merge to stable-23-3: filestore-vhost local mode + async read/write in local service; lazy filestore counters init; support for large unaligned writes in ThreeStageWrite mode

cloud/filestore/config/storage.proto

@@ -309,4 +309,7 @@ message TStorageConfig
     // Blob compression experiment params.
     optional uint32 BlobCompressionRate = 364;
     optional string BlobCompressionCodec = 365;
+
+    // Enables ThreeStageWrite for unaligned requests.
+    optional bool UnalignedThreeStageWriteEnabled = 366;
 }

cloud/filestore/libs/daemon/common/CMakeLists.darwin-x86_64.txt

@@ -24,6 +24,7 @@ target_link_libraries(filestore-libs-daemon-common PUBLIC
   filestore-libs-service
   filestore-libs-storage-core
   libs-storage-init
+  core-libs-aio
   core-libs-common
   core-libs-daemon
   core-libs-diagnostics

cloud/filestore/libs/daemon/common/CMakeLists.linux-aarch64.txt

@@ -25,6 +25,7 @@ target_link_libraries(filestore-libs-daemon-common PUBLIC
   filestore-libs-service
   filestore-libs-storage-core
   libs-storage-init
+  core-libs-aio
   core-libs-common
   core-libs-daemon
   core-libs-diagnostics

cloud/filestore/libs/daemon/common/CMakeLists.linux-x86_64.txt

@@ -25,6 +25,7 @@ target_link_libraries(filestore-libs-daemon-common PUBLIC
   filestore-libs-service
   filestore-libs-storage-core
   libs-storage-init
+  core-libs-aio
   core-libs-common
   core-libs-daemon
   core-libs-diagnostics

cloud/filestore/libs/daemon/common/CMakeLists.windows-x86_64.txt

@@ -24,6 +24,7 @@ target_link_libraries(filestore-libs-daemon-common PUBLIC
   filestore-libs-service
   filestore-libs-storage-core
   libs-storage-init
+  core-libs-aio
   core-libs-common
   core-libs-daemon
   core-libs-diagnostics

cloud/filestore/libs/daemon/common/bootstrap.cpp

@@ -12,6 +12,8 @@
 #include <cloud/filestore/libs/storage/core/config.h>
 #include <cloud/filestore/libs/storage/init/actorsystem.h>
 
+#include <cloud/storage/core/libs/aio/service.h>
+#include <cloud/storage/core/libs/common/file_io_service.h>
 #include <cloud/storage/core/libs/common/scheduler.h>
 #include <cloud/storage/core/libs/common/task_queue.h>
 #include <cloud/storage/core/libs/common/thread_pool.h>
@@ -116,6 +118,7 @@ void TBootstrapCommon::Start()
 {
     FILESTORE_LOG_START_COMPONENT(Logging);
     FILESTORE_LOG_START_COMPONENT(Monitoring);
+    FILESTORE_LOG_START_COMPONENT(FileIOService);
     FILESTORE_LOG_START_COMPONENT(Metrics);
     FILESTORE_LOG_START_COMPONENT(TraceProcessor);
     FILESTORE_LOG_START_COMPONENT(TraceSerializer);
@@ -160,6 +163,7 @@ void TBootstrapCommon::Stop()
     FILESTORE_LOG_STOP_COMPONENT(TraceSerializer);
     FILESTORE_LOG_STOP_COMPONENT(TraceProcessor);
     FILESTORE_LOG_STOP_COMPONENT(Metrics);
+    FILESTORE_LOG_STOP_COMPONENT(FileIOService);
     FILESTORE_LOG_STOP_COMPONENT(Monitoring);
     FILESTORE_LOG_STOP_COMPONENT(Logging);
 }
@@ -216,6 +220,8 @@ void TBootstrapCommon::Init()
         InitActorSystem();
     }
 
