From 8588d44e45dd241e86b36d1b015644a7646dd7c5 Mon Sep 17 00:00:00 2001
From: yegorskii <137099343+yegorskii@users.noreply.github.com>
Date: Sat, 3 Aug 2024 14:05:20 +0200
Subject: [PATCH] collect barrier gen should match current current tablet
 generation (#1715)

* collect barrier gen should match current current tablet generation

* update

* fix failed uts because of shifted LastStep

---------

Co-authored-by: yegorskii <yegorskii@yandex-team.ru>
---
 .../libs/storage/tablet/tablet_state.cpp      |  4 +-
 .../libs/storage/tablet/tablet_ut_data.cpp    | 81 ++++++++++++++++++-
 2 files changed, 81 insertions(+), 4 deletions(-)

diff --git a/cloud/filestore/libs/storage/tablet/tablet_state.cpp b/cloud/filestore/libs/storage/tablet/tablet_state.cpp
index d29ed36f1b6..ac39a29cc63 100644
--- a/cloud/filestore/libs/storage/tablet/tablet_state.cpp
+++ b/cloud/filestore/libs/storage/tablet/tablet_state.cpp
@@ -84,7 +84,9 @@ void TIndexTabletState::LoadState(
     const TThrottlerConfig& throttlerConfig)
 {
     Generation = generation;
-    LastStep = 0;
+    // https://github.com/ydb-platform/nbs/issues/1714
+    // because of possible race in vdisks we should not start with 0
+    LastStep = 1;
     LastCollectCounter = 0;
 
     TruncateBlocksThreshold = config.GetMaxBlocksPerTruncateTx();
diff --git a/cloud/filestore/libs/storage/tablet/tablet_ut_data.cpp b/cloud/filestore/libs/storage/tablet/tablet_ut_data.cpp
index c6a55cc4270..cd1ba6543b7 100644
--- a/cloud/filestore/libs/storage/tablet/tablet_ut_data.cpp
+++ b/cloud/filestore/libs/storage/tablet/tablet_ut_data.cpp
@@ -4032,7 +4032,7 @@ Y_UNIT_TEST_SUITE(TIndexTabletTest_Data)
             NKikimr::LogoBlobIDFromLogoBlobID(blobPieces[0].GetBlobId());
         UNIT_ASSERT_VALUES_EQUAL(tabletId, blobId0.TabletID());
         UNIT_ASSERT_VALUES_EQUAL(2, blobId0.Generation());
-        UNIT_ASSERT_VALUES_EQUAL(3, blobId0.Step());
+        UNIT_ASSERT_VALUES_EQUAL(4, blobId0.Step());
         UNIT_ASSERT_VALUES_EQUAL(3, blobId0.Channel());
         UNIT_ASSERT_VALUES_EQUAL(0, blobId0.Cookie());
         UNIT_ASSERT_VALUES_EQUAL(0, blobId0.PartId());
@@ -4066,7 +4066,7 @@ Y_UNIT_TEST_SUITE(TIndexTabletTest_Data)
             NKikimr::LogoBlobIDFromLogoBlobID(blobPieces[1].GetBlobId());
         UNIT_ASSERT_VALUES_EQUAL(tabletId, blobId1.TabletID());
         UNIT_ASSERT_VALUES_EQUAL(2, blobId1.Generation());
-        UNIT_ASSERT_VALUES_EQUAL(10, blobId1.Step());
+        UNIT_ASSERT_VALUES_EQUAL(11, blobId1.Step());
         // 4_KB block leads to 4 blobs, 16_KB block leads to 2 blobs => this
         // blob becomes blob #4 => we start from channel #3, write our 1st blob
         // to it, next blob goes to channel #4, then #5 and the 4th blob goes
@@ -4150,7 +4150,7 @@ Y_UNIT_TEST_SUITE(TIndexTabletTest_Data)
             NKikimr::LogoBlobIDFromLogoBlobID(blobPieces[0].GetBlobId());
         UNIT_ASSERT_VALUES_EQUAL(tabletId, blobId0.TabletID());
         UNIT_ASSERT_VALUES_EQUAL(2, blobId0.Generation());
-        UNIT_ASSERT_VALUES_EQUAL(5, blobId0.Step());
+        UNIT_ASSERT_VALUES_EQUAL(6, blobId0.Step());
         UNIT_ASSERT_VALUES_EQUAL(4, blobId0.Channel());
         UNIT_ASSERT_VALUES_EQUAL(0, blobId0.Cookie());
         UNIT_ASSERT_VALUES_EQUAL(0, blobId0.PartId());
@@ -5586,6 +5586,81 @@ Y_UNIT_TEST_SUITE(TIndexTabletTest_Data)
         }
     }
 
