General SelfHeal fixes (#4315)

ydb-platform · May 6, 2024 · 77a85af · 77a85af
1 parent 068e829
commit 77a85af
Show file tree

Hide file tree

Showing 9 changed files with 98 additions and 145 deletions.
diff --git a/ydb/core/mind/bscontroller/config_cmd.cpp b/ydb/core/mind/bscontroller/config_cmd.cpp
@@ -99,6 +99,9 @@ namespace NKikimr::NBsController {
                         for (bool value : settings.GetEnableDonorMode()) {
                             Self->DonorMode = value;
                             db.Table<T>().Key(true).Update<T::DonorModeEnable>(Self->DonorMode);
+                            auto ev = std::make_unique<TEvControllerUpdateSelfHealInfo>();
+                            ev->DonorMode = Self->DonorMode;
+                            Self->Send(Self->SelfHealId, ev.release());
                         }
                         for (ui64 value : settings.GetScrubPeriodicitySeconds()) {
                             Self->ScrubPeriodicity = TDuration::Seconds(value);

diff --git a/ydb/core/mind/bscontroller/impl.h b/ydb/core/mind/bscontroller/impl.h
@@ -471,8 +471,8 @@ class TBlobStorageController : public TActor<TBlobStorageController>, public TTa
                 || Status == NKikimrBlobStorage::EDriveStatus::INACTIVE;
         }
 
-        std::tuple<bool, bool> GetSelfHealStatusTuple() const {
-            return {ShouldBeSettledBySelfHeal(), BadInTermsOfSelfHeal()};
+        auto GetSelfHealStatusTuple() const {
+            return std::make_tuple(ShouldBeSettledBySelfHeal(), BadInTermsOfSelfHeal(), Decommitted(), IsSelfHealReasonDecommit());
         }
 
         bool AcceptsNewSlots() const {
@@ -2222,12 +2222,15 @@ class TBlobStorageController : public TActor<TBlobStorageController>, public TTa
         const TMonotonic now = TActivationContext::Monotonic();
         THashSet<TGroupInfo*> groups;
 
-        auto sh = std::make_unique<TEvControllerUpdateSelfHealInfo>();
+        std::vector<TEvControllerUpdateSelfHealInfo::TVDiskStatusUpdate> updates;
         for (auto it = VSlotReadyTimestampQ.begin(); it != VSlotReadyTimestampQ.end() && it->first <= now;
                 it = VSlotReadyTimestampQ.erase(it)) {
             Y_DEBUG_ABORT_UNLESS(!it->second->IsReady);
 
-            sh->VDiskIsReadyUpdate.emplace_back(it->second->GetVDiskId(), true);
+            updates.push_back({
+                .VDiskId = it->second->GetVDiskId(),
+                .IsReady = true,
+            });
             it->second->IsReady = true;
             it->second->ResetVSlotReadyTimestampIter();
             if (const TGroupInfo *group = it->second->Group) {
@@ -2245,8 +2248,8 @@ class TBlobStorageController : public TActor<TBlobStorageController>, public TTa
         if (!timingQ.empty()) {
             Execute(CreateTxUpdateLastSeenReady(std::move(timingQ)));
         }
-        if (sh->VDiskIsReadyUpdate) {
-            Send(SelfHealId, sh.release());
+        if (!updates.empty()) {
+            Send(SelfHealId, new TEvControllerUpdateSelfHealInfo(std::move(updates)));
         }
     }
 

diff --git a/ydb/core/mind/bscontroller/register_node.cpp b/ydb/core/mind/bscontroller/register_node.cpp
@@ -549,25 +549,29 @@ void TBlobStorageController::OnWardenDisconnected(TNodeId nodeId, TActorId serve
         SysViewChangedPDisks.insert(it->first);
     }
     const TVSlotId startingId(nodeId, Min<Schema::VSlot::PDiskID::Type>(), Min<Schema::VSlot::VSlotID::Type>());
-    auto sh = MakeHolder<TEvControllerUpdateSelfHealInfo>();
+    std::vector<TEvControllerUpdateSelfHealInfo::TVDiskStatusUpdate> updates;
     for (auto it = VSlots.lower_bound(startingId); it != VSlots.end() && it->first.NodeId == nodeId; ++it) {
         if (const TGroupInfo *group = it->second->Group) {
             if (it->second->IsReady) {
                 NotReadyVSlotIds.insert(it->second->VSlotId);
-                sh->VDiskIsReadyUpdate.emplace_back(it->second->GetVDiskId(), false);
             }
             it->second->SetStatus(NKikimrBlobStorage::EVDiskStatus::ERROR, mono, now, false);
             timingQ.emplace_back(*it->second);
-            sh->VDiskStatusUpdate.emplace_back(it->second->GetVDiskId(), it->second->Status, false);
+            updates.push_back({
+                .VDiskId = it->second->GetVDiskId(),
+                .IsReady = it->second->IsReady,
+                .VDiskStatus = it->second->Status,
+            });
             ScrubState.UpdateVDiskState(&*it->second);
         }
     }
     for (auto it = StaticVSlots.lower_bound(startingId); it != StaticVSlots.end() && it->first.NodeId == nodeId; ++it) {
-        it->second.VDiskStatus = NKikimrBlobStorage::EVDiskStatus::ERROR;
-        it->second.ReadySince = TMonotonic::Max();
+        auto& slot = it->second;
+        slot.ReadySince = TMonotonic::Max();
+        slot.VDiskStatus = NKikimrBlobStorage::EVDiskStatus::ERROR;
     }
-    if (sh->VDiskStatusUpdate) {
-        Send(SelfHealId, sh.Release());
+    if (!updates.empty()) {
+        Send(SelfHealId, new TEvControllerUpdateSelfHealInfo(std::move(updates)));
     }
     ScrubState.OnNodeDisconnected(nodeId);
     EraseKnownDrivesOnDisconnected(&node);