add checksum mismatch sensor per volume (#1612) (#1616)

ydb-platform · Jul 16, 2024 · 97bc0cf · 97bc0cf
1 parent 40030ab
commit 97bc0cf
Show file tree

Hide file tree

Showing 5 changed files with 12 additions and 0 deletions.
diff --git a/cloud/blockstore/libs/storage/core/disk_counters.h b/cloud/blockstore/libs/storage/core/disk_counters.h
@@ -203,6 +203,10 @@ struct TSimpleDiskCounters
         EPublishingPolicy::DiskRegistryBased,
         TSimpleCounter::ECounterType::Generic,
         ECounterExpirationPolicy::Permanent};
+    TCounter ChecksumMismatches{
+        EPublishingPolicy::DiskRegistryBased,
+        TSimpleCounter::ECounterType::Generic,
+        ECounterExpirationPolicy::Permanent};
 
     static constexpr TMeta AllCounters[] = {
         MakeMeta<&TSimpleDiskCounters::BytesCount>(),
@@ -233,6 +237,7 @@ struct TSimpleDiskCounters
         MakeMeta<&TSimpleDiskCounters::HasBrokenDevice>(),
         MakeMeta<&TSimpleDiskCounters::HasBrokenDeviceSilent>(),
         MakeMeta<&TSimpleDiskCounters::ScrubbingProgress>(),
+        MakeMeta<&TSimpleDiskCounters::ChecksumMismatches>(),
     };
 };
 static_assert(

diff --git a/cloud/blockstore/libs/storage/partition_nonrepl/part_mirror_actor.cpp b/cloud/blockstore/libs/storage/partition_nonrepl/part_mirror_actor.cpp
@@ -204,6 +204,7 @@ void TMirrorPartitionActor::CompareChecksums(const TActorContext& ctx)
                 DescribeRange(GetScrubbingRange()).c_str(),
                 checksums[i]);
         }
+        ++ChecksumMismatches;
         ReportMirroredDiskChecksumMismatch();
 
         if (Config->GetResyncRangeAfterScrubbing()) {

diff --git a/cloud/blockstore/libs/storage/partition_nonrepl/part_mirror_actor.h b/cloud/blockstore/libs/storage/partition_nonrepl/part_mirror_actor.h
@@ -77,6 +77,7 @@ class TMirrorPartitionActor final
     ui64 ScrubbingThroughput = 0;
     TInstant ScrubbingRangeStarted;
     bool ResyncRangeStarted = false;
+    ui32 ChecksumMismatches = 0;
 
 public:
     TMirrorPartitionActor(

diff --git a/cloud/blockstore/libs/storage/partition_nonrepl/part_mirror_actor_stats.cpp b/cloud/blockstore/libs/storage/partition_nonrepl/part_mirror_actor_stats.cpp
@@ -62,6 +62,7 @@ void TMirrorPartitionActor::SendStats(const TActorContext& ctx)
         }
     }
 
+    stats->Simple.ChecksumMismatches.Value = ChecksumMismatches;
     stats->Simple.ScrubbingProgress.Value =
         100 * GetScrubbingRange().Start / State.GetBlockCount();
     stats->Cumulative.ScrubbingThroughput.Value = ScrubbingThroughput;

diff --git a/cloud/blockstore/libs/storage/partition_nonrepl/part_mirror_ut.cpp b/cloud/blockstore/libs/storage/partition_nonrepl/part_mirror_ut.cpp
@@ -1184,6 +1184,7 @@ Y_UNIT_TEST_SUITE(TMirrorPartitionTest)
             2 * 4_MB,
             counters.Cumulative.ScrubbingThroughput.Value);
         UNIT_ASSERT_VALUES_EQUAL(33, counters.Simple.ScrubbingProgress.Value);
+        UNIT_ASSERT_VALUES_EQUAL(0, counters.Simple.ChecksumMismatches.Value);
     }
 
     Y_UNIT_TEST(ShouldFindChecksumMismatch)
@@ -1234,6 +1235,7 @@ Y_UNIT_TEST_SUITE(TMirrorPartitionTest)
             true);
 
         UNIT_ASSERT_VALUES_EQUAL(2, mirroredDiskChecksumMismatch->Val());
+        UNIT_ASSERT_VALUES_EQUAL(2, counters.Simple.ChecksumMismatches.Value);
 
         const auto range3 = TBlockRange64::WithLength(1025, 50);
         env.WriteMirror(range3, 'A');
@@ -1250,6 +1252,7 @@ Y_UNIT_TEST_SUITE(TMirrorPartitionTest)
             runtime.AdvanceCurrentTime(UpdateCountersInterval);
             runtime.DispatchEvents({}, TDuration::MilliSeconds(50));
         }
+        UNIT_ASSERT_VALUES_EQUAL(3, counters.Simple.ChecksumMismatches.Value);
         UNIT_ASSERT_VALUES_EQUAL(3, mirroredDiskChecksumMismatch->Val());
 
         // check that all ranges was resynced and there is no more mismatches
@@ -1264,6 +1267,7 @@ Y_UNIT_TEST_SUITE(TMirrorPartitionTest)
             runtime.AdvanceCurrentTime(UpdateCountersInterval);
             runtime.DispatchEvents({}, TDuration::MilliSeconds(50));
         }
+        UNIT_ASSERT_VALUES_EQUAL(3, counters.Simple.ChecksumMismatches.Value);
         UNIT_ASSERT_VALUES_EQUAL(3, mirroredDiskChecksumMismatch->Val());
     }