Fix error logs,

Add markers for logs in tablet_req_rebuildhistory.cpp

5 files changed, 31 insertions, 27 deletions

diff --git a/ydb/core/blobstorage/dsproxy/dsproxy_get.cpp b/ydb/core/blobstorage/dsproxy/dsproxy_get.cpp
index e4043c36754..0b89004dd24 100644
--- a/ydb/core/blobstorage/dsproxy/dsproxy_get.cpp
+++ b/ydb/core/blobstorage/dsproxy/dsproxy_get.cpp
@@ -331,7 +331,7 @@ class TBlobStorageGroupGetRequest : public TBlobStorageGroupRequestActor<TBlobSt
                 TStringStream errorReason;
                 errorReason << "Got VMultiPutResult itemStatus# " << itemStatus << " from VDiskId# " << vDiskId;
                 ErrorReason = errorReason.Str();
-                GetImpl.PrepareReply(itemStatus, LogCtx, outGetResult);
+                GetImpl.PrepareReply(itemStatus, ErrorReason, LogCtx, outGetResult);
                 return true;
             } else {
                 R_LOG_DEBUG_S("BPG27", "Handle TEvVMultiPutResultItem"
@@ -543,8 +543,7 @@ public:
     friend class TBlobStorageGroupRequestActor<TBlobStorageGroupGetRequest>;
     void ReplyAndDie(NKikimrProto::EReplyStatus status) {
         TAutoPtr<TEvBlobStorage::TEvGetResult> getResult;
-        GetImpl.PrepareReply(status, LogCtx, getResult);
-        getResult->ErrorReason = ErrorReason;
+        GetImpl.PrepareReply(status, ErrorReason, LogCtx, getResult);
         SendReplyAndDie(getResult);
         return;
     }
diff --git a/ydb/core/blobstorage/dsproxy/dsproxy_get_impl.cpp b/ydb/core/blobstorage/dsproxy/dsproxy_get_impl.cpp
index f1d01776b6c..9729b909811 100644
--- a/ydb/core/blobstorage/dsproxy/dsproxy_get_impl.cpp
+++ b/ydb/core/blobstorage/dsproxy/dsproxy_get_impl.cpp
@@ -16,11 +16,12 @@
 
 namespace NKikimr {
 
-void TGetImpl::PrepareReply(NKikimrProto::EReplyStatus status, TLogContext &logCtx,
+void TGetImpl::PrepareReply(NKikimrProto::EReplyStatus status, TString errorReason, TLogContext &logCtx,
         TAutoPtr<TEvBlobStorage::TEvGetResult> &outGetResult) {
     outGetResult.Reset(new TEvBlobStorage::TEvGetResult(status, QuerySize, Info->GroupID));
     ReplyBytes = 0;
     outGetResult->BlockedGeneration = BlockedGeneration;
+    outGetResult->ErrorReason = errorReason;
 
     if (status != NKikimrProto::OK) {
         for (ui32 i = 0, e = QuerySize; i != e; ++i) {
diff --git a/ydb/core/blobstorage/dsproxy/dsproxy_get_impl.h b/ydb/core/blobstorage/dsproxy/dsproxy_get_impl.h
index 815ec33c285..146c0270c37 100644
--- a/ydb/core/blobstorage/dsproxy/dsproxy_get_impl.h
+++ b/ydb/core/blobstorage/dsproxy/dsproxy_get_impl.h
@@ -248,7 +248,7 @@ public:
             TDeque<std::unique_ptr<TEvBlobStorage::TEvVMultiPut>> &outVMultiPuts,
             TAutoPtr<TEvBlobStorage::TEvGetResult> &outGetResult);
 
