[Filestore] issue-1751: support guest writeback cache

Pass FUSE_CAP_WRITEBACK_CACHE to fuse client when GuestWritebackCacheEnabled flag is set. This flag ensures that individual write request may be buffered and merged in the kernel of the filesystem client
ydb-platform · Oct 31, 2024 · 9328308 · 9328308
1 parent 11aab3b
commit 9328308
Show file tree

Hide file tree

Showing 16 changed files with 44 additions and 5 deletions.
diff --git a/cloud/filestore/config/filesystem.proto b/cloud/filestore/config/filesystem.proto
@@ -48,4 +48,7 @@ message TFileSystemConfig
 
     // Aligment needed for buffers when using direct io
     optional uint32 DirectIoAlign = 14;
+
+    // Enable Writeback cache on guest (fuse client)
+    optional bool GuestWritebackCacheEnabled = 15;
 }
diff --git a/cloud/filestore/config/server.proto b/cloud/filestore/config/server.proto
@@ -89,6 +89,10 @@ message TLocalServiceConfig
 
     // Aligment needed for buffers when using direct io
     optional uint32 DirectIoAlign = 10;
+
+    // Enable Writeback cache on guest (fuse client)
+    optional bool GuestWritebackCacheEnabled = 11;
+
 }
 
 ////////////////////////////////////////////////////////////////////////////////

diff --git a/cloud/filestore/config/storage.proto b/cloud/filestore/config/storage.proto
@@ -456,4 +456,7 @@ message TStorageConfig
 
     // Disables TwoStageRead for HDD filesystems.
     optional bool TwoStageReadDisabledForHDD = 404;
+
+    // Enable Writeback cache on guest (fuse client)
+    optional bool GuestWritebackCacheEnabled = 405;
 }
diff --git a/cloud/filestore/libs/service_local/config.cpp b/cloud/filestore/libs/service_local/config.cpp
@@ -21,6 +21,7 @@ namespace {
     xxx(MaxHandlePerSessionCount,    ui32,          10000                     )\
     xxx(DirectIoEnabled,             bool,          false                     )\
     xxx(DirectIoAlign,               ui32,          4_KB                      )\
+    xxx(GuestWritebackCacheEnabled,  bool,          false                     )\
 // FILESTORE_SERVICE_CONFIG
 
 #define FILESTORE_SERVICE_DECLARE_CONFIG(name, type, value)                    \

diff --git a/cloud/filestore/libs/service_local/config.h b/cloud/filestore/libs/service_local/config.h
@@ -31,6 +31,7 @@ class TLocalFileStoreConfig
     ui32 GetMaxHandlePerSessionCount() const;
     bool GetDirectIoEnabled() const;
     ui32 GetDirectIoAlign() const;
+    bool GetGuestWritebackCacheEnabled() const;
 
     void Dump(IOutputStream& out) const;
     void DumpHtml(IOutputStream& out) const;

diff --git a/cloud/filestore/libs/service_local/fs_session.cpp b/cloud/filestore/libs/service_local/fs_session.cpp
@@ -78,6 +78,8 @@ NProto::TCreateSessionResponse TLocalFileSystem::CreateSession(
     auto* features = response.MutableFileStore()->MutableFeatures();
     features->SetDirectIoEnabled(Config->GetDirectIoEnabled());
     features->SetDirectIoAlign(Config->GetDirectIoAlign());
+    features->SetGuestWritebackCacheEnabled(
+        Config->GetGuestWritebackCacheEnabled());
 
     return response;
 }

diff --git a/cloud/filestore/libs/storage/core/config.cpp b/cloud/filestore/libs/storage/core/config.cpp
@@ -43,6 +43,7 @@ using TAliases = NProto::TStorageConfig::TFilestoreAliases;
     xxx(GarbageCompactionThresholdAverage,  ui32,   20                        )\
     xxx(CompactRangeGarbagePercentageThreshold, ui32,    0                    )\
     xxx(CompactRangeAverageBlobSizeThreshold,   ui32,    0                    )\
+    xxx(GuestWritebackCacheEnabled,         bool,   false                     )\
     xxx(NewCompactionEnabled,               bool,   false                     )\
     xxx(UseMixedBlocksInsteadOfAliveBlocksInCompaction, bool,   false         )\
     xxx(CollectGarbageThreshold,            ui32,   4_MB                      )\

diff --git a/cloud/filestore/libs/storage/core/config.h b/cloud/filestore/libs/storage/core/config.h
@@ -286,6 +286,8 @@ class TStorageConfig
 
     bool GetTwoStageReadDisabledForHDD() const;
     bool GetThreeStageWriteDisabledForHDD() const;
+
+    bool GetGuestWritebackCacheEnabled() const;
 };
 
 }   // namespace NCloud::NFileStore::NStorage
diff --git a/cloud/filestore/libs/storage/tablet/tablet_actor_createsession.cpp b/cloud/filestore/libs/storage/tablet/tablet_actor_createsession.cpp
@@ -43,6 +43,9 @@ void FillFeatures(const TStorageConfig& config, NProto::TFileStore& fileStore)
         config.GetAsyncDestroyHandleEnabled());
     features->SetAsyncHandleOperationPeriod(
         config.GetAsyncHandleOperationPeriod().MilliSeconds());
+
+    features->SetGuestWritebackCacheEnabled(
+        config.GetGuestWritebackCacheEnabled());
 }
 
 ////////////////////////////////////////////////////////////////////////////////

diff --git a/cloud/filestore/libs/vfs_fuse/config.cpp b/cloud/filestore/libs/vfs_fuse/config.cpp
@@ -32,6 +32,8 @@ namespace {
                                                                                \
     xxx(DirectIoEnabled,            bool,       false                         )\
     xxx(DirectIoAlign,              ui32,       4_KB                          )\
+                                                                               \
+    xxx(GuestWritebackCacheEnabled, bool,       false                         )\
 // FILESTORE_FUSE_CONFIG
 
 #define FILESTORE_FILESYSTEM_DECLARE_CONFIG(name, type, value)                 \

diff --git a/cloud/filestore/libs/vfs_fuse/config.h b/cloud/filestore/libs/vfs_fuse/config.h
@@ -42,6 +42,8 @@ struct TFileSystemConfig
     bool GetDirectIoEnabled() const;
     ui32 GetDirectIoAlign() const;
 
+    bool GetGuestWritebackCacheEnabled() const;
+
     void Dump(IOutputStream& out) const;
     void DumpHtml(IOutputStream& out) const;
 };

diff --git a/cloud/filestore/libs/vfs_fuse/loop.cpp b/cloud/filestore/libs/vfs_fuse/loop.cpp
@@ -508,6 +508,7 @@ class TFileSystemLoop final
     std::shared_ptr<TCompletionQueue> CompletionQueue;
     IRequestStatsPtr RequestStats;
     IFileSystemPtr FileSystem;
+    TFileSystemConfigPtr FileSystemConfig;
 
     bool HandleOpsQueueInitialized = false;
 
@@ -785,13 +786,13 @@ class TFileSystemLoop final
                 RequestStats,
                 Log,
                 StorageMediaKind);
-            auto filestoreConfig = MakeFileSystemConfig(filestore);
+            FileSystemConfig = MakeFileSystemConfig(filestore);
             std::unique_ptr<THandleOpsQueue> handleOpsQueue;
 
-            if (filestoreConfig->GetAsyncDestroyHandleEnabled()) {
+            if (FileSystemConfig->GetAsyncDestroyHandleEnabled()) {
                 TString path