Add Debug Info to BufferWithGaps,

Add DebugInfo string to BufferWithGaps

7 files changed, 48 insertions, 6 deletions

diff --git a/ydb/core/blobstorage/base/bufferwithgaps.h b/ydb/core/blobstorage/base/bufferwithgaps.h
index 28fc2f78ed9..506a2c7f641 100644
--- a/ydb/core/blobstorage/base/bufferwithgaps.h
+++ b/ydb/core/blobstorage/base/bufferwithgaps.h
@@ -6,6 +6,7 @@
 #include <util/generic/map.h>
 #include <util/generic/vector.h>
 #include <util/generic/algorithm.h>
+#include <ydb/core/util/yverify_stream.h>
 
 namespace NKikimr {
 
@@ -25,6 +26,8 @@ namespace NKikimr {
         ui32 Offset; // Data's offset in Gaps space
         bool IsCommited;
 
+        std::function<TString()> DebugInfoGenerator;
+
     public:
         TBufferWithGaps()
             : Offset(0)
@@ -66,20 +69,21 @@ namespace NKikimr {
         }
 
         TString ToString() const {
-            Y_VERIFY(IsReadable(), "returned data is corrupt (or was never written) and therefore could not be used safely");
+            Y_VERIFY_S(IsReadable(), "returned data is corrupt (or was never written) and therefore could not be used safely, State# "
+                    << PrintState());
             return Data;
         }
 
         TString Substr(ui32 offset, ui32 len) const {
-            Y_VERIFY(IsReadable(offset, len), "returned data is corrupt (or was never written) at offset# %" PRIu32
-                   " len# %" PRIu32 " and therefore could not be used safely", offset, len);
+            Y_VERIFY_S(IsReadable(offset, len), "returned data is corrupt (or was never written) at offset# %" << offset
+                   << " len# " << len << " and therefore could not be used safely, State# " << PrintState());
             return Data.substr(offset, len);
         }
 
         template<typename T>
         const T *DataPtr(ui32 offset, ui32 len = sizeof(T)) const {
-            Y_VERIFY(IsReadable(offset, len), "returned data is corrupt (or was never written) at offset# %" PRIu32
-                   " len# %" PRIu32 " and therefore could not be used safely", offset, len);
+            Y_VERIFY_S(IsReadable(offset, len), "returned data is corrupt (or was never written) at offset# " << offset
+                   << " len# " << len << " and therefore could not be used safely, State# " << PrintState());
             return reinterpret_cast<T *>(Data.data() + offset);
         }
 
@@ -173,6 +177,29 @@ namespace NKikimr {
                 }
             }
         }
+
+        void SetDebugInfoGenerator(const std::function<TString()>& debugInfoGenerator) {
+            DebugInfoGenerator = debugInfoGenerator;
+        }
+
+        void SetDebugInfoGenerator(std::function<TString()>&& debugInfoGenerator) {
+            DebugInfoGenerator = std::move(debugInfoGenerator);
+        }
+
+        TString PrintState() const {
+            TStringStream str;
+            str << "Offset# " << Offset;
+            str << " Gaps# { ";
+            for (auto [begin, size] : Gaps) {
+                str << "{ Begin# " << begin << " Size# " << size << " } "; 
+            }
+            str << "}";
+            str << " IsCommitted# " << IsCommited;
+            if (DebugInfoGenerator) {
+                str << " DebugInfo# " << DebugInfoGenerator();
+            }
+            return str.Str();
+        }
     };
 