-    void PrepareReply(NKikimrProto::EReplyStatus status, TLogContext &logCtx,
+    void PrepareReply(NKikimrProto::EReplyStatus status, TString errorReason, TLogContext &logCtx,
             TAutoPtr<TEvBlobStorage::TEvGetResult> &outGetResult);
 
     template <typename TVPutEvent>
@@ -301,12 +301,11 @@ protected:
             }
 
             case EStrategyOutcome::ERROR:
-                PrepareReply(NKikimrProto::ERROR, logCtx, outGetResult);
-                outGetResult->ErrorReason = outcome.ErrorReason;
+                PrepareReply(NKikimrProto::ERROR, outcome.ErrorReason, logCtx, outGetResult);
                 return false;
 
             case EStrategyOutcome::DONE:
-                PrepareReply(NKikimrProto::OK, logCtx, outGetResult);
+                PrepareReply(NKikimrProto::OK, "", logCtx, outGetResult);
                 return false;
         }
     }
diff --git a/ydb/core/blobstorage/dsproxy/dsproxy_range.cpp b/ydb/core/blobstorage/dsproxy/dsproxy_range.cpp
index 319e7e11ca3..299dfa2af7d 100644
--- a/ydb/core/blobstorage/dsproxy/dsproxy_range.cpp
+++ b/ydb/core/blobstorage/dsproxy/dsproxy_range.cpp
@@ -95,6 +95,7 @@ class TBlobStorageGroupRangeRequest : public TBlobStorageGroupRequestActor<TBlob
             case NKikimrProto::OUT_OF_SPACE:
                 FailedDisks |= TBlobStorageGroupInfo::TGroupVDisks(&Info->GetTopology(), vdisk);
                 if (!Info->GetQuorumChecker().CheckFailModelForGroup(FailedDisks)) {
+                    ErrorReason = "Failed disks check fails on non-OK event status";
                     return ReplyAndDie(NKikimrProto::ERROR);
                 }
                 break;
@@ -106,6 +107,7 @@ class TBlobStorageGroupRangeRequest : public TBlobStorageGroupRequestActor<TBlob
                             " TEvVGetResult# " << ev->Get()->ToString());
                     FailedDisks |= TBlobStorageGroupInfo::TGroupVDisks(&Info->GetTopology(), vdisk);
                     if (!Info->GetQuorumChecker().CheckFailModelForGroup(FailedDisks)) {
+                        ErrorReason = "Failed disks check fails on OK event status";
                         return ReplyAndDie(NKikimrProto::ERROR);
                     }
                 }
@@ -193,6 +195,7 @@ class TBlobStorageGroupRangeRequest : public TBlobStorageGroupRequestActor<TBlob
                 switch (tracker.GetBlobState(Info.Get(), &lostByIngress)) {
                     case TBlobStorageGroupInfo::EBS_DISINTEGRATED:
                         R_LOG_ERROR_S("DSR02", "disintegrated");
+                        ErrorReason = "BS disintegrated";
                         return ReplyAndDie(NKikimrProto::ERROR);
 
                     case TBlobStorageGroupInfo::EBS_UNRECOVERABLE_FRAGMENTARY:
@@ -269,6 +272,7 @@ class TBlobStorageGroupRangeRequest : public TBlobStorageGroupRequestActor<TBlob
         NKikimrProto::EReplyStatus status = getResult.Status;
         if (status != NKikimrProto::OK) {
             R_LOG_ERROR_S("DSR03", "Handle TEvGetResult status# " << NKikimrProto::EReplyStatus_Name(status).data());
+            ErrorReason = getResult.ErrorReason;
             ReplyAndDie(status);
             return;
         }
@@ -289,6 +293,7 @@ class TBlobStorageGroupRangeRequest : public TBlobStorageGroupRequestActor<TBlob
                     << " Response[" << i << "]# " << NKikimrProto::EReplyStatus_Name(response.Status)
                     << " BlobsToGet# " << DumpBlobsToGet()
                     << " TEvGetResult# " << ev->Get()->ToString());
+                ErrorReason = getResult.ErrorReason;
                 ReplyAndDie(NKikimrProto::ERROR);
                 return;
             }
diff --git a/ydb/core/tablet/tablet_req_rebuildhistory.cpp b/ydb/core/tablet/tablet_req_rebuildhistory.cpp
index c12f6c93ad4..6c2dd2ce46e 100644
--- a/ydb/core/tablet/tablet_req_rebuildhistory.cpp
+++ b/ydb/core/tablet/tablet_req_rebuildhistory.cpp
@@ -16,11 +16,11 @@
 #error log macro definition clash
 #endif
 
