Refactor bridge mode info returned in TEvNodeWardenStorageConfig (#18744)

10 files changed, 82 insertions, 30 deletions

diff --git a/ydb/core/base/bridge.h b/ydb/core/base/bridge.h
new file mode 100644
index 00000000000..2694bb45b64
--- /dev/null
+++ b/ydb/core/base/bridge.h
@@ -0,0 +1,28 @@
+#pragma once
+
+#include "defs.h"
+#include "blobstorage_common.h"
+
+#include <ydb/core/protos/bridge.pb.h>
+
+namespace NKikimr {
+
+    struct TBridgeInfo {
+        struct TPile {
+            TBridgePileId BridgePileId; // essentially an index into TBridgeInfo::Piles array
+            std::vector<ui32> StaticNodeIds; // a sorted set of static node ids belonging to this very pile
+            NKikimrBridge::TClusterState::EPileState State = {}; // state of this pile
+            bool IsPrimary = false; // is this pile selected as primary
+            bool IsBeingPromoted = false; // is this pile being promoted right now (and not the primary, but in sync with one)
+        };
+
+        THashMap<ui32, const TPile*> StaticNodeIdToPile; // node to pile map for static nodes
+        std::vector<TPile> Piles; // a vector of piles
+        const TPile *SelfNodePile = nullptr; // a reference to pile this node belongs to
+        const TPile *PrimaryPile = nullptr; // a reference to the primary pile
+        const TPile *BeingPromotedPile = nullptr; // a reference to the pile being promoted, or nullptr if none are promoted
+
+        using TPtr = std::shared_ptr<const TBridgeInfo>;
+    };
+
+} // NKikimr
diff --git a/ydb/core/blobstorage/base/blobstorage_events.cpp b/ydb/core/blobstorage/base/blobstorage_events.cpp
index 49c0ee82de8..a1144a07390 100644
--- a/ydb/core/blobstorage/base/blobstorage_events.cpp
+++ b/ydb/core/blobstorage/base/blobstorage_events.cpp
@@ -5,14 +5,13 @@ namespace NKikimr {
 
     TEvNodeWardenStorageConfig::TEvNodeWardenStorageConfig(const NKikimrBlobStorage::TStorageConfig& config,
             const NKikimrBlobStorage::TStorageConfig *proposedConfig, bool selfManagementEnabled,
-            bool isPrimary, bool isBeingPromoted)
+            TBridgeInfo::TPtr bridgeInfo)
         : Config(std::make_unique<NKikimrBlobStorage::TStorageConfig>(config))
         , ProposedConfig(proposedConfig
             ? std::make_unique<NKikimrBlobStorage::TStorageConfig>(*proposedConfig)
             : nullptr)
         , SelfManagementEnabled(selfManagementEnabled)
-        , IsPrimary(isPrimary)
-        , IsBeingPromoted(isBeingPromoted)
+        , BridgeInfo(std::move(bridgeInfo))
     {}
 
     TEvNodeWardenStorageConfig::~TEvNodeWardenStorageConfig()
diff --git a/ydb/core/blobstorage/base/blobstorage_events.h b/ydb/core/blobstorage/base/blobstorage_events.h
index d6ddbf0e1c7..8883feebf38 100644
--- a/ydb/core/blobstorage/base/blobstorage_events.h
+++ b/ydb/core/blobstorage/base/blobstorage_events.h
@@ -3,6 +3,7 @@
 
 #include "blobstorage_vdiskid.h"
 #include <ydb/core/base/blobstorage.h>
+#include <ydb/core/base/bridge.h>
 #include <ydb/core/blobstorage/groupinfo/blobstorage_groupinfo.h>
 #include <ydb/core/blobstorage/pdisk/blobstorage_pdisk_config.h>
 #include <ydb/core/blobstorage/pdisk/blobstorage_pdisk_defs.h>
@@ -583,12 +584,11 @@ namespace NKikimr {
         std::unique_ptr<NKikimrBlobStorage::TStorageConfig> Config;
         std::unique_ptr<NKikimrBlobStorage::TStorageConfig> ProposedConfig;
         bool SelfManagementEnabled;
-        bool IsPrimary; // if bridge mode is enabled
-        bool IsBeingPromoted; // if bridge mode is enabled
+        TBridgeInfo::TPtr BridgeInfo;
 
