issue-2086: use multiple aio services (#2126)

* issue-2086: use multiple aio services
ydb-platform · Sep 27, 2024 · f4cf1bb · f4cf1bb
1 parent 04cca17
commit f4cf1bb
Show file tree

Hide file tree

Showing 19 changed files with 113 additions and 75 deletions.
diff --git a/cloud/blockstore/config/disk.proto b/cloud/blockstore/config/disk.proto
@@ -251,6 +251,13 @@ message TDiskAgentConfig
 
     // Max. count of concurrently processing operations (io_setup).
     optional uint32 MaxAIOContextEvents = 33;
+
+    // Number of paths per each FileIOService.
+    // 0 means that only one IO service will be created.
+    // 1 means that one IO service per file path will be created.
+    // 2 means that one IO service will be created for every two file paths.
+    // etc.
+    optional uint32 PathsPerFileIOService = 34;
 }
 
 ////////////////////////////////////////////////////////////////////////////////

diff --git a/cloud/blockstore/libs/daemon/common/bootstrap.cpp b/cloud/blockstore/libs/daemon/common/bootstrap.cpp
@@ -56,6 +56,7 @@
 #include <cloud/blockstore/libs/service/service_filtered.h>
 #include <cloud/blockstore/libs/service/service_null.h>
 #include <cloud/blockstore/libs/service/storage_provider.h>
+#include <cloud/blockstore/libs/service_local/file_io_service_provider.h>
 #include <cloud/blockstore/libs/service_local/service_local.h>
 #include <cloud/blockstore/libs/service_local/storage_aio.h>
 #include <cloud/blockstore/libs/service_local/storage_null.h>
@@ -740,7 +741,8 @@ void TBootstrapBase::InitLocalService()
         ? *Configs->ServerConfig->GetLocalServiceConfig()
         : NProto::TLocalServiceConfig();
 
-    FileIOService = CreateAIOService();
+    FileIOServiceProvider =
+        CreateSingleFileIOServiceProvider(CreateAIOService());
 
     NvmeManager = CreateNvmeManager(
         Configs->DiskAgentConfig->GetSecureEraseTimeout());
@@ -749,7 +751,7 @@ void TBootstrapBase::InitLocalService()
         config,
         DiscoveryService,
         CreateAioStorageProvider(
-            FileIOService,
+            FileIOServiceProvider,
             NvmeManager,
             false,  // directIO
             EAioSubmitQueueOpt::DontUse
@@ -869,7 +871,7 @@ void TBootstrapBase::Start()
     START_KIKIMR_COMPONENT(StatsUploader);
     START_COMMON_COMPONENT(Spdk);
     START_KIKIMR_COMPONENT(ActorSystem);
-    START_COMMON_COMPONENT(FileIOService);
+    START_COMMON_COMPONENT(FileIOServiceProvider);
     START_COMMON_COMPONENT(EndpointProxyClient);
     START_COMMON_COMPONENT(EndpointManager);
     START_COMMON_COMPONENT(Service);
@@ -939,7 +941,7 @@ void TBootstrapBase::Stop()
     STOP_COMMON_COMPONENT(Service);
     STOP_COMMON_COMPONENT(EndpointManager);
     STOP_COMMON_COMPONENT(EndpointProxyClient);
-    STOP_COMMON_COMPONENT(FileIOService);
+    STOP_COMMON_COMPONENT(FileIOServiceProvider);
     STOP_KIKIMR_COMPONENT(ActorSystem);
     STOP_COMMON_COMPONENT(Spdk);
     STOP_KIKIMR_COMPONENT(StatsUploader);

diff --git a/cloud/blockstore/libs/daemon/common/bootstrap.h b/cloud/blockstore/libs/daemon/common/bootstrap.h
@@ -61,7 +61,7 @@ class TBootstrapBase
     NVhost::IServerPtr VhostServer;
     NVhost::TVhostCallbacks VhostCallbacks;
     NBD::IServerPtr NbdServer;
-    IFileIOServicePtr FileIOService;
+    IFileIOServiceProviderPtr FileIOServiceProvider;
     IStorageProviderPtr StorageProvider;
     IKmsKeyProviderPtr KmsKeyProvider;
     IRootKmsKeyProviderPtr RootKmsKeyProvider;

