Report status for static group (#1116)

Author: alexvru

ydb/core/mind/bscontroller/bsc.cpp

@@ -75,21 +75,23 @@ void TBlobStorageController::TGroupInfo::CalculateGroupStatus() {
                 failedByPDisk |= {Topology.get(), slot->GetShortVDiskId()};
             }
         }
-        auto deriveStatus = [&](const auto& failed) {
-            auto& checker = *Topology->QuorumChecker;
-            if (!failed.GetNumSetItems()) { // all disks of group are operational
-                return NKikimrBlobStorage::TGroupStatus::FULL;
-            } else if (!checker.CheckFailModelForGroup(failed)) { // fail model exceeded
-                return NKikimrBlobStorage::TGroupStatus::DISINTEGRATED;
-            } else if (checker.IsDegraded(failed)) { // group degraded
-                return NKikimrBlobStorage::TGroupStatus::DEGRADED;
-            } else if (failed.GetNumSetItems()) { // group partially available, but not degraded
-                return NKikimrBlobStorage::TGroupStatus::PARTIAL;
-            } else {
-                Y_ABORT("unexpected case");
-            }
-        };
-        Status.MakeWorst(deriveStatus(failed), deriveStatus(failed | failedByPDisk));
+        Status.MakeWorst(DeriveStatus(Topology.get(), failed), DeriveStatus(Topology.get(), failed | failedByPDisk));
+    }
+}
+
+NKikimrBlobStorage::TGroupStatus::E TBlobStorageController::DeriveStatus(const TBlobStorageGroupInfo::TTopology *topology,
+        const TBlobStorageGroupInfo::TGroupVDisks& failed) {
+    auto& checker = *topology->QuorumChecker;
+    if (!failed.GetNumSetItems()) { // all disks of group are operational
+        return NKikimrBlobStorage::TGroupStatus::FULL;
+    } else if (!checker.CheckFailModelForGroup(failed)) { // fail model exceeded
+        return NKikimrBlobStorage::TGroupStatus::DISINTEGRATED;
+    } else if (checker.IsDegraded(failed)) { // group degraded
+        return NKikimrBlobStorage::TGroupStatus::DEGRADED;
+    } else if (failed.GetNumSetItems()) { // group partially available, but not degraded
+        return NKikimrBlobStorage::TGroupStatus::PARTIAL;
+    } else {
+        Y_ABORT("unexpected case");
     }
 }
 

ydb/core/mind/bscontroller/cmds_storage_pool.cpp

@@ -568,6 +568,7 @@ namespace NKikimr::NBsController {
             if (const auto& s = Self.StorageConfig; s.HasBlobStorageConfig()) {
                 if (const auto& bsConfig = s.GetBlobStorageConfig(); bsConfig.HasServiceSet()) {
                     const auto& ss = bsConfig.GetServiceSet();
+                    const TMonotonic mono = TActivationContext::Monotonic();
                     for (const auto& group : ss.GetGroups()) {
                         auto *x = pb->AddGroup();
                         x->SetGroupId(group.GetGroupID());
@@ -581,6 +582,45 @@ namespace NKikimr::NBsController {
                                 }
                             }
                         }
+
+                        TStringStream err;
+                        auto info = TBlobStorageGroupInfo::Parse(group, nullptr, &err);
+                        Y_VERIFY_DEBUG_S(info, "failed to parse static group, error# " << err.Str());
+                        if (info) {
+                            const auto *topology = &info->GetTopology();
+
+                            TBlobStorageGroupInfo::TGroupVDisks failed(topology);
+                            TBlobStorageGroupInfo::TGroupVDisks failedByPDisk(topology);
+
+                            ui32 realmIdx = 0;
+                            for (const auto& realm : group.GetRings()) {
+                                ui32 domainIdx = 0;
+                                for (const auto& domain : realm.GetFailDomains()) {
+                                    ui32 vdiskIdx = 0;
+                                    for (const auto& location : domain.GetVDiskLocations()) {
+                                        const TVSlotId vslotId(location.GetNodeID(), location.GetPDiskID(), location.GetVDiskSlotID());
+                                        const TVDiskIdShort vdiskId(realmIdx, domainIdx, vdiskIdx);
+
+                                        if (const auto it = StaticVSlots.find(vslotId); it != StaticVSlots.end()) {
+                                            if (mono <= it->second.ReadySince) { // VDisk can't be treated as READY one
+                                                failed |= {topology, vdiskId};
+                                            } else if (const TPDiskInfo *pdisk = PDisks.Find(vslotId.ComprisingPDiskId()); !pdisk || !pdisk->HasGoodExpectedStatus()) {
+                                                failedByPDisk |= {topology, vdiskId};
+                                            }
+                                        } else {
+                                            failed |= {topology, vdiskId};
+                                        }
+
+                                        ++vdiskIdx;
+                                    }
+                                    ++domainIdx;
+                                }
+                                ++realmIdx;
+                            }
+
+                            x->SetOperatingStatus(DeriveStatus(topology, failed));
+                            x->SetExpectedStatus(DeriveStatus(topology, failed | failedByPDisk));
+                        }
                     }
                 }
             }

