22-2: Reply ALREADY to TEvBlock races, don't hide errors by retrying blocking, KIKIMR-15235

Merge from trunk: r9662422

REVIEW: 2695485
x-ydb-stable-ref: 8bd7b98b3b008990ac0cd4adb76f30a51bbc2014

6 files changed, 10 insertions, 16 deletions

diff --git a/ydb/core/blobstorage/dsproxy/dsproxy_block.cpp b/ydb/core/blobstorage/dsproxy/dsproxy_block.cpp
index dba0a2fd43..fdfd95b37a 100644
--- a/ydb/core/blobstorage/dsproxy/dsproxy_block.cpp
+++ b/ydb/core/blobstorage/dsproxy/dsproxy_block.cpp
@@ -59,7 +59,7 @@ class TBlobStorageGroupBlockRequest : public TBlobStorageGroupRequestActor<TBlob
         std::vector<TVDiskID> queryStatus, resend;
         if (status == NKikimrProto::ALREADY) {
             // ALREADY means that newly arrived Block is the same or older than existing one; we treat it as ERROR here
-            // and reply with RACE only when no quorum could be obtained during the whole operation
+            // and reply with ALREADY only when no quorum could be obtained during the whole operation
             SeenAlready = true;
             status = NKikimrProto::ERROR;
         }
@@ -74,7 +74,7 @@ class TBlobStorageGroupBlockRequest : public TBlobStorageGroupRequestActor<TBlob
 
             case NKikimrProto::ERROR: {
                 TStringStream err;
-                newStatus = SeenAlready ? NKikimrProto::RACE : NKikimrProto::ERROR;
+                newStatus = SeenAlready ? NKikimrProto::ALREADY : NKikimrProto::ERROR;
                 err << "Status# " << NKikimrProto::EReplyStatus_Name(newStatus)
                     << " From# " << vdisk.ToString()
                     << " NodeId# " << Info->GetActorId(vdisk).NodeId()
diff --git a/ydb/core/blobstorage/dsproxy/mock/model.h b/ydb/core/blobstorage/dsproxy/mock/model.h
index 6161a6b857..f8a7f78d4a 100644
--- a/ydb/core/blobstorage/dsproxy/mock/model.h
+++ b/ydb/core/blobstorage/dsproxy/mock/model.h
@@ -137,7 +137,7 @@ namespace NFake {
             if (it == Blocks.end()) {
                 Blocks.emplace(msg->TabletId, msg->Generation);
             } else if (msg->Generation <= it->second) {
-                status = NKikimrProto::RACE;
+                status = NKikimrProto::ALREADY;
             } else {
                 it->second = msg->Generation;
             }
diff --git a/ydb/core/blobstorage/nodewarden/blobstorage_node_warden_ut.cpp b/ydb/core/blobstorage/nodewarden/blobstorage_node_warden_ut.cpp
index e3c74cce7b..aa9ab4027e 100644
--- a/ydb/core/blobstorage/nodewarden/blobstorage_node_warden_ut.cpp
+++ b/ydb/core/blobstorage/nodewarden/blobstorage_node_warden_ut.cpp
@@ -461,8 +461,8 @@ Y_UNIT_TEST_SUITE(TBlobStorageWardenTest) {
         ui64 tabletId = 1234;
         ui32 generation = 1;
         BlockGroup(runtime, sender0, tabletId, groupId, generation, true);
-        BlockGroup(runtime, sender0, tabletId, groupId, generation, true, NKikimrProto::EReplyStatus::RACE);
-        BlockGroup(runtime, sender0, tabletId, groupId, generation-1, true, NKikimrProto::EReplyStatus::RACE);
+        BlockGroup(runtime, sender0, tabletId, groupId, generation, true, NKikimrProto::EReplyStatus::ALREADY);
+        BlockGroup(runtime, sender0, tabletId, groupId, generation-1, true, NKikimrProto::EReplyStatus::ALREADY);
 