diff --git a/cloud/blockstore/libs/daemon/ydb/bootstrap.cpp b/cloud/blockstore/libs/daemon/ydb/bootstrap.cpp
@@ -32,6 +32,7 @@
 #include <cloud/blockstore/libs/service/service_auth.h>
 #include <cloud/blockstore/libs/service_kikimr/auth_provider_kikimr.h>
 #include <cloud/blockstore/libs/service_kikimr/service_kikimr.h>
+#include <cloud/blockstore/libs/service_local/file_io_service_provider.h>
 #include <cloud/blockstore/libs/service_local/storage_aio.h>
 #include <cloud/blockstore/libs/service_local/storage_null.h>
 #include <cloud/blockstore/libs/spdk/iface/env.h>
@@ -433,23 +434,30 @@ void TBootstrapYdb::InitKikimrService()
 
     InitSpdk();
 
-    FileIOService = CreateAIOService();
-
-    if (Configs->DiskAgentConfig->GetEnabled() &&
-        Configs->DiskAgentConfig->GetBackend() == NProto::DISK_AGENT_BACKEND_AIO &&
+    if (const auto& config = *Configs->DiskAgentConfig;
+        config.GetEnabled() &&
+        config.GetBackend() == NProto::DISK_AGENT_BACKEND_AIO &&
         !AioStorageProvider)
     {
-        Y_ABORT_UNLESS(FileIOService);
-
         NvmeManager = CreateNvmeManager(
             Configs->DiskAgentConfig->GetSecureEraseTimeout());
 
+        auto factory = [events = config.GetMaxAIOContextEvents()] {
+            return CreateAIOService(events);
+        };
+
+        FileIOServiceProvider =
+            config.GetPathsPerFileIOService()
+                ? CreateFileIOServiceProvider(
+                      config.GetPathsPerFileIOService(),
+                      factory)
+                : CreateSingleFileIOServiceProvider(factory());
+
         AioStorageProvider = CreateAioStorageProvider(
-            FileIOService,
+            FileIOServiceProvider,
             NvmeManager,
-            !Configs->DiskAgentConfig->GetDirectIoFlagDisabled(),
-            EAioSubmitQueueOpt::DontUse
-        );
+            !config.GetDirectIoFlagDisabled(),
+            EAioSubmitQueueOpt::DontUse);
 
         STORAGE_INFO("AioStorageProvider initialized");
     }
@@ -465,8 +473,6 @@ void TBootstrapYdb::InitKikimrService()
         STORAGE_INFO("AioStorageProvider (null) initialized");
     }
 
-    Y_ABORT_UNLESS(FileIOService);
-
     Allocator = CreateCachingAllocator(
         Spdk ? Spdk->GetAllocator() : TDefaultAllocator::Instance(),
         Configs->DiskAgentConfig->GetPageSize(),
@@ -534,7 +540,6 @@ void TBootstrapYdb::InitKikimrService()
     args.DiscoveryService = DiscoveryService;
     args.Spdk = Spdk;
     args.Allocator = Allocator;
-    args.FileIOService = FileIOService;
     args.AioStorageProvider = AioStorageProvider;
     args.ProfileLog = ProfileLog;
     args.BlockDigestGenerator = BlockDigestGenerator;

diff --git a/cloud/blockstore/libs/disk_agent/bootstrap.cpp b/cloud/blockstore/libs/disk_agent/bootstrap.cpp
@@ -19,6 +19,7 @@
 #include <cloud/blockstore/libs/rdma/iface/probes.h>
 #include <cloud/blockstore/libs/rdma/iface/server.h>
 #include <cloud/blockstore/libs/server/config.h>
+#include <cloud/blockstore/libs/service_local/file_io_service_provider.h>
 #include <cloud/blockstore/libs/service_local/storage_aio.h>
 #include <cloud/blockstore/libs/service_local/storage_null.h>
 #include <cloud/blockstore/libs/spdk/iface/config.h>
@@ -430,20 +431,30 @@ bool TBootstrap::InitKikimrService()
                 break;
             }
 
-            case NProto::DISK_AGENT_BACKEND_AIO:
-                FileIOService =
-                    CreateAIOService(config.GetMaxAIOContextEvents());
+            case NProto::DISK_AGENT_BACKEND_AIO: {
                 NvmeManager = CreateNvmeManager(config.GetSecureEraseTimeout());
 