ydb/core/mind/bscontroller/impl.h

@@ -82,6 +82,9 @@ class TBlobStorageController : public TActor<TBlobStorageController>, public TTa
 
     using TVSlotReadyTimestampQ = std::list<std::pair<TMonotonic, TVSlotInfo*>>;
 
+    // VDisk will be considered READY during this period after reporting its READY state
+    static constexpr TDuration ReadyStablePeriod = TDuration::Seconds(15);
+
     class TVSlotInfo : public TIndirectReferable<TVSlotInfo> {
     public:
         using Table = Schema::VSlot;
@@ -121,9 +124,6 @@ class TBlobStorageController : public TActor<TBlobStorageController>, public TTa
         TVSlotReadyTimestampQ& VSlotReadyTimestampQ;
         TVSlotReadyTimestampQ::iterator VSlotReadyTimestampIter;
 
-        // VDisk will be considered READY during this period after reporting its READY state
-        static constexpr TDuration ReadyStablePeriod = TDuration::Seconds(15);
-
     public:
         NKikimrBlobStorage::EVDiskStatus Status = NKikimrBlobStorage::EVDiskStatus::INIT_PENDING;
         bool IsReady = false;
@@ -2237,6 +2237,7 @@ class TBlobStorageController : public TActor<TBlobStorageController>, public TTa
 
         std::optional<NKikimrBlobStorage::TVDiskMetrics> VDiskMetrics;
         NKikimrBlobStorage::EVDiskStatus VDiskStatus = NKikimrBlobStorage::EVDiskStatus::ERROR;
+        TMonotonic ReadySince = TMonotonic::Max(); // when IsReady becomes true for this disk; Max() in non-READY state
 
         TStaticVSlotInfo(const NKikimrBlobStorage::TNodeWardenServiceSet::TVDisk& vdisk)
             : VDiskId(VDiskIDFromVDiskID(vdisk.GetVDiskID()))
@@ -2306,6 +2307,9 @@ class TBlobStorageController : public TActor<TBlobStorageController>, public TTa
         const TGroupInfo& group, const TVSlotFinder& finder);
     static void SerializeGroupInfo(NKikimrBlobStorage::TGroupInfo *group, const TGroupInfo& groupInfo,
         const TString& storagePoolName, const TMaybe<TKikimrScopeId>& scopeId);
+
+    static NKikimrBlobStorage::TGroupStatus::E DeriveStatus(const TBlobStorageGroupInfo::TTopology *topology,
+        const TBlobStorageGroupInfo::TGroupVDisks& failed);
 };
 
 } //NBsController

ydb/core/mind/bscontroller/self_heal.cpp

@@ -971,6 +971,11 @@ namespace NKikimr::NBsController {
             }
             if (const auto it = StaticVSlots.find(vslotId); it != StaticVSlots.end() && it->second.VDiskId == vdiskId) {
                 it->second.VDiskStatus = m.GetStatus();
+                if (it->second.VDiskStatus == NKikimrBlobStorage::EVDiskStatus::READY) {
+                    it->second.ReadySince = Min(it->second.ReadySince, mono + ReadyStablePeriod);
+                } else {
+                    it->second.ReadySince = TMonotonic::Max();
+                }
             }
         }