Make cooldown timeout on VDisk flexible

ydb-platform · May 15, 2024 · 737db6b · 737db6b
1 parent ed038e2
commit 737db6b
Show file tree

Hide file tree

Showing 4 changed files with 18 additions and 36 deletions.
diff --git a/ydb/core/blobstorage/dsproxy/dsproxy_put.cpp b/ydb/core/blobstorage/dsproxy/dsproxy_put.cpp
@@ -138,7 +138,7 @@ class TBlobStorageGroupPutRequest : public TBlobStorageGroupRequestActor<TBlobSt
     }
 
     void HandleIncarnation(TMonotonic timestamp, ui32 orderNumber, ui64 incarnationGuid) {
-        timestamp += TDuration::Seconds(15); // TODO: cooldown timeout
+        timestamp += VDiskCooldownTimeoutOnProxy;
 
         Y_ABORT_UNLESS(orderNumber < IncarnationRecords.size());
         auto& record = IncarnationRecords[orderNumber];

diff --git a/ydb/core/blobstorage/nodewarden/node_warden_vdisk.cpp b/ydb/core/blobstorage/nodewarden/node_warden_vdisk.cpp
@@ -6,8 +6,6 @@
 
 namespace NKikimr::NStorage {
 
-    constexpr TDuration PDISK_CONFIDENCE_DELAY = TDuration::Seconds(15);
-
     void TNodeWarden::DestroyLocalVDisk(TVDiskRecord& vdisk) {
         STLOG(PRI_INFO, BS_NODE, NW35, "DestroyLocalVDisk", (VDiskId, vdisk.GetVDiskId()), (VSlotId, vdisk.GetVSlotId()));
         Y_ABORT_UNLESS(!vdisk.RuntimeData);
@@ -296,7 +294,7 @@ namespace NKikimr::NStorage {
             StartLocalVDiskActor(record, TDuration::Zero());
         } else if (record.RuntimeData->DonorMode < record.Config.HasDonorMode() || record.RuntimeData->ReadOnly != record.Config.GetReadOnly()) {
             PoisonLocalVDisk(record);
-            StartLocalVDiskActor(record, PDISK_CONFIDENCE_DELAY);
+            StartLocalVDiskActor(record, VDiskCooldownTimeout);
         }
     }
 
@@ -323,7 +321,7 @@ namespace NKikimr::NStorage {
             auto& record = it->second;
             if (record.GetVDiskId() == vDiskId) {
                 PoisonLocalVDisk(record);
-                StartLocalVDiskActor(record, PDISK_CONFIDENCE_DELAY);
+                StartLocalVDiskActor(record, VDiskCooldownTimeout);
                 break;
             }
         }

diff --git a/ydb/core/blobstorage/vdisk/localrecovery/localrecovery_public.cpp b/ydb/core/blobstorage/vdisk/localrecovery/localrecovery_public.cpp
@@ -82,8 +82,6 @@ namespace NKikimr {
         bool HugeKeeperInitialized = false;
         ui64 RecoveredLsn = 0;
         ui64 SyncLogMaxLsnStored = 0;
-        bool CooldownTimerHit = false;
-        bool DatabaseLoaded = false;
         NKikimrVDiskData::TScrubEntrypoint ScrubEntrypoint;
         ui64 ScrubEntrypointLsn = 0;
 
@@ -215,11 +213,8 @@ namespace NKikimr {
             ActiveActors.Erase(ev->Sender);
             auto *msg = ev->Get();
             RecoveredLsn = msg->RecoveredLsn;
-            if (RecoveredLsn) {
-                CooldownTimerHit = true; // we do not apply cooldown timer when there are entries in the log
-            }
             if (msg->Status == NKikimrProto::OK) {
-                PostReplayRecoveryLog(ctx);
+                SignalSuccessAndDie(ctx);
             } else {
                 SignalErrorAndDie(ctx, msg->Status, msg->ErrorReason);
             }
@@ -560,19 +555,6 @@ namespace NKikimr {
                     AfterDatabaseLoaded(ctx);
             }
         }
-        void PostReplayRecoveryLog(const TActorContext &ctx) {
-            DatabaseLoaded = true;
-            if (DatabaseLoaded && CooldownTimerHit) {
-                SignalSuccessAndDie(ctx);
-            }
-        }
-
-        void HandleWakeup(const TActorContext &ctx) {
-            CooldownTimerHit = true;
-            if (DatabaseLoaded && CooldownTimerHit) {
-                SignalSuccessAndDie(ctx);
-            }
-        }
 
         void SendYardInit(const TActorContext &ctx, TDuration yardInitDelay) {
             auto ev = std::make_unique<NPDisk::TEvYardInit>(Config->BaseInfo.InitOwnerRound, SelfVDiskId,
@@ -596,8 +578,8 @@ namespace NKikimr {
             LOG_NOTICE(ctx, BS_LOCALRECOVERY,
                        VDISKP(LocRecCtx->VCtx->VDiskLogPrefix, "LocalRecovery START"));
 
