more detailed information about error from blob cache (#16365)

5 files changed, 16 insertions, 12 deletions

diff --git a/ydb/core/tx/columnshard/blob_cache.cpp b/ydb/core/tx/columnshard/blob_cache.cpp
index f404a25536..391f57535c 100644
--- a/ydb/core/tx/columnshard/blob_cache.cpp
+++ b/ydb/core/tx/columnshard/blob_cache.cpp
@@ -237,7 +237,7 @@ private:
         if (it != Cache.End()) {
             Hits->Inc();
             HitsBytes->Add(blobRange.Size);
-            SendResult(sender, blobRange, NKikimrProto::OK, it.Value(), ctx, true);
+            SendResult(sender, blobRange, NKikimrProto::OK, it.Value(),  {}, ctx, true);
             return true;
         }
 
@@ -423,10 +423,10 @@ private:
     }
 
     void SendResult(const TActorId& to, const TBlobRange& blobRange, NKikimrProto::EReplyStatus status,
-                    const TString& data, const TActorContext& ctx, const bool fromCache = false) {
+                    const TString& data, const TString& detailedError, const TActorContext& ctx, const bool fromCache = false) {
         LOG_S_DEBUG("Send result: " << blobRange << " to: " << to << " status: " << status);
 
-        ctx.Send(to, new TEvBlobCache::TEvReadBlobRangeResult(blobRange, status, data, fromCache));
+        ctx.Send(to, new TEvBlobCache::TEvReadBlobRangeResult(blobRange, status, data, detailedError, fromCache));
     }
 
     void Handle(TEvBlobStorage::TEvGetResult::TPtr& ev, const TActorContext& ctx) {
@@ -436,7 +436,9 @@ private:
             Y_ABORT("Unexpected reply from blobstorage");
         }
 
+        TString detailedError;
         if (ev->Get()->Status != NKikimrProto::EReplyStatus::OK) {
+            detailedError = ev->Get()->ToString();
             AFL_WARN(NKikimrServices::BLOB_CACHE)("fail", ev->Get()->ToString());
             ReadSimpleFailedBytes->Add(ev->Get()->ResponseSz);
             ReadSimpleFailedCount->Add(1);
@@ -458,14 +460,14 @@ private:
 
         for (size_t i = 0; i < ev->Get()->ResponseSz; ++i) {
             const auto& res = ev->Get()->Responses[i];
-            ProcessSingleRangeResult(blobRanges[i], readCookie, res.Status, res.Buffer.ConvertToString(), ctx);
+            ProcessSingleRangeResult(blobRanges[i], readCookie, res.Status, res.Buffer.ConvertToString(), detailedError, ctx);
         }
 
         MakeReadRequests(ctx);
     }
 
     void ProcessSingleRangeResult(const TBlobRange& blobRange, const ui64 readCookie,
-        ui32 status, const TString& data, const TActorContext& ctx) noexcept
+        ui32 status, const TString& data, const TString& detailedError, const TActorContext& ctx) noexcept
     {
         AFL_DEBUG(NKikimrServices::BLOB_CACHE)("ProcessSingleRangeResult", blobRange);
         auto readIt = OutstandingReads.find(blobRange);
@@ -500,7 +502,7 @@ private:
         AFL_DEBUG(NKikimrServices::BLOB_CACHE)("ProcessSingleRangeResult", blobRange)("send_replies", readIt->second.Waiting.size());
         // Send results to all waiters
         for (const auto& to : readIt->second.Waiting) {
-            SendResult(to, blobRange, (NKikimrProto::EReplyStatus)status, data, ctx);
+            SendResult(to, blobRange, (NKikimrProto::EReplyStatus)status, data, detailedError, ctx);
         }
 
         OutstandingReads.erase(readIt);
@@ -525,7 +527,7 @@ private:
 
             for (size_t i = 0; i < blobRanges.size(); ++i) {
                 Y_ABORT_UNLESS(blobRanges[i].BlobId.GetTabletId() == tabletId);
-                ProcessSingleRangeResult(blobRanges[i], readCookie, NKikimrProto::EReplyStatus::NOTREADY, {}, ctx);
+                ProcessSingleRangeResult(blobRanges[i], readCookie, NKikimrProto::EReplyStatus::NOTREADY, {}, {}, ctx);
             }
         }
 
diff --git a/ydb/core/tx/columnshard/blob_cache.h b/ydb/core/tx/columnshard/blob_cache.h
index 75e0ccf0a6..61aa6edcf9 100644
--- a/ydb/core/tx/columnshard/blob_cache.h
+++ b/ydb/core/tx/columnshard/blob_cache.h
@@ -72,14 +72,16 @@ struct TEvBlobCache {
         TBlobRange BlobRange;
         NKikimrProto::EReplyStatus Status;
         TString Data;
+        TString DetailedError;
         const bool FromCache = false;
         const TInstant ConstructTime = Now();
         const TString DataSourceId;
 
