Add cluster state info into state storage config (#18864)

5 files changed, 88 insertions, 1 deletions

diff --git a/ydb/core/blobstorage/nodewarden/distconf.cpp b/ydb/core/blobstorage/nodewarden/distconf.cpp
index c8c181cb669..b417c16e673 100644
--- a/ydb/core/blobstorage/nodewarden/distconf.cpp
+++ b/ydb/core/blobstorage/nodewarden/distconf.cpp
@@ -442,6 +442,74 @@ namespace NKikimr::NStorage {
         return std::nullopt;
     }
 
+    std::optional<TString> UpdateClusterState(NKikimrBlobStorage::TStorageConfig *config) {
+        // copy bridge info into state storage configs for easier access in replica processors/proxies
+        if (config->HasClusterState()) {
+            auto fillInBridge = [&](auto *pb) -> std::optional<TString> {
+                auto& clusterState = config->GetClusterState();
+                if (!pb->RingGroupsSize() || pb->HasRing()) {
+                    return "configuration has Ring field set or no RingGroups";
+                }
+                auto *groups = pb->MutableRingGroups();
+                for (int i = 0, count = groups->size(); i < count; ++i) {
+                    auto *group = groups->Mutable(i);
+                    if (!group->HasBridgePileId()) {
+                        return "bridge pile id is not set for a ring group";
+                    } else if (ui32 pileId = group->GetBridgePileId(); pileId < clusterState.PerPileStateSize()) {
+                        using T = NKikimrConfig::TDomainsConfig::TStateStorage;
+                        std::optional<T::EPileState> state;
+                        if (pileId == clusterState.GetPrimaryPile()) {
+                            state = pileId == clusterState.GetPromotedPile()
+                                ? T::PRIMARY
+                                : T::DEMOTED;
+                        } else if (pileId == clusterState.GetPromotedPile()) {
+                            state = T::PROMOTED;
+                        } else {
+                            switch (clusterState.GetPerPileState(pileId)) {
+                                case NKikimrBridge::TClusterState::DISCONNECTED:
+                                    state = T::DISCONNECTED;
+                                    break;
+
+                                case NKikimrBridge::TClusterState::NOT_SYNCHRONIZED:
+                                    state = T::NOT_SYNCHRONIZED;
+                                    break;
+
+                                case NKikimrBridge::TClusterState::SYNCHRONIZED:
+                                    state = T::SYNCHRONIZED;
+                                    break;
+
+                                case NKikimrBridge::TClusterState_EPileState_TClusterState_EPileState_INT_MIN_SENTINEL_DO_NOT_USE_:
+                                case NKikimrBridge::TClusterState_EPileState_TClusterState_EPileState_INT_MAX_SENTINEL_DO_NOT_USE_:
+                                    Y_DEBUG_ABORT("unexpected value");
+                            }
+                        }
+                        if (!state) {
+                            return "can't determine correct pile state for ring group";
+                        }
+                        group->SetPileState(*state);
+                    } else {
+                        return "bridge pile id is out of bounds";
+                    }
+                }
+                return std::nullopt;
+            };
+
+            std::optional<TString> error;
+            if (!error && config->HasStateStorageConfig()) {
+                error = fillInBridge(config->MutableStateStorageConfig());
+            }
+            if (!error && config->HasStateStorageBoardConfig()) {
+                error = fillInBridge(config->MutableStateStorageBoardConfig());
+            }
+            if (!error && config->HasSchemeBoardConfig()) {
+                error = fillInBridge(config->MutableSchemeBoardConfig());
+            }
+            return error;
+        }
+
+        return std::nullopt;
+    }
+
 } // NKikimr::NStorage
 
 template<>
diff --git a/ydb/core/blobstorage/nodewarden/distconf.h b/ydb/core/blobstorage/nodewarden/distconf.h
index d8164d078a4..039e71cc6fa 100644
--- a/ydb/core/blobstorage/nodewarden/distconf.h
+++ b/ydb/core/blobstorage/nodewarden/distconf.h
@@ -752,6 +752,8 @@ namespace NKikimr::NStorage {
     std::optional<TString> DecomposeConfig(const TString& configComposite, TString *mainConfigYaml,
         ui64 *mainConfigVersion, TString *mainConfigFetchYaml);
 