         TEvNodeWardenStorageConfig(const NKikimrBlobStorage::TStorageConfig& config,
                 const NKikimrBlobStorage::TStorageConfig *proposedConfig, bool selfManagementEnabled,
-                bool isPrimary, bool isBeingPromoted);
+                TBridgeInfo::TPtr bridgeInfo);
         ~TEvNodeWardenStorageConfig();
     };
 
diff --git a/ydb/core/blobstorage/nodewarden/distconf.cpp b/ydb/core/blobstorage/nodewarden/distconf.cpp
index a3fce1f7e20..c24bf5e9ae9 100644
--- a/ydb/core/blobstorage/nodewarden/distconf.cpp
+++ b/ydb/core/blobstorage/nodewarden/distconf.cpp
@@ -102,21 +102,51 @@ namespace NKikimr::NStorage {
                 config.GetSelfManagementConfig().GetEnabled() &&
                 config.GetGeneration();
 
-            if (config.HasClusterState()) {
+            if (config.HasClusterState() && Cfg->BridgeConfig) {
                 const auto& state = config.GetClusterState();
-                bool found = false;
+
+                // prepare empty structure
+                auto bridgeInfo = std::make_shared<TBridgeInfo>();
+                bridgeInfo->Piles.resize(Cfg->BridgeConfig->PilesSize());
+
                 for (const auto& node : config.GetAllNodes()) {
-                    if (node.GetNodeId() == SelfNode.NodeId() && node.HasBridgePileId()) {
-                        IsPrimary = node.GetBridgePileId() == state.GetPrimaryPile();
-                        IsBeingPromoted = !IsPrimary && node.GetBridgePileId() == state.GetPromotedPile();
-                        found = true;
-                        break;
+                    if (node.HasBridgePileId()) {
+                        const ui32 nodeId = node.GetNodeId();
+                        const ui32 pileId = node.GetBridgePileId();
+                        Y_ABORT_UNLESS(pileId < bridgeInfo->Piles.size());
+                        auto& pile = bridgeInfo->Piles[pileId];
+                        pile.StaticNodeIds.push_back(node.GetNodeId());
+                        bridgeInfo->StaticNodeIdToPile[nodeId] = &pile;
+                        if (nodeId == SelfNode.NodeId()) {
+                            bridgeInfo->SelfNodePile = &pile;
+                        }
                     }
                 }
-                if (!found) {
-                    // TODO(alexvru): think about better handling of node removal
-                    IsPrimary = IsBeingPromoted = false;
+
+                Y_ABORT_UNLESS(state.PerPileStateSize() == Cfg->BridgeConfig->PilesSize());
+                for (size_t i = 0; i < state.PerPileStateSize(); ++i) {
+                    auto& pile = bridgeInfo->Piles[i];
+                    pile.BridgePileId = TBridgePileId::FromValue(i);
+                    pile.State = state.GetPerPileState(i);
+                    std::ranges::sort(pile.StaticNodeIds);
                 }
+
+                const ui32 primary = state.GetPrimaryPile();
+                Y_ABORT_UNLESS(primary < Cfg->BridgeConfig->PilesSize());
+                bridgeInfo->Piles[primary].IsPrimary = true;
+                bridgeInfo->PrimaryPile = &bridgeInfo->Piles[primary];
+
+                if (const ui32 promoted = state.GetPromotedPile(); promoted != primary) {
+                    Y_ABORT_UNLESS(promoted < Cfg->BridgeConfig->PilesSize());
+                    auto& pile = bridgeInfo->Piles[promoted];
+                    Y_ABORT_UNLESS(pile.State == NKikimrBridge::TClusterState::SYNCHRONIZED);
+                    pile.IsBeingPromoted = true;
+                    bridgeInfo->BeingPromotedPile = &pile;
+                }
+
+                BridgeInfo = std::move(bridgeInfo);
+            } else {
+                Y_ABORT_UNLESS(!BridgeInfo);
             }
 