+    FileIOService = CreateAIOService();
+
     InitDiagnostics();
     InitComponents();
 

cloud/filestore/libs/daemon/common/bootstrap.h

@@ -63,6 +63,7 @@ class TBootstrapCommon
     ISchedulerPtr BackgroundScheduler;
     ILoggingServicePtr Logging;
     IMonitoringServicePtr Monitoring;
+    IFileIOServicePtr FileIOService;
     NMetrics::IMetricsServicePtr Metrics;
     IRequestStatsRegistryPtr StatsRegistry;
     IStatsUpdaterPtr RequestStatsUpdater;

cloud/filestore/libs/daemon/common/ya.make

@@ -16,6 +16,7 @@ PEERDIR(
     cloud/filestore/libs/storage/core
     cloud/filestore/libs/storage/init
 
+    cloud/storage/core/libs/aio
     cloud/storage/core/libs/common
     cloud/storage/core/libs/daemon
     cloud/storage/core/libs/diagnostics

cloud/filestore/libs/daemon/server/bootstrap.cpp

@@ -157,6 +157,7 @@ void TBootstrapServer::InitLocalService()
         Timer,
         Scheduler,
         Logging,
+        FileIOService,
         ThreadPool);
 }
 

cloud/filestore/libs/daemon/vhost/bootstrap.cpp

@@ -292,6 +292,7 @@ void TBootstrapVhost::InitEndpoints()
             Timer,
             Scheduler,
             Logging,
+            FileIOService,
             ThreadPool);
 
         STORAGE_INFO("initialized LocalService: %s",
@@ -375,7 +376,7 @@ void TBootstrapVhost::StartComponents()
     NVhost::StartServer();
 
     FILESTORE_LOG_START_COMPONENT(ThreadPool);
-    FILESTORE_LOG_START_COMPONENT(LocalService);
+    // LocalService is started inside FileStoreEndpoints
     FILESTORE_LOG_START_COMPONENT(FileStoreEndpoints);
     FILESTORE_LOG_START_COMPONENT(EndpointManager);
     FILESTORE_LOG_START_COMPONENT(Server);
@@ -390,7 +391,7 @@ void TBootstrapVhost::StopComponents()
     FILESTORE_LOG_STOP_COMPONENT(Server);
     FILESTORE_LOG_STOP_COMPONENT(EndpointManager);
     FILESTORE_LOG_STOP_COMPONENT(FileStoreEndpoints);
-    FILESTORE_LOG_STOP_COMPONENT(LocalService);
+    // LocalService is stopped inside FileStoreEndpoints
     FILESTORE_LOG_STOP_COMPONENT(ThreadPool);
 
     NVhost::StopServer();

cloud/filestore/libs/diagnostics/critical_events.h

@@ -33,6 +33,7 @@ namespace NCloud::NFileStore{
     xxx(ChildRefIsNull)                                                        \
     xxx(NewChildNodeIsNull)                                                    \
     xxx(IndexOutOfBounds)                                                      \
+    xxx(CheckFreshBytesFailed)                                                 \
 // FILESTORE_IMPOSSIBLE_EVENTS
 
 ////////////////////////////////////////////////////////////////////////////////

cloud/filestore/libs/diagnostics/request_stats.cpp

@@ -38,9 +38,14 @@ bool IsReadWriteRequest(EFileStoreRequest rt)
         || rt == EFileStoreRequest::ReadData;
 }
 