 } // NKikimr
diff --git a/ydb/core/blobstorage/pdisk/blobstorage_pdisk_actor.cpp b/ydb/core/blobstorage/pdisk/blobstorage_pdisk_actor.cpp
index b1dd7ac5954..ec48d98fccf 100644
--- a/ydb/core/blobstorage/pdisk/blobstorage_pdisk_actor.cpp
+++ b/ydb/core/blobstorage/pdisk/blobstorage_pdisk_actor.cpp
@@ -682,6 +682,8 @@ public:
         PDisk->Mon.GetReadCounter(evChunkRead.PriorityClass)->CountRequest(0);
         THolder<NPDisk::TEvChunkReadResult> result = MakeHolder<NPDisk::TEvChunkReadResult>(NKikimrProto::CORRUPTED,
             evChunkRead.ChunkIdx, evChunkRead.Offset, evChunkRead.Cookie, 0, "PDisk is in error state");
+        result->Data.SetDebugInfoGenerator(PDisk->DebugInfoGenerator);
+
         LOG_DEBUG(*TlsActivationContext, NKikimrServices::BS_PDISK, "PDiskId# %" PRIu32 " %s To: %" PRIu64 " Marker# BSY02",
             (ui32)PDisk->PDiskId, result->ToString().c_str(), (ui64)ev->Sender.LocalId());
         Send(ev->Sender, result.Release());
@@ -801,6 +803,7 @@ public:
     void Handle(NPDisk::TEvChunkRead::TPtr &ev) {
         double burstMs;
         TChunkRead* request = PDisk->ReqCreator.CreateChunkRead(*ev->Get(), ev->Sender, burstMs, std::move(ev->TraceId));
+        request->DebugInfoGenerator = PDisk->DebugInfoGenerator;
         CheckBurst(request->IsSensitive, burstMs);
         PDisk->InputRequest(request);
     }
diff --git a/ydb/core/blobstorage/pdisk/blobstorage_pdisk_completion_impl.cpp b/ydb/core/blobstorage/pdisk/blobstorage_pdisk_completion_impl.cpp
index f0ecc64539d..e1c10e70acd 100644
--- a/ydb/core/blobstorage/pdisk/blobstorage_pdisk_completion_impl.cpp
+++ b/ydb/core/blobstorage/pdisk/blobstorage_pdisk_completion_impl.cpp
@@ -327,6 +327,9 @@ void TCompletionChunkRead::ReplyError(TActorSystem *actorSystem, TString reason)
     auto result = MakeHolder<TEvChunkReadResult>(NKikimrProto::CORRUPTED,
             Read->ChunkIdx, Read->Offset, Read->Cookie,
             PDisk->GetStatusFlags(Read->Owner, Read->OwnerGroupType), error.Str());
+
+    result->Data.SetDebugInfoGenerator(PDisk->DebugInfoGenerator);
+
     LOG_WARN_S(*actorSystem, NKikimrServices::BS_PDISK, error.Str());
     actorSystem->Send(Read->Sender, result.Release());
     Read->IsReplied = true;
diff --git a/ydb/core/blobstorage/pdisk/blobstorage_pdisk_impl.cpp b/ydb/core/blobstorage/pdisk/blobstorage_pdisk_impl.cpp
index 7e804978e31..e4bcef03b0f 100644
--- a/ydb/core/blobstorage/pdisk/blobstorage_pdisk_impl.cpp
+++ b/ydb/core/blobstorage/pdisk/blobstorage_pdisk_impl.cpp
@@ -95,6 +95,10 @@ TPDisk::TPDisk(const TIntrusivePtr<TPDiskConfig> cfg, const TIntrusivePtr<::NMon
     JointLogWrites.reserve(16 << 10);
     JointCommits.reserve(16 << 10);
     JointChunkForgets.reserve(16 << 10);
+
+    DebugInfoGenerator = [id = PDiskId, type = PDiskCategory]() {
+        return TStringBuilder() << "PDisk DebugInfo# { Id# " << id << " Type# " << type.TypeStrLong() << " }";
+    };
 }
 
 TString TPDisk::DynamicStateToString(bool isMultiline) {
@@ -3595,7 +3599,6 @@ void TPDisk::AddCbsSet(ui32 ownerId) {
     SchedulerConfigure(conf);
 }
 
-
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
 // External interface
 ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
diff --git a/ydb/core/blobstorage/pdisk/blobstorage_pdisk_impl.h b/ydb/core/blobstorage/pdisk/blobstorage_pdisk_impl.h
index e2c12017f39..3e80f29a484 100644
--- a/ydb/core/blobstorage/pdisk/blobstorage_pdisk_impl.h
+++ b/ydb/core/blobstorage/pdisk/blobstorage_pdisk_impl.h
@@ -185,6 +185,9 @@ public:
     // Chunk locking
     TMap<TOwner, ui32> OwnerLocks;
 
+    // Debug
+    std::function<TString()> DebugInfoGenerator;
+
     ////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
     // Initialization
     TPDisk(const TIntrusivePtr<TPDiskConfig> cfg, const TIntrusivePtr<::NMonitoring::TDynamicCounters>& counters);
diff --git a/ydb/core/blobstorage/pdisk/blobstorage_pdisk_requestimpl.cpp b/ydb/core/blobstorage/pdisk/blobstorage_pdisk_requestimpl.cpp
index 27f26b1f4d7..cc476b18213 100644
--- a/ydb/core/blobstorage/pdisk/blobstorage_pdisk_requestimpl.cpp
+++ b/ydb/core/blobstorage/pdisk/blobstorage_pdisk_requestimpl.cpp
@@ -52,6 +52,7 @@ void TChunkRead::Abort(TActorSystem* actorSystem) {
             <NPDisk::TEvChunkReadResult>(NKikimrProto::ERROR,
                     ChunkIdx, Offset, Cookie,
                     NKikimrBlobStorage::StatusIsValid, error.Str());
+        result->Data.SetDebugInfoGenerator(std::move(DebugInfoGenerator));
         actorSystem->Send(Sender, result.Release());
         IsReplied = true;
     }
diff --git a/ydb/core/blobstorage/pdisk/blobstorage_pdisk_requestimpl.h b/ydb/core/blobstorage/pdisk/blobstorage_pdisk_requestimpl.h
index a819fae0d80..60208432cb5 100644
--- a/ydb/core/blobstorage/pdisk/blobstorage_pdisk_requestimpl.h
+++ b/ydb/core/blobstorage/pdisk/blobstorage_pdisk_requestimpl.h
@@ -366,6 +366,8 @@ public:
 
     const ui64 DoubleFreeCanary;
 
+    std::function<TString()> DebugInfoGenerator;
+
     TChunkRead(const NPDisk::TEvChunkRead &ev, const TActorId &sender, TReqId reqId, NWilson::TTraceId traceId)
         : TRequestBase(sender, reqId, ev.Owner, ev.OwnerRound, ev.PriorityClass, std::move(traceId))
         , ChunkIdx(ev.ChunkIdx)