+                auto factory = [events = config.GetMaxAIOContextEvents()] {
+                    return CreateAIOService(events);
+                };
+
+                FileIOServiceProvider =
+                    config.GetPathsPerFileIOService()
+                        ? CreateFileIOServiceProvider(
+                              config.GetPathsPerFileIOService(),
+                              factory)
+                        : CreateSingleFileIOServiceProvider(factory());
+
                 AioStorageProvider = CreateAioStorageProvider(
-                    FileIOService,
+                    FileIOServiceProvider,
                     NvmeManager,
                     !config.GetDirectIoFlagDisabled(),
                     EAioSubmitQueueOpt::Use
                 );
 
                 STORAGE_INFO("Aio backend initialized");
                 break;
+            }
             case NProto::DISK_AGENT_BACKEND_NULL:
                 NvmeManager = CreateNvmeManager(config.GetSecureEraseTimeout());
                 AioStorageProvider = CreateNullStorageProvider();
@@ -489,7 +500,6 @@ bool TBootstrap::InitKikimrService()
     args.AsyncLogger = AsyncLogger;
     args.Spdk = Spdk;
     args.Allocator = Allocator;
-    args.FileIOService = FileIOService;
     args.AioStorageProvider = AioStorageProvider;
     args.ProfileLog = ProfileLog;
     args.BlockDigestGenerator = BlockDigestGenerator;
@@ -606,7 +616,7 @@ void TBootstrap::Start()
     START_COMPONENT(TraceProcessor);
     START_COMPONENT(Spdk);
     START_COMPONENT(RdmaServer);
-    START_COMPONENT(FileIOService);
+    START_COMPONENT(FileIOServiceProvider);
     START_COMPONENT(ActorSystem);
 
     // we need to start scheduler after all other components for 2 reasons:
@@ -644,9 +654,9 @@ void TBootstrap::Stop()
     STOP_COMPONENT(Scheduler);
 
     STOP_COMPONENT(ActorSystem);
-    // stop FileIOService after ActorSystem to ensure that there are no
+    // stop FileIOServiceProvider after ActorSystem to ensure that there are no
     // in-flight I/O requests from TDiskAgentActor
-    STOP_COMPONENT(FileIOService);
+    STOP_COMPONENT(FileIOServiceProvider);
 
     STOP_COMPONENT(Spdk);
     STOP_COMPONENT(RdmaServer);

diff --git a/cloud/blockstore/libs/disk_agent/bootstrap.h b/cloud/blockstore/libs/disk_agent/bootstrap.h
@@ -64,7 +64,7 @@ class TBootstrap
     ITraceProcessorPtr TraceProcessor;
     IProfileLogPtr ProfileLog;
     IBlockDigestGeneratorPtr BlockDigestGenerator;
-    IFileIOServicePtr FileIOService;
+    IFileIOServiceProviderPtr FileIOServiceProvider;
     NSpdk::ISpdkEnvPtr Spdk;
     std::function<void(TLog& log)> SpdkLogInitializer;
     ICachingAllocatorPtr Allocator;

diff --git a/cloud/blockstore/libs/service_local/file_io_service_provider.cpp b/cloud/blockstore/libs/service_local/file_io_service_provider.cpp
@@ -64,12 +64,12 @@ class TFileIOServiceProvider final
 
 ////////////////////////////////////////////////////////////////////////////////
 
-struct TFileIOServiceProviderStub final
+struct TSingleFileIOServiceProvider final
     : IFileIOServiceProvider
 {
     IFileIOServicePtr FileIO;
 
-    explicit TFileIOServiceProviderStub(
+    explicit TSingleFileIOServiceProvider(
             IFileIOServicePtr fileIO)
         : FileIO{std::move(fileIO)}
     {}
@@ -96,10 +96,10 @@ struct TFileIOServiceProviderStub final
 
 ////////////////////////////////////////////////////////////////////////////////
 
-IFileIOServiceProviderPtr CreateFileIOServiceProviderStub(
+IFileIOServiceProviderPtr CreateSingleFileIOServiceProvider(
     IFileIOServicePtr fileIO)
 {
-    return std::make_shared<TFileIOServiceProviderStub>(std::move(fileIO));
+    return std::make_shared<TSingleFileIOServiceProvider>(std::move(fileIO));
 }
 
 IFileIOServiceProviderPtr CreateFileIOServiceProvider(

diff --git a/cloud/blockstore/libs/service_local/file_io_service_provider.h b/cloud/blockstore/libs/service_local/file_io_service_provider.h
@@ -21,7 +21,7 @@ struct IFileIOServiceProvider
 