-#define BLOG_D(stream) LOG_DEBUG_S(*TlsActivationContext, NKikimrServices::TABLET_MAIN, "TabletId# " << Info->TabletID << (FollowerCookie ? "f " : " ") << stream)
-#define BLOG_W(stream) LOG_WARN_S(*TlsActivationContext, NKikimrServices::TABLET_MAIN, "TabletId# " << Info->TabletID << (FollowerCookie ? "f " : " ") << stream)
-#define BLOG_ERROR(stream) LOG_ERROR_S(*TlsActivationContext, NKikimrServices::TABLET_MAIN, "TabletId# " << Info->TabletID << (FollowerCookie ? "f " : " ") << stream)
-#define BLOG_TRACE(stream) LOG_TRACE_S(*TlsActivationContext, NKikimrServices::TABLET_MAIN, "TabletId# " << Info->TabletID << (FollowerCookie ? "f " : " ") << stream)
-#define BLOG_CRIT(stream) LOG_CRIT_S(*TlsActivationContext, NKikimrServices::TABLET_MAIN, "TabletId# " << Info->TabletID << (FollowerCookie ? "f " : " ") << stream)
+#define BLOG_D(stream, marker) LOG_DEBUG_S(*TlsActivationContext, NKikimrServices::TABLET_MAIN, "TabletId# " << Info->TabletID << (FollowerCookie ? "f " : " ") << stream << " Marker# " << marker)
+#define BLOG_W(stream, marker) LOG_WARN_S(*TlsActivationContext, NKikimrServices::TABLET_MAIN, "TabletId# " << Info->TabletID << (FollowerCookie ? "f " : " ") << stream << " Marker# " << marker)
+#define BLOG_ERROR(stream, marker) LOG_ERROR_S(*TlsActivationContext, NKikimrServices::TABLET_MAIN, "TabletId# " << Info->TabletID << (FollowerCookie ? "f " : " ") << stream << " Marker# " << marker)
+#define BLOG_TRACE(stream, marker) LOG_TRACE_S(*TlsActivationContext, NKikimrServices::TABLET_MAIN, "TabletId# " << Info->TabletID << (FollowerCookie ? "f " : " ") << stream << " Marker# " << marker)
+#define BLOG_CRIT(stream, marker) LOG_CRIT_S(*TlsActivationContext, NKikimrServices::TABLET_MAIN, "TabletId# " << Info->TabletID << (FollowerCookie ? "f " : " ") << stream << " Marker# " << marker)
 
 namespace NKikimr {
 
@@ -228,7 +228,7 @@ class TTabletReqRebuildHistoryGraph : public TActorBootstrapped<TTabletReqRebuil
 
     void ProcessZeroEntry(ui32 gen, NKikimrTabletBase::TTabletLogEntry &logEntry) {
 
-        BLOG_D("TTabletReqRebuildHistoryGraph::ProcessZeroEntry - generation " << gen);
+        BLOG_D("TTabletReqRebuildHistoryGraph::ProcessZeroEntry - generation " << gen, "TRRH01");
         if (IntrospectionTrace) {
             IntrospectionTrace->Attach(MakeHolder<NTracing::TOnProcessZeroEntry>(gen, Snapshot, Confirmed));
         }
@@ -262,7 +262,7 @@ class TTabletReqRebuildHistoryGraph : public TActorBootstrapped<TTabletReqRebuil
             const ui32 prevGeneration = confirmed.first;
 
             if (prevGeneration < Snapshot.first) {
-                BLOG_CRIT("snapshot overrun in gen " << gen << " zero entry, declared prev gen " << prevGeneration << " while known snapshot is " << Snapshot.first << ":" << Snapshot.second);
+                BLOG_CRIT("snapshot overrun in gen " << gen << " zero entry, declared prev gen " << prevGeneration << " while known snapshot is " << Snapshot.first << ":" << Snapshot.second, "TRRH02");
 