+    TABLET_TEST(ShouldNotCollectGarbageWithPreviousGeneration)
+    {
+        const auto block = tabletConfig.BlockSize;
+
+        NProto::TStorageConfig storageConfig;
+        storageConfig.SetCompactionThreshold(999'999);
+        storageConfig.SetCleanupThreshold(999'999);
+        storageConfig.SetWriteBlobThreshold(block);
+        storageConfig.SetCollectGarbageThreshold(block);
+
+        TTestEnv env({}, std::move(storageConfig));
+        env.CreateSubDomain("nfs");
+
+        ui32 nodeIdx = env.CreateNode("nfs");
+        ui64 tabletId = env.BootIndexTablet(nodeIdx);
+
+        ui32 barrierGen = 0;
+        ui32 perGenerationCounter = 0;
+        ui32 recordGeneration = 0;
+        ui32 collectStep = 0;
+        bool firstMessageSeen = false;
+        env.GetRuntime().SetEventFilter([&] (auto& runtime, auto& event) {
+            Y_UNUSED(runtime);
+
+            switch (event->GetTypeRewrite()) {
+                case TEvBlobStorage::EvCollectGarbage: {
+                    const auto* msg =
+                        event->template Get<TEvBlobStorage::TEvCollectGarbage>();
+                    if (msg->TabletId == tabletId && !firstMessageSeen) {
+                        barrierGen = msg->CollectGeneration;
+                        perGenerationCounter = msg->PerGenerationCounter;
+                        recordGeneration = msg->RecordGeneration;
+                        collectStep = msg->CollectStep;
+                        firstMessageSeen = true;
+                        return true;
+                    }
+                }
+            }
+
+            return false;
+        });
+
+        TIndexTabletClient tablet(env.GetRuntime(), nodeIdx, tabletId);
+        tablet.InitSession("client", "session");
+
+        auto id = CreateNode(tablet, TCreateNodeArgs::File(RootNodeId, "test"));
+        auto handle = CreateHandle(tablet, id);
+
+        // directly written blob
+        tablet.WriteData(handle, 0, block, 'c');
+
+        TDispatchOptions options;
+        options.CustomFinalCondition = [&] {
+            return barrierGen;
+        };
+        env.GetRuntime().DispatchEvents(options);
+
+        UNIT_ASSERT_VALUES_EQUAL(recordGeneration, barrierGen);
+
+        auto oldBarrierGen = barrierGen;
+        barrierGen = 0;
+        perGenerationCounter = 0;
+        recordGeneration = 0;
+        collectStep = 0;
+        firstMessageSeen = false;
+
+        tablet.RebootTablet();
+
+        env.GetRuntime().DispatchEvents(options);
+
+        UNIT_ASSERT_VALUES_EQUAL(0, collectStep);
+        UNIT_ASSERT_VALUES_EQUAL(oldBarrierGen + 1, barrierGen);
+        UNIT_ASSERT_VALUES_EQUAL(1, perGenerationCounter);
+    }
+
 #undef TABLET_TEST
 }