)
       crimson::get_logger(ceph_subsys_osd).debug("start_recovery_ops: AllReplicasRecovered for pg: {}",
                     pg->get_pgid());
       using LocalPeeringEvent = crimson::osd::LocalPeeringEvent;
-      (void) pg->get_shard_services().start_operation<LocalPeeringEvent>(
-       static_cast<crimson::osd::PG*>(pg),
-       pg->get_shard_services(),
-       pg->get_pg_whoami(),
-       pg->get_pgid(),
-       pg->get_osdmap_epoch(),
-       pg->get_osdmap_epoch(),
-       PeeringState::AllReplicasRecovered{});
+      if (!pg->get_peering_state().needs_backfill()) {
+        logger().debug("start_recovery_ops: AllReplicasRecovered for pg: {}",
+                      pg->get_pgid());
+        (void) pg->get_shard_services().start_operation<LocalPeeringEvent>(
+          static_cast<crimson::osd::PG*>(pg),
+          pg->get_shard_services(),
+          pg->get_pg_whoami(),
+          pg->get_pgid(),
+          pg->get_osdmap_epoch(),
+          pg->get_osdmap_epoch(),
+          PeeringState::AllReplicasRecovered{});
+      } else {
+        logger().debug("start_recovery_ops: RequestBackfill for pg: {}",
+                      pg->get_pgid());
+        (void) pg->get_shard_services().start_operation<LocalPeeringEvent>(
+          static_cast<crimson::osd::PG*>(pg),
+          pg->get_shard_services(),
+          pg->get_pg_whoami(),
+          pg->get_pgid(),
+          pg->get_osdmap_epoch(),
+          pg->get_osdmap_epoch(),
+          PeeringState::RequestBackfill{});
+      }
     }
     return seastar::make_ready_future<bool>(!done);
   });
@@ -255,15 +277,6 @@ size_t PGRecovery::start_replica_recovery_ops(
   return started;
 }
 
-size_t PGRecovery::start_backfill_ops(
-  size_t max_to_start,
-  std::vector<crimson::osd::blocking_future<>> *out)
-{
-  if (pg->get_peering_state().get_backfill_targets().empty())
-    return 0;
-  ceph_abort("not implemented!");
-}
-
 std::optional<crimson::osd::blocking_future<>> PGRecovery::recover_missing(
   const hobject_t &soid, eversion_t need)
 {

diff --git a/src/crimson/osd/pg_recovery.h b/src/crimson/osd/pg_recovery.h
index 20d462789a85437ccbe566a7b7fd7d1254bacd76..77ee1122f2e817e250499b30723f7251b73bb2b5 100644 (file)
--- a/src/crimson/osd/pg_recovery.h
+++ b/src/crimson/osd/pg_recovery.h
@@ -31,9 +31,6 @@ private:
   size_t start_replica_recovery_ops(
     size_t max_to_start,
     std::vector<crimson::osd::blocking_future<>> *out);
-  size_t start_backfill_ops(
-    size_t max_to_start,
-    std::vector<crimson::osd::blocking_future<>> *out);
 
   std::vector<pg_shard_t> get_replica_recovery_order() const {
     return pg->get_replica_recovery_order();