             StorageConfig.emplace(config);
@@ -327,7 +357,7 @@ namespace NKikimr::NStorage {
             ? &ProposedStorageConfig.value()
             : nullptr;
         auto ev = std::make_unique<TEvNodeWardenStorageConfig>(*config, proposedConfig, SelfManagementEnabled,
-            IsPrimary, IsBeingPromoted);
+            BridgeInfo);
         Send(wardenId, ev.release(), 0, cookie);
     }
 
diff --git a/ydb/core/blobstorage/nodewarden/distconf.h b/ydb/core/blobstorage/nodewarden/distconf.h
index 7a317a01554..e6e8b4b754b 100644
--- a/ydb/core/blobstorage/nodewarden/distconf.h
+++ b/ydb/core/blobstorage/nodewarden/distconf.h
@@ -175,8 +175,7 @@ namespace NKikimr::NStorage {
         const bool IsSelfStatic = false;
         TIntrusivePtr<TNodeWardenConfig> Cfg;
         bool SelfManagementEnabled = false;
-        bool IsPrimary = false;
-        bool IsBeingPromoted = false;
+        TBridgeInfo::TPtr BridgeInfo;
 
         // currently active storage config
         std::optional<NKikimrBlobStorage::TStorageConfig> StorageConfig;
diff --git a/ydb/core/blobstorage/nodewarden/node_warden_impl.h b/ydb/core/blobstorage/nodewarden/node_warden_impl.h
index e4816e7bb63..2539937f553 100644
--- a/ydb/core/blobstorage/nodewarden/node_warden_impl.h
+++ b/ydb/core/blobstorage/nodewarden/node_warden_impl.h
@@ -639,8 +639,7 @@ namespace NKikimr::NStorage {
 
         NKikimrBlobStorage::TStorageConfig StorageConfig;
         bool SelfManagementEnabled = false;
-        bool IsPrimary = false;
-        bool IsBeingPromoted = false;
+        TBridgeInfo::TPtr BridgeInfo;
         THashSet<TActorId> StorageConfigSubscribers;
 
         void Handle(TEvNodeWardenQueryStorageConfig::TPtr ev);
diff --git a/ydb/core/blobstorage/nodewarden/node_warden_resource.cpp b/ydb/core/blobstorage/nodewarden/node_warden_resource.cpp
index 96a7be0d32c..f3d2636f867 100644
--- a/ydb/core/blobstorage/nodewarden/node_warden_resource.cpp
+++ b/ydb/core/blobstorage/nodewarden/node_warden_resource.cpp
@@ -77,8 +77,7 @@ void TNodeWarden::ApplyServiceSet(const NKikimrBlobStorage::TNodeWardenServiceSe
 }
 
 void TNodeWarden::Handle(TEvNodeWardenQueryStorageConfig::TPtr ev) {
-    Send(ev->Sender, new TEvNodeWardenStorageConfig(StorageConfig, nullptr, SelfManagementEnabled, IsPrimary,
-        IsBeingPromoted));
+    Send(ev->Sender, new TEvNodeWardenStorageConfig(StorageConfig, nullptr, SelfManagementEnabled, BridgeInfo));
     if (ev->Get()->Subscribe) {
         StorageConfigSubscribers.insert(ev->Sender);
     }
@@ -88,8 +87,7 @@ void TNodeWarden::Handle(TEvNodeWardenStorageConfig::TPtr ev) {
     auto *msg = ev->Get();
     msg->Config->Swap(&StorageConfig);
     SelfManagementEnabled = msg->SelfManagementEnabled;
-    IsPrimary = msg->IsPrimary;
-    IsBeingPromoted = msg->IsBeingPromoted;
+    BridgeInfo = std::move(msg->BridgeInfo);
 