                 if (IntrospectionTrace)
                     IntrospectionTrace->Attach(MakeHolder<NTracing::TErrorRebuildGraph>(gen, 0));
@@ -403,7 +403,7 @@ class TTabletReqRebuildHistoryGraph : public TActorBootstrapped<TTabletReqRebuil
     void ProcessKeyEntry(const TLogoBlobID &id, const TString &logBody) {
         NKikimrTabletBase::TTabletLogEntry logEntry;
         if (!logEntry.ParseFromString(logBody)) {
-            BLOG_ERROR("TTabletReqRebuildHistoryGraph::ProcessKeyEntry logBody ParseFromString error, id# " << id);
+            BLOG_ERROR("TTabletReqRebuildHistoryGraph::ProcessKeyEntry logBody ParseFromString error, id# " << id, "TRRH03");
             if (IntrospectionTrace) {
                 IntrospectionTrace->Attach(MakeHolder<NTracing::TErrorParsingFromString>(id));
             }
@@ -415,7 +415,7 @@ class TTabletReqRebuildHistoryGraph : public TActorBootstrapped<TTabletReqRebuil
 
         BLOG_D("TTabletReqRebuildHistoryGraph::ProcessKeyEntry, LastBlobID: " << id.ToString()
             << " Snap: " << Snapshot.first << ":" << Snapshot.second
-            << " for " << Info->TabletID);
+            << " for " << Info->TabletID, "TRRH04");
 
         const bool isZeroStep = id.Step() == 0;
         const bool isSynthEntry = id.Cookie() != 0;
@@ -467,7 +467,7 @@ class TTabletReqRebuildHistoryGraph : public TActorBootstrapped<TTabletReqRebuil
 
             NKikimrTabletBase::TTabletLogEntry logEntry;
             if (!logEntry.ParseFromString(it->Buffer)) {
-                BLOG_ERROR("TTabletReqRebuildHistoryGraph::ApplyDiscoveryRange it->Buffer ParseFromString error, id# " << id);
+                BLOG_ERROR("TTabletReqRebuildHistoryGraph::ApplyDiscoveryRange it->Buffer ParseFromString error, id# " << id, "TRRH05");
                 return ReplyAndDie(NKikimrProto::ERROR, "Log entry parse failed");
             }
 
@@ -494,7 +494,7 @@ class TTabletReqRebuildHistoryGraph : public TActorBootstrapped<TTabletReqRebuil
 
         for (auto &xpair : RefsToCheckByGroup) {
             if (!SendRefsCheck(xpair.second, xpair.first)) {
-                BLOG_ERROR("TTabletReqRebuildHistoryGraph::MakeHistory SendRefsCheck A error");
+                BLOG_ERROR("TTabletReqRebuildHistoryGraph::MakeHistory SendRefsCheck A error", "TRRH06");
                 if (IntrospectionTrace) {
                     IntrospectionTrace->Attach(MakeHolder<NTracing::TErrorSendRefsCheck>());
                 }
@@ -590,11 +590,11 @@ class TTabletReqRebuildHistoryGraph : public TActorBootstrapped<TTabletReqRebuil
                 GroupReadOps[std::make_pair(response.Id.Channel(), msg->GroupId)] += 1;
                 break;
             case NKikimrProto::NODATA:
-                BLOG_W("TTabletReqRebuildHistoryGraph::CheckReferences - NODATA for blob " << response.Id);
+                BLOG_W("TTabletReqRebuildHistoryGraph::CheckReferences - NODATA for blob " << response.Id, "TRRH07");
                 break; // must left as unchecked
             default:
                 BLOG_ERROR("TTabletReqRebuildHistoryGraph::CheckReferences - blob " << response.Id
-                            << " Status# " << NKikimrProto::EReplyStatus_Name(response.Status));
+                            << " Status# " << NKikimrProto::EReplyStatus_Name(response.Status), "TRRH08");
                 if (IntrospectionTrace) {
                     IntrospectionTrace->Attach(MakeHolder<NTracing::TErrorUnknownStatus>(response.Status, msg->ErrorReason));
                 }
@@ -619,7 +619,7 @@ class TTabletReqRebuildHistoryGraph : public TActorBootstrapped<TTabletReqRebuil
             bool hasSnapshotInGeneration = (generation == 0);
 