 ////////////////////////////////////////////////////////////////////////////////
 
-IFileIOServiceProviderPtr CreateFileIOServiceProviderStub(
+IFileIOServiceProviderPtr CreateSingleFileIOServiceProvider(
     IFileIOServicePtr fileIO);
 
 IFileIOServiceProviderPtr CreateFileIOServiceProvider(

diff --git a/cloud/blockstore/libs/service_local/file_io_service_provider_ut.cpp b/cloud/blockstore/libs/service_local/file_io_service_provider_ut.cpp
@@ -98,7 +98,7 @@ Y_UNIT_TEST_SUITE(TFileIOServiceProviderTest)
         auto fileIO = std::make_shared<TTestFileIOService>();
         UNIT_ASSERT_VALUES_EQUAL(0, fileIO->Started);
 
-        auto provider = CreateFileIOServiceProviderStub(fileIO);
+        auto provider = CreateSingleFileIOServiceProvider(fileIO);
         provider->Start();
 
         UNIT_ASSERT_VALUES_EQUAL(1, fileIO->Started);

diff --git a/cloud/blockstore/libs/service_local/storage_aio.cpp b/cloud/blockstore/libs/service_local/storage_aio.cpp
@@ -1,5 +1,7 @@
 #include "storage_aio.h"
 
+#include "file_io_service_provider.h"
+
 #include <cloud/blockstore/libs/common/iovector.h>
 #include <cloud/blockstore/libs/nvme/nvme.h>
 #include <cloud/blockstore/libs/service/context.h>
@@ -882,7 +884,7 @@ class TSafeDeallocator
         const NProto::TError& error,
         ui32 bytes)
     {
-        auto checker = static_cast<TSafeDeallocator*>(completion);
+        auto* checker = static_cast<TSafeDeallocator*>(completion);
         checker->ReadBlockComplete(error, bytes);
     }
 };
@@ -916,18 +918,18 @@ class TAioStorageProvider final
 {
 private:
     ITaskQueuePtr SubmitQueue;
-    IFileIOServicePtr FileIOService;
+    IFileIOServiceProviderPtr FileIOServiceProvider;
     INvmeManagerPtr NvmeManager;
     const bool DirectIO;
 
 public:
     explicit TAioStorageProvider(
             ITaskQueuePtr submitQueue,
-            IFileIOServicePtr fileIO,
+            IFileIOServiceProviderPtr fileIOProvider,
             INvmeManagerPtr nvmeManager,
             bool directIO)
         : SubmitQueue(std::move(submitQueue))
-        , FileIOService(std::move(fileIO))
+        , FileIOServiceProvider(std::move(fileIOProvider))
         , NvmeManager(std::move(nvmeManager))
         , DirectIO(directIO)
     {}
@@ -957,7 +959,7 @@ class TAioStorageProvider final
 
         auto storage = std::make_shared<TAioStorage>(
             SubmitQueue,
-            FileIOService,
+            FileIOServiceProvider->CreateFileIOService(filePath),
             NvmeManager,
             blockSize,
             volume.GetStartIndex(),
@@ -974,7 +976,7 @@ class TAioStorageProvider final
 ////////////////////////////////////////////////////////////////////////////////
 
 IStorageProviderPtr CreateAioStorageProvider(
-    IFileIOServicePtr fileIO,
+    IFileIOServiceProviderPtr fileIOProvider,
     INvmeManagerPtr nvmeManager,
     bool directIO,
     EAioSubmitQueueOpt submitQueueOpt)
@@ -986,7 +988,7 @@ IStorageProviderPtr CreateAioStorageProvider(
 
     return std::make_shared<TAioStorageProvider>(
         std::move(submitQueue),
-        std::move(fileIO),
+        std::move(fileIOProvider),
         std::move(nvmeManager),
         directIO);
 }

diff --git a/cloud/blockstore/libs/service_local/storage_aio.h b/cloud/blockstore/libs/service_local/storage_aio.h
@@ -19,7 +19,7 @@ enum class EAioSubmitQueueOpt : bool {
 ////////////////////////////////////////////////////////////////////////////////
 
 IStorageProviderPtr CreateAioStorageProvider(
-    IFileIOServicePtr fileIO,
+    IFileIOServiceProviderPtr fileIOProvider,
     NNvme::INvmeManagerPtr nvmeManager,
     bool directIO,
     EAioSubmitQueueOpt submitQueueOpt);