+const auto REQUEST_COUNTERS_OPTIONS =
+    TRequestCounters::EOption::ReportDataPlaneHistogram |
+    TRequestCounters::EOption::ReportControlPlaneHistogram;
+
 TRequestCountersPtr MakeRequestCounters(
     ITimerPtr timer,
-    TDynamicCounters& counters)
+    TDynamicCounters& counters,
+    TRequestCounters::EOptions options)
 {
     auto requestCounters = std::make_shared<TRequestCounters>(
         std::move(timer),
@@ -51,8 +56,7 @@ TRequestCountersPtr MakeRequestCounters(
         [] (TRequestCounters::TRequestType t) {
             return IsReadWriteRequest(static_cast<EFileStoreRequest>(t));
         },
-        TRequestCounters::EOption::ReportDataPlaneHistogram |
-        TRequestCounters::EOption::ReportControlPlaneHistogram
+        options
     );
     requestCounters->Register(counters);
     return requestCounters;
@@ -157,13 +161,18 @@ class TRequestStats final
             TDuration totalTimeThreshold)
         : TRequestLogger(executionTimeThreshold, totalTimeThreshold)
         , RootCounters(std::move(counters))
-        , TotalCounters(MakeRequestCounters(timer, *RootCounters))
+        , TotalCounters(MakeRequestCounters(
+            timer,
+            *RootCounters,
+            REQUEST_COUNTERS_OPTIONS))
         , SsdCounters(MakeRequestCounters(
             timer,
-            *RootCounters->GetSubgroup("type", "ssd")))
+            *RootCounters->GetSubgroup("type", "ssd"),
+            REQUEST_COUNTERS_OPTIONS))
         , HddCounters(MakeRequestCounters(
             timer,
-            *RootCounters->GetSubgroup("type", "hdd")))
+            *RootCounters->GetSubgroup("type", "hdd"),
+            REQUEST_COUNTERS_OPTIONS))
     {
         auto revisionGroup =
             RootCounters->GetSubgroup("revision", GetFullVersionString());
@@ -420,7 +429,11 @@ class TFileSystemStats final
         : TRequestLogger{executionTimeThreshold, totalTimeThreshold}
         , FileSystemId{std::move(fileSystemId)}
         , ClientId{std::move(clientId)}
-        , Counters{MakeRequestCounters(timer, *counters)}
+        , Counters{MakeRequestCounters(
+            timer,
+            *counters,
+            REQUEST_COUNTERS_OPTIONS
+            | TRequestCounters::EOption::LazyRequestInitialization)}
         , Predictor{std::move(predictor)}
         , PredictorStats{counters, std::move(timer)}
     {}

cloud/filestore/libs/diagnostics/request_stats_ut.cpp

@@ -772,6 +772,81 @@ Y_UNIT_TEST_SUITE(TRequestStatRegistryTest)
             UNIT_ASSERT(!counters);
         }
     }
+
+    Y_UNIT_TEST(ShouldNotReportZeroCounters)
+    {
+        const TString FS = "test";
+        const TString CLIENT = "client";
+
+        TBootstrap bootstrap;
+
+        auto fsCounters = bootstrap.Counters
+            ->FindSubgroup("component", METRIC_FS_COMPONENT);