-        TEvReadBlobRangeResult(const TBlobRange& blobRange, NKikimrProto::EReplyStatus status, const TString& data, const bool fromCache = false, const TString& dataSourceId = Default<TString>())
+        TEvReadBlobRangeResult(const TBlobRange& blobRange, NKikimrProto::EReplyStatus status, const TString& data, const TString& detailedError, const bool fromCache = false, const TString& dataSourceId = Default<TString>())
             : BlobRange(blobRange)
             , Status(status)
             , Data(data)
+            , DetailedError(detailedError)
             , FromCache(fromCache)
             , DataSourceId(dataSourceId)
         {}
diff --git a/ydb/core/tx/columnshard/blobs_action/tier/adapter.cpp b/ydb/core/tx/columnshard/blobs_action/tier/adapter.cpp
index d0136aa983..8d42d230ee 100644
--- a/ydb/core/tx/columnshard/blobs_action/tier/adapter.cpp
+++ b/ydb/core/tx/columnshard/blobs_action/tier/adapter.cpp
@@ -16,11 +16,11 @@ std::unique_ptr<NActors::IEventBase> TRepliesAdapter::RebuildReplyEvent(std::uni
     }
     if (ev->IsSuccess()) {
         AFL_VERIFY(!!ev->Body)("key", ev->Key)("interval_from", ev->GetReadInterval().first)("interval_to", ev->GetReadInterval().second);
-        return std::make_unique<NBlobCache::TEvBlobCache::TEvReadBlobRangeResult>(bRange, NKikimrProto::EReplyStatus::OK, ev->Body, false, StorageId);
+        return std::make_unique<NBlobCache::TEvBlobCache::TEvReadBlobRangeResult>(bRange, NKikimrProto::EReplyStatus::OK, ev->Body, TString{}, false, StorageId);
     } else {
         AFL_DEBUG(NKikimrServices::TX_TIERING)("event", "s3_request_failed")("request_type", "get_object")(
             "exception", ev->GetError().GetExceptionName())("message", ev->GetError().GetMessage())("storage_id", StorageId)("blob", logoBlobId);
-        return std::make_unique<NBlobCache::TEvBlobCache::TEvReadBlobRangeResult>(bRange, NKikimrProto::EReplyStatus::ERROR, TStringBuilder() << ev->Result, false, StorageId);
+        return std::make_unique<NBlobCache::TEvBlobCache::TEvReadBlobRangeResult>(bRange, NKikimrProto::EReplyStatus::ERROR, TStringBuilder() << ev->Result, TStringBuilder{} << ev->GetError().GetExceptionName() << ", " << ev->GetError().GetMessage(), false, StorageId);
     }
 }
 
diff --git a/ydb/core/tx/columnshard/blobs_reader/actor.cpp b/ydb/core/tx/columnshard/blobs_reader/actor.cpp
index 3d27b9d8a0..2121d30893 100644
--- a/ydb/core/tx/columnshard/blobs_reader/actor.cpp
+++ b/ydb/core/tx/columnshard/blobs_reader/actor.cpp
@@ -15,7 +15,7 @@ void TActor::Handle(NBlobCache::TEvBlobCache::TEvReadBlobRangeResult::TPtr& ev)
     bool aborted = false;
     if (event.Status != NKikimrProto::EReplyStatus::OK) {
         WaitingBlobsCount.Sub(Task->GetWaitingRangesCount());
-        if (!Task->AddError(event.DataSourceId, event.BlobRange, IBlobsReadingAction::TErrorStatus::Fail(event.Status, "cannot get blob: " + event.Data.substr(0, 1024)))) {
+        if (!Task->AddError(event.DataSourceId, event.BlobRange, IBlobsReadingAction::TErrorStatus::Fail(event.Status, "cannot get blob: " + event.Data.substr(0, 1024) + ", detailed error: " + event.DetailedError))) {
             aborted = true;
         }
     } else {
diff --git a/ydb/core/tx/columnshard/blobs_reader/read_coordinator.cpp b/ydb/core/tx/columnshard/blobs_reader/read_coordinator.cpp
index df9a740d23..207b41f00e 100644
--- a/ydb/core/tx/columnshard/blobs_reader/read_coordinator.cpp
+++ b/ydb/core/tx/columnshard/blobs_reader/read_coordinator.cpp
@@ -18,7 +18,7 @@ void TReadCoordinatorActor::Handle(NBlobCache::TEvBlobCache::TEvReadBlobRangeRes
     auto tasks = BlobTasks.Extract(event.DataSourceId, event.BlobRange);
     for (auto&& i : tasks) {
         if (event.Status != NKikimrProto::EReplyStatus::OK) {
-            i->AddError(event.DataSourceId, event.BlobRange, IBlobsReadingAction::TErrorStatus::Fail(event.Status, "cannot get blob"));
+            i->AddError(event.DataSourceId, event.BlobRange, IBlobsReadingAction::TErrorStatus::Fail(event.Status, "cannot get blob, detailed error: " + event.DetailedError));
         } else {
             i->AddData(event.DataSourceId, event.BlobRange, event.Data);
         }