             BLOG_D("TTabletReqRebuildHistoryGraph::BuildHistory - Process generation " << generation
-                << " from " << (ui32)gx.Base << " with " << gx.Body.size() << " steps");
+                << " from " << (ui32)gx.Base << " with " << gx.Body.size() << " steps", "TRRH09");
 
             for (ui32 i = 0, e = (ui32)gx.Body.size(); i != e; ++i) {
                 const ui32 step = gx.Base + i;
@@ -700,7 +700,7 @@ class TTabletReqRebuildHistoryGraph : public TActorBootstrapped<TTabletReqRebuil
                                     hasSnapshotInGeneration = true;
                                 }
                             } else {
-                                BLOG_D("TTabletReqRebuildHistoryGraph::BuildHistory - THE TAIL - miss " << id.first << ":" << id.second);
+                                BLOG_D("TTabletReqRebuildHistoryGraph::BuildHistory - THE TAIL - miss " << id.first << ":" << id.second, "TRRH10");
                             }
                         }
                         break;
@@ -801,7 +801,7 @@ class TTabletReqRebuildHistoryGraph : public TActorBootstrapped<TTabletReqRebuil
         case NKikimrProto::OK:
             if (FollowerCookie == 0 && msg->Latest.Generation() > BlockedGen) {
                 BLOG_ERROR("TTabletReqRebuildHistoryGraph - Found entry beyond blocked generation"
-                    << " LastBlobID: " << msg->Latest.ToString() << ". Blocked: " << BlockedGen);
+                    << " LastBlobID: " << msg->Latest.ToString() << ". Blocked: " << BlockedGen, "TRRH11");
                 if (IntrospectionTrace) {
                     IntrospectionTrace->Attach(MakeHolder<NTracing::TErrorEntryBeyondBlocked>(msg->Latest, BlockedGen));
                 }
@@ -818,7 +818,7 @@ class TTabletReqRebuildHistoryGraph : public TActorBootstrapped<TTabletReqRebuil
             return ReplyAndDie(msg->Status, msg->ErrorReason); // valid condition, nothing known in blob-storage
         default:
             BLOG_ERROR("TTabletReqRebuildHistoryGraph::Handle TEvFindLatestLogEntryResult"
-                << " Status# " << NKikimrProto::EReplyStatus_Name(msg->Status));
+                << " Status# " << NKikimrProto::EReplyStatus_Name(msg->Status), "TRRH12");
             if (IntrospectionTrace) {
                 IntrospectionTrace->Attach(MakeHolder<NTracing::TErrorUnknownStatus>(msg->Status, msg->ErrorReason));
             }
@@ -838,7 +838,7 @@ class TTabletReqRebuildHistoryGraph : public TActorBootstrapped<TTabletReqRebuil
         default:
             BLOG_ERROR("TTabletReqRebuildHistoryGraph::HandleDiscover TEvRangeResult"
                 << " Status# " << NKikimrProto::EReplyStatus_Name(msg->Status)
-                << " Result# " << msg->Print(false));
+                << " Result# " << msg->Print(false), "TRRH13");
             if (IntrospectionTrace) {
                 IntrospectionTrace->Attach(MakeHolder<NTracing::TErrorUnknownStatus>(msg->Status, msg->ErrorReason));
             }
@@ -862,7 +862,7 @@ class TTabletReqRebuildHistoryGraph : public TActorBootstrapped<TTabletReqRebuil
         default:
             BLOG_ERROR("TTabletReqRebuildHistoryGraph::Handle TEvGetResult"
                 << " Status# " << NKikimrProto::EReplyStatus_Name(msg->Status)
-                << " Result# " << msg->Print(false));
+                << " Result# " << msg->Print(false), "TRRH14");
             if (IntrospectionTrace) {
                 IntrospectionTrace->Attach(MakeHolder<NTracing::TErrorUnknownStatus>(msg->Status, msg->ErrorReason));
             }