+        UNIT_ASSERT(fsCounters);
+        fsCounters = fsCounters->FindSubgroup("host", "cluster");
+        UNIT_ASSERT(fsCounters);
+
+        auto stats = bootstrap.Registry->GetFileSystemStats(FS, CLIENT);
+        UNIT_ASSERT(stats);
+        fsCounters = fsCounters->FindSubgroup("filesystem", FS);
+        UNIT_ASSERT(fsCounters);
+        fsCounters = fsCounters->FindSubgroup("client", CLIENT);
+        UNIT_ASSERT(fsCounters);
+
+        // non lazy-init request
+        auto readData = fsCounters->FindSubgroup("request", "ReadData");
+        UNIT_ASSERT(readData);
+
+        // lazy-init request
+        auto createHandle = fsCounters->FindSubgroup("request", "CreateHandle");
+        UNIT_ASSERT(createHandle);
+
+        // these ones should always be present
+        auto readDataCount = readData->FindCounter("Count");
+        auto readDataErrors = readData->FindCounter("Errors");
+
+        // non lazy-init counters
+        auto createHandleCount = createHandle->FindCounter("Count");
+        auto createHandleErrorsFatal =
+            createHandle->FindCounter("Errors/Fatal");
+        // lazy-init counter
+        auto createHandleErrors = createHandle->FindCounter("Errors");
+
+        UNIT_ASSERT(readDataCount);
+        UNIT_ASSERT_VALUES_EQUAL(0, readDataCount->Val());
+        UNIT_ASSERT(readDataErrors);
+        UNIT_ASSERT_VALUES_EQUAL(0, readDataErrors->Val());
+        UNIT_ASSERT(createHandleCount);
+        UNIT_ASSERT_VALUES_EQUAL(0, createHandleCount->Val());
+        UNIT_ASSERT(createHandleErrorsFatal);
+        UNIT_ASSERT_VALUES_EQUAL(0, createHandleErrorsFatal->Val());
+        UNIT_ASSERT(!createHandleErrors);
+
+        {
+            auto context = MakeIntrusive<TCallContext>(FS);
+            context->RequestType = EFileStoreRequest::CreateHandle;
+            stats->RequestStarted(*context);
+            stats->RequestCompleted(*context, MakeError(S_OK));
+
+            // we got a request - now all counters for this request should be
+            // initialized
+            createHandleErrors = createHandle->FindCounter("Errors");
+            UNIT_ASSERT(createHandleErrors);
+            UNIT_ASSERT_VALUES_EQUAL(0, createHandleErrors->Val());
+            UNIT_ASSERT_VALUES_EQUAL(1, createHandleCount->Val());
+        }
+
+        {
+            auto context = MakeIntrusive<TCallContext>(FS);
+            context->RequestType = EFileStoreRequest::CreateHandle;
+            stats->RequestStarted(*context);
+            stats->RequestCompleted(*context, MakeError(E_REJECTED));
+
+            // and now we successfully registered an error
+            UNIT_ASSERT_VALUES_EQUAL(1, createHandleErrors->Val());
+            UNIT_ASSERT_VALUES_EQUAL(1, createHandleCount->Val());
+        }
+    }
 }
 
