Merge pull request ceph#63125 from Matan-B/wip-matanb-crimson-tentacle-62847

tentacle: crimson/osd/pg_recovery: rework start_recovery_ops

Reviewed-by: Aishwarya Mathuria <[email protected]>

Author: Matan-B

src/crimson/common/interruptible_future.h

@@ -1486,6 +1486,17 @@ struct interruptor
 	futurize_invoke_if_func(std::forward<FutOrFuncs>(fut_or_funcs))...);
   }
 
+  // This is a simpler implemation than seastar::when_all_succeed.
+  // We are not using ::seastar::internal::complete_when_all
+  template <typename T>
+  static inline auto when_all_succeed(std::vector<interruptible_future<InterruptCond, T>>&& futures) noexcept {
+    return interruptor::parallel_for_each(futures,
+    [] (auto&& ifut) -> interruptible_future<InterruptCond, T> {
+      return std::move(ifut);
+    });
+  }
+
+
   template <typename Func,
 	    typename Result = futurize_t<std::invoke_result_t<Func>>>
   static inline Result async(Func&& func) {

src/crimson/osd/backfill_state.cc

@@ -122,7 +122,7 @@ void BackfillState::Enqueuing::maybe_update_range()
       ceph_assert(primary_bi.version == eversion_t());
       return;
     }
-    DEBUGDPP("{}: bi is old, ({}) can be updated with log to {}",
+    DEBUGDPP("bi is old, ({}) can be updated with log to {}",
 	     pg(),
 	     primary_bi.version,
 	     pg().get_projected_last_update());
@@ -183,9 +183,9 @@ void BackfillState::Enqueuing::maybe_update_range()
           }
         }
       };
-    DEBUGDPP("{}: scanning pg log first", pg());
+    DEBUGDPP("scanning pg log first", pg());
     peering_state().scan_log_after(primary_bi.version, func);
-    DEBUGDPP("{}: scanning projected log", pg());
+    DEBUGDPP("scanning projected log", pg());
     pg().get_projected_log().scan_log_after(primary_bi.version, func);
     primary_bi.version = pg().get_projected_last_update();
   } else {

src/crimson/osd/osd_operations/background_recovery.cc

@@ -81,14 +81,8 @@ seastar::future<> BackgroundRecoveryT<T>::start()
       return interruptor::with_interruption([this] {
        return do_recovery();
       }, [](std::exception_ptr) {
-       return seastar::make_ready_future<bool>(false);
-      }, pg, epoch_started).then([](bool do_recovery) {
-       if (do_recovery) {
-         return seastar::stop_iteration::no;
-       } else {
-         return seastar::stop_iteration::yes;
-       }
-      });
+       return seastar::make_ready_future<seastar::stop_iteration>(seastar::stop_iteration::yes);
+      }, pg, epoch_started);
     });
   });
 }
@@ -105,13 +99,13 @@ UrgentRecovery::UrgentRecovery(
 {
 }
 
-UrgentRecovery::interruptible_future<bool>
+UrgentRecovery::interruptible_future<seastar::stop_iteration>
 UrgentRecovery::do_recovery()
 {
   LOG_PREFIX(UrgentRecovery::do_recovery);
   DEBUGDPPI("{}: {}", *pg, __func__, *this);
   if (pg->has_reset_since(epoch_started)) {
-    return seastar::make_ready_future<bool>(false);
+    return seastar::make_ready_future<seastar::stop_iteration>(seastar::stop_iteration::yes);
   }
 
   return pg->find_unfound(epoch_started
@@ -121,7 +115,7 @@ UrgentRecovery::do_recovery()
       return pg->get_recovery_handler()->recover_missing(
 	trigger, soid, need, false);
     }).then_interruptible([] {
-      return seastar::make_ready_future<bool>(false);
+      return seastar::make_ready_future<seastar::stop_iteration>(seastar::stop_iteration::yes);
     });
   });
 }
@@ -157,13 +151,13 @@ PglogBasedRecovery::PglogBasedRecovery(
       delay)
 {}
 
-PglogBasedRecovery::interruptible_future<bool>
+PglogBasedRecovery::interruptible_future<seastar::stop_iteration>
 PglogBasedRecovery::do_recovery()
 {
   LOG_PREFIX(PglogBasedRecovery::do_recovery);
   DEBUGDPPI("{}: {}", *pg, __func__, *this);
   if (pg->has_reset_since(epoch_started)) {
-    return seastar::make_ready_future<bool>(false);
+    return seastar::make_ready_future<seastar::stop_iteration>(seastar::stop_iteration::yes);
   }
   return pg->find_unfound(epoch_started
   ).then_interruptible([this] {

src/crimson/osd/osd_operations/background_recovery.h

@@ -44,7 +44,7 @@ class BackgroundRecoveryT : public PhasedOperationT<T> {
       scheduler_class
     };
   }
-  using do_recovery_ret_t = typename PhasedOperationT<T>::template interruptible_future<bool>;
+  using do_recovery_ret_t = typename PhasedOperationT<T>::template interruptible_future<seastar::stop_iteration>;
   virtual do_recovery_ret_t do_recovery() = 0;
   ShardServices &ss;
   const crimson::osd::scheduler::scheduler_class_t scheduler_class;
@@ -71,7 +71,7 @@ class UrgentRecovery final : public BackgroundRecoveryT<UrgentRecovery> {
 
 private:
   void dump_detail(Formatter* f) const final;
-  interruptible_future<bool> do_recovery() override;
+  interruptible_future<seastar::stop_iteration> do_recovery() override;
   const hobject_t soid;
   const eversion_t need;
 };
@@ -100,7 +100,7 @@ class PglogBasedRecovery final : public BackgroundRecoveryT<PglogBasedRecovery>
     return epoch_started;
   }
 private:
-  interruptible_future<bool> do_recovery() override;
+  interruptible_future<seastar::stop_iteration> do_recovery() override;
   bool cancelled = false;
 };
 

src/crimson/osd/pg.cc

@@ -630,7 +630,7 @@ void PG::on_active_actmap()
       publish_stats_to_osd();
     });
   } else {
-    logger().debug("{}: pg not clean, skipping snap trim");
+    logger().debug("pg not clean, skipping snap trim");
     ceph_assert(!peering_state.state_test(PG_STATE_SNAPTRIM));
   }
 }

src/crimson/osd/pg.h

@@ -309,7 +309,7 @@ class PG : public boost::intrusive_ref_counter<
     LOG_PREFIX(PG::request_remote_recovery_reservation);
     SUBDEBUGDPP(
       osd, "priority {} on_grant {} on_preempt {}",
-      *this, on_grant->get_desc(), on_preempt->get_desc());
+      *this, priority, on_grant->get_desc(), on_preempt->get_desc());
     shard_services.remote_request_reservation(
       orderer,
       pgid,