     if (StorageConfig.HasBlobStorageConfig()) {
         if (const auto& bsConfig = StorageConfig.GetBlobStorageConfig(); bsConfig.HasServiceSet()) {
@@ -115,8 +113,7 @@ void TNodeWarden::Handle(TEvNodeWardenStorageConfig::TPtr ev) {
     }
 
     for (const TActorId& subscriber : StorageConfigSubscribers) {
-        Send(subscriber, new TEvNodeWardenStorageConfig(StorageConfig, nullptr, SelfManagementEnabled,
-            IsPrimary, IsBeingPromoted));
+        Send(subscriber, new TEvNodeWardenStorageConfig(StorageConfig, nullptr, SelfManagementEnabled, BridgeInfo));
     }
 
     if (StorageConfig.HasConfigComposite()) {
diff --git a/ydb/core/blobstorage/ut_blobstorage/lib/node_warden_mock_bsc.cpp b/ydb/core/blobstorage/ut_blobstorage/lib/node_warden_mock_bsc.cpp
index 3fcf0854a21..fac9cc3c782 100644
--- a/ydb/core/blobstorage/ut_blobstorage/lib/node_warden_mock_bsc.cpp
+++ b/ydb/core/blobstorage/ut_blobstorage/lib/node_warden_mock_bsc.cpp
@@ -226,7 +226,7 @@ void TNodeWardenMockActor::Handle(TEvBlobStorage::TEvControllerNodeServiceSetUpd
 }
 
 void TNodeWardenMockActor::Handle(TEvNodeWardenQueryStorageConfig::TPtr ev) {
-    Send(ev->Sender, new TEvNodeWardenStorageConfig(NKikimrBlobStorage::TStorageConfig(), nullptr, false, false, false));
+    Send(ev->Sender, new TEvNodeWardenStorageConfig(NKikimrBlobStorage::TStorageConfig(), nullptr, false, nullptr));
 }
 
 void TNodeWardenMockActor::HandleUnsubscribe(STATEFN_SIG) {
diff --git a/ydb/core/mind/bscontroller/ut_bscontroller/main.cpp b/ydb/core/mind/bscontroller/ut_bscontroller/main.cpp
index 3daa37d7dc2..1c76d926086 100644
--- a/ydb/core/mind/bscontroller/ut_bscontroller/main.cpp
+++ b/ydb/core/mind/bscontroller/ut_bscontroller/main.cpp
@@ -243,7 +243,7 @@ struct TEnvironmentSetup {
             {}
 
             void Handle(TEvNodeWardenQueryStorageConfig::TPtr ev) {
-                Send(ev->Sender, new TEvNodeWardenStorageConfig(NKikimrBlobStorage::TStorageConfig(), nullptr, false, false, false));
+                Send(ev->Sender, new TEvNodeWardenStorageConfig(NKikimrBlobStorage::TStorageConfig(), nullptr, false, nullptr));
             }
 
             STATEFN(StateFunc) {
diff --git a/ydb/core/mind/bscontroller/ut_selfheal/node_warden_mock.h b/ydb/core/mind/bscontroller/ut_selfheal/node_warden_mock.h
index 16e9e78be3b..966849f54b8 100644
--- a/ydb/core/mind/bscontroller/ut_selfheal/node_warden_mock.h
+++ b/ydb/core/mind/bscontroller/ut_selfheal/node_warden_mock.h
@@ -180,7 +180,7 @@ public:
     }
 
     void Handle(TEvNodeWardenQueryStorageConfig::TPtr ev) {
-        Send(ev->Sender, new TEvNodeWardenStorageConfig(NKikimrBlobStorage::TStorageConfig(), nullptr, false, false, false));
+        Send(ev->Sender, new TEvNodeWardenStorageConfig(NKikimrBlobStorage::TStorageConfig(), nullptr, false, nullptr));
     }
 
     STRICT_STFUNC(StateFunc, {