-            SendYardInit(ctx, Config->BaseInfo.YardInitDelay);
-            Become(&TThis::StateInitialize, ctx, VDiskCooldownTimeout, new TEvents::TEvWakeup);
+            SendYardInit(ctx, TDuration::Zero());
+            Become(&TThis::StateInitialize);
             VDiskMonGroup.VDiskLocalRecoveryState() = TDbMon::TDbLocalRecovery::YardInit;
         }
 
@@ -649,28 +631,24 @@ namespace NKikimr {
             HFunc(TEvents::TEvUndelivered, HandleUndelivered)
             CFunc(NActors::TEvents::TSystem::PoisonPill, HandlePoison)
             HFunc(NMon::TEvHttpInfo, Handle)
-            CFunc(TEvents::TSystem::Wakeup, HandleWakeup);
         )
 
         STRICT_STFUNC(StateLoadDatabase,
             HFunc(THullIndexLoaded, Handle)
             CFunc(NActors::TEvents::TSystem::PoisonPill, HandlePoison)
             HFunc(NMon::TEvHttpInfo, Handle)
-            CFunc(TEvents::TSystem::Wakeup, HandleWakeup);
         )
 
         STRICT_STFUNC(StateLoadBulkFormedSegments,
             HFunc(TEvBulkSstsLoaded, Handle)
             CFunc(NActors::TEvents::TSystem::PoisonPill, HandlePoison)
             HFunc(NMon::TEvHttpInfo, Handle)
-            CFunc(TEvents::TSystem::Wakeup, HandleWakeup);
         )
 
         STRICT_STFUNC(StateApplyRecoveryLog,
             HFunc(TEvRecoveryLogReplayDone, Handle)
             CFunc(NActors::TEvents::TSystem::PoisonPill, HandlePoison)
             HFunc(NMon::TEvHttpInfo, Handle)
-            CFunc(TEvents::TSystem::Wakeup, HandleWakeup);
         )
 
     public:
@@ -693,11 +671,7 @@ namespace NKikimr {
             , LocRecCtx(std::make_shared<TLocalRecoveryContext>(vctx))
             , Arena(std::move(arena))
             , VDiskMonGroup(vctx->VDiskCounters, "subsystem", "state")
-        {
-            if (!Config->EnableVDiskCooldownTimeout || LocRecCtx->VCtx->Top->GType.BlobSubgroupSize() == 1) {
-                CooldownTimerHit = true;
-            }
-        }
+        {}
     };
 
 

diff --git a/ydb/core/blobstorage/vdisk/syncer/blobstorage_syncer.cpp b/ydb/core/blobstorage/vdisk/syncer/blobstorage_syncer.cpp
@@ -298,10 +298,19 @@ namespace NKikimr {
             }
 
             Become(&TThis::RecoverLostDataStateFunc);
+            Phase = TPhaseVal::PhaseRecoverLostData;
+
+            if (SyncerCtx->Config->EnableVDiskCooldownTimeout) {
+                Schedule(SyncerCtx->Config->BaseInfo.YardInitDelay, new TEvents::TEvWakeup);
+            } else {
+                RecoverLostDataResumeAfterDelay(ctx);
+            }
+        }
+
+        void RecoverLostDataResumeAfterDelay(const TActorContext& ctx) {
             const TVDiskEternalGuid guid = GuidRecovOutcome->Guid;
             RecoverLostDataId = ctx.Register(CreateSyncerRecoverLostDataActor(SyncerCtx, GInfo, CommitterId, ctx.SelfID, guid));
             ActiveActors.Insert(RecoverLostDataId, __FILE__, __LINE__, ctx, NKikimrServices::BLOBSTORAGE);
-            Phase = TPhaseVal::PhaseRecoverLostData;
         }
 
         void Handle(TEvSyncerLostDataRecovered::TPtr &ev, const TActorContext &ctx) {
@@ -322,6 +331,7 @@ namespace NKikimr {
             HFunc(TEvents::TEvPoisonPill, HandlePoison)
             HFunc(TEvSublogLine, Handle)
             HFunc(TEvVGenerationChange, RecoverLostDataModeHandle)
+            CFunc(TEvents::TSystem::Wakeup, RecoverLostDataResumeAfterDelay);
         )
 
         ////////////////////////////////////////////////////////////////////////