+    std::optional<TString> UpdateClusterState(NKikimrBlobStorage::TStorageConfig *config);
+
 } // NKikimr::NStorage
 
 template<>
diff --git a/ydb/core/blobstorage/nodewarden/distconf_generate.cpp b/ydb/core/blobstorage/nodewarden/distconf_generate.cpp
index 0d92992f2d3..d7b7e7a8097 100644
--- a/ydb/core/blobstorage/nodewarden/distconf_generate.cpp
+++ b/ydb/core/blobstorage/nodewarden/distconf_generate.cpp
@@ -120,6 +120,10 @@ namespace NKikimr::NStorage {
 
         config->SetSelfAssemblyUUID(selfAssemblyUUID);
 
+        if (auto error = UpdateClusterState(config)) {
+            return error;
+        }
+
         return std::nullopt;
     }
 
diff --git a/ydb/core/blobstorage/nodewarden/distconf_invoke_common.cpp b/ydb/core/blobstorage/nodewarden/distconf_invoke_common.cpp
index f918f2f220d..092567590c3 100644
--- a/ydb/core/blobstorage/nodewarden/distconf_invoke_common.cpp
+++ b/ydb/core/blobstorage/nodewarden/distconf_invoke_common.cpp
@@ -194,6 +194,9 @@ namespace NKikimr::NStorage {
 
     void TInvokeRequestHandlerActor::StartProposition(NKikimrBlobStorage::TStorageConfig *config, bool updateFields) {
         if (updateFields) {
+            if (auto error = UpdateClusterState(config)) {
+                return FinishWithError(TResult::ERROR, *error);
+            }
             config->MutablePrevConfig()->CopyFrom(*Self->StorageConfig);
             config->MutablePrevConfig()->ClearPrevConfig();
             UpdateFingerprint(config);
diff --git a/ydb/core/protos/config.proto b/ydb/core/protos/config.proto
index 04e0020101d..64cb3871c86 100644
--- a/ydb/core/protos/config.proto
+++ b/ydb/core/protos/config.proto
@@ -186,6 +186,15 @@ message TDynamicNameserviceConfig {
 
 message TDomainsConfig {
     message TStateStorage {
+        enum EPileState {
+            DISCONNECTED = 0; // this pile is disconnected from primary one
+            NOT_SYNCHRONIZED = 1; // this pile is connected to primary pile, but not yet fully synchronized
+            SYNCHRONIZED = 2; // this pile is connected to primary pile and in sync
+            PRIMARY = 3; // this is the primary pile
+            PROMOTED = 4; // this is the synchronized pile which is being promoted to become primary
+            DEMOTED = 5; // this is currently primary pile which is being demoted
+        }
+
         message TRing {
             optional uint32 NToSelect = 1;
             repeated TRing Ring = 2;
@@ -195,6 +204,8 @@ message TDomainsConfig {
             optional bool IsDisabled = 6 [default = false]; // used in the intermediate reconfiguration step
             optional bool WriteOnly = 7 [default = false]; // used for ring groups
             optional uint32 RingGroupActorIdOffset = 8 [default = 0]; // used for ActorId difference for multiple replicas on one node
+            optional uint32 BridgePileId = 9; // bridge pile id for this ring group when bridge mode is enabled
+            optional EPileState PileState = 10; // state for this ring group's pile
         }
 
         optional uint32 SSId = 1 [default = 1];
@@ -202,7 +213,6 @@ message TDomainsConfig {
         optional uint32 StateStorageVersion = 3 [default = 0];
         repeated uint32 CompatibleVersions = 4;
         repeated TRing RingGroups = 5;
-
     }
 
     message TStoragePoolType {