 } // namespace NCloud::NFileStore::NStorage

cloud/filestore/libs/service_local/fs.cpp

@@ -10,11 +10,13 @@ TLocalFileSystem::TLocalFileSystem(
         TFsPath root,
         ITimerPtr timer,
         ISchedulerPtr scheduler,
-        ILoggingServicePtr logging)
+        ILoggingServicePtr logging,
+        IFileIOServicePtr fileIOService)
     : Config(std::move(config))
     , Root(std::move(root))
     , Timer(std::move(timer))
     , Scheduler(std::move(scheduler))
+    , FileIOService(std::move(fileIOService))
     , Store(std::move(store))
 {
     Log = logging->CreateLog(Store.GetFileSystemId());

cloud/filestore/libs/service_local/fs.h

@@ -28,6 +28,56 @@ namespace NCloud::NFileStore {
 
 ////////////////////////////////////////////////////////////////////////////////
 
+#define FILESTORE_DATA_METHODS_LOCAL_SYNC(xxx, ...)                            \
+    xxx(StatFileStore,                      __VA_ARGS__)                       \
+                                                                               \
+    xxx(SubscribeSession,                   __VA_ARGS__)                       \
+    xxx(GetSessionEvents,                   __VA_ARGS__)                       \
+    xxx(ResetSession,                       __VA_ARGS__)                       \
+                                                                               \
+    xxx(ResolvePath,                        __VA_ARGS__)                       \
+    xxx(CreateNode,                         __VA_ARGS__)                       \
+    xxx(UnlinkNode,                         __VA_ARGS__)                       \
+    xxx(RenameNode,                         __VA_ARGS__)                       \
+    xxx(AccessNode,                         __VA_ARGS__)                       \
+    xxx(ListNodes,                          __VA_ARGS__)                       \
+    xxx(ReadLink,                           __VA_ARGS__)                       \
+                                                                               \
+    xxx(SetNodeAttr,                        __VA_ARGS__)                       \
+    xxx(GetNodeAttr,                        __VA_ARGS__)                       \
+    xxx(SetNodeXAttr,                       __VA_ARGS__)                       \
+    xxx(GetNodeXAttr,                       __VA_ARGS__)                       \
+    xxx(ListNodeXAttr,                      __VA_ARGS__)                       \
+    xxx(RemoveNodeXAttr,                    __VA_ARGS__)                       \
+                                                                               \
+    xxx(CreateHandle,                       __VA_ARGS__)                       \
+    xxx(DestroyHandle,                      __VA_ARGS__)                       \
+                                                                               \
+    xxx(AcquireLock,                        __VA_ARGS__)                       \
+    xxx(ReleaseLock,                        __VA_ARGS__)                       \
+    xxx(TestLock,                           __VA_ARGS__)                       \
+                                                                               \
+    xxx(AllocateData,                       __VA_ARGS__)                       \
+// FILESTORE_DATA_METHODS_LOCAL_SYNC
+
+#define FILESTORE_DATA_METHODS_LOCAL_ASYNC(xxx, ...)                           \
+    xxx(ReadData,                           __VA_ARGS__)                       \
+    xxx(WriteData,                          __VA_ARGS__)                       \
+// FILESTORE_DATA_METHODS_LOCAL_ASYNC
+
+#define FILESTORE_SERVICE_LOCAL_SYNC(xxx, ...)                                 \
+    xxx(Ping,                               __VA_ARGS__)                       \
+    xxx(PingSession,                        __VA_ARGS__)                       \
+    FILESTORE_SERVICE_METHODS(xxx,          __VA_ARGS__)                       \
+    FILESTORE_DATA_METHODS_LOCAL_SYNC(xxx,  __VA_ARGS__)                       \
+// FILESTORE_SERVICE_LOCAL_SYNC
+
+#define FILESTORE_SERVICE_LOCAL_ASYNC(xxx, ...)                                \
+    FILESTORE_DATA_METHODS_LOCAL_ASYNC(xxx,  __VA_ARGS__)                      \
+// FILESTORE_SERVICE_LOCAL_SYNC
+
+////////////////////////////////////////////////////////////////////////////////
+
 class TLocalFileSystem final
     : public std::enable_shared_from_this<TLocalFileSystem>
 {
@@ -38,6 +88,7 @@ class TLocalFileSystem final
     const TFsPath Root;
     const ITimerPtr Timer;
     const ISchedulerPtr Scheduler;
+    const IFileIOServicePtr FileIOService;
 
     NProto::TFileStore Store;
     TLog Log;
@@ -59,16 +110,24 @@ class TLocalFileSystem final
         TFsPath root,
         ITimerPtr timer,
         ISchedulerPtr scheduler,
-        ILoggingServicePtr logging);
+        ILoggingServicePtr logging,
+        IFileIOServicePtr fileIOService);
 
-#define FILESTORE_DECLARE_METHOD(name, ...)                                    \
+#define FILESTORE_DECLARE_METHOD_SYNC(name, ...)                               \
     NProto::T##name##Response name(                                            \
         const NProto::T##name##Request& request);                              \
-// FILESTORE_DECLARE_METHOD
+// FILESTORE_DECLARE_METHOD_SYNC
+
+#define FILESTORE_DECLARE_METHOD_ASYNC(name, ...)                              \
+    NThreading::TFuture<NProto::T##name##Response> name##Async(                \
+        NProto::T##name##Request& request);                                    \
+// FILESTORE_DECLARE_METHOD_SYNC
 
-    FILESTORE_SERVICE(FILESTORE_DECLARE_METHOD)
+    FILESTORE_SERVICE_LOCAL_SYNC(FILESTORE_DECLARE_METHOD_SYNC)
+    FILESTORE_SERVICE_LOCAL_ASYNC(FILESTORE_DECLARE_METHOD_ASYNC)
 
-#undef FILESTORE_DECLARE_METHOD
+#undef FILESTORE_DECLARE_METHOD_SYNC
+#undef FILESTORE_DECLARE_METHOD_ASYNC
 
     NProto::TFileStore GetConfig() const
     {