         auto describePool = DescribeStoragePool(runtime, DOMAIN_ID, "test_storage");
         {
diff --git a/ydb/core/blobstorage/testload/test_load_write.cpp b/ydb/core/blobstorage/testload/test_load_write.cpp
index d30f9d4a31..a7ac7823e8 100644
--- a/ydb/core/blobstorage/testload/test_load_write.cpp
+++ b/ydb/core/blobstorage/testload/test_load_write.cpp
@@ -271,9 +271,9 @@ class TLogWriterTestLoadActor : public TActorBootstrapped<TLogWriterTestLoadActo
             auto callback = [this] (IEventBase *event, const TActorContext& ctx) {
                 auto *res = dynamic_cast<TEvBlobStorage::TEvBlockResult *>(event);
                 Y_VERIFY(res);
-                if (!CheckStatus(ctx, res, {NKikimrProto::EReplyStatus::OK, NKikimrProto::EReplyStatus::RACE})) {
+                if (!CheckStatus(ctx, res, {NKikimrProto::EReplyStatus::OK, NKikimrProto::EReplyStatus::ALREADY})) {
                     return;
-                } else if (res->Status == NKikimrProto::EReplyStatus::RACE && GroupBlockRetries-- > 0) {
+                } else if (res->Status == NKikimrProto::EReplyStatus::ALREADY && GroupBlockRetries-- > 0) {
                     LOG_INFO_S(ctx, NKikimrServices::BS_LOAD_TEST, PrintMe() << " recieved " << res->ToString());
                     IssueTEvBlock(ctx);
                     return;
diff --git a/ydb/core/mind/hive/tx__block_storage_result.cpp b/ydb/core/mind/hive/tx__block_storage_result.cpp
index 28315ec93b..ee2cfe8763 100644
--- a/ydb/core/mind/hive/tx__block_storage_result.cpp
+++ b/ydb/core/mind/hive/tx__block_storage_result.cpp
@@ -41,6 +41,7 @@ public:
         TLeaderTabletInfo* tablet = Self->FindTabletEvenInDeleting(TabletId);
         if (tablet != nullptr) {
             if (msg->Status == NKikimrProto::OK
+                    || msg->Status == NKikimrProto::ALREADY
                     || msg->Status == NKikimrProto::RACE
                     || msg->Status == NKikimrProto::BLOCKED
                     || msg->Status == NKikimrProto::NO_GROUP) {
diff --git a/ydb/core/tablet/tablet_req_blockbs.cpp b/ydb/core/tablet/tablet_req_blockbs.cpp
index 7e2e94ec95..481ca9fc6c 100644
--- a/ydb/core/tablet/tablet_req_blockbs.cpp
+++ b/ydb/core/tablet/tablet_req_blockbs.cpp
@@ -5,14 +5,11 @@
 
 namespace NKikimr {
 
-constexpr ui32 MAX_ATTEMPTS = 3;
-
 class TTabletReqBlockBlobStorageGroup : public TActorBootstrapped<TTabletReqBlockBlobStorageGroup> {
     const TActorId Owner;
     const ui64 TabletId;
     const ui32 GroupId;
     const ui32 Generation;
-    ui32 ErrorCount;
 
     void ReplyAndDie(NKikimrProto::EReplyStatus status, const TString &reason = { }) {
         Send(Owner, new TEvTabletBase::TEvBlockBlobStorageResult(status, TabletId, reason));
@@ -36,17 +33,14 @@ class TTabletReqBlockBlobStorageGroup : public TActorBootstrapped<TTabletReqBloc
         switch (msg->Status) {
         case NKikimrProto::OK:
             return ReplyAndDie(NKikimrProto::OK);
+        case NKikimrProto::ALREADY:
         case NKikimrProto::BLOCKED:
         case NKikimrProto::RACE:
         case NKikimrProto::NO_GROUP:
             // The request will never succeed
             return ReplyAndDie(msg->Status, msg->ErrorReason);
         default:
-            ++ErrorCount;
-            if (ErrorCount >= MAX_ATTEMPTS) {
-                return ReplyAndDie(NKikimrProto::ERROR, msg->ErrorReason);
-            }
-            return SendRequest();
+            return ReplyAndDie(NKikimrProto::ERROR, msg->ErrorReason);
         }
     }
 
@@ -74,7 +68,6 @@ public:
         , TabletId(tabletId)
         , GroupId(groupId)
         , Generation(gen)
-        , ErrorCount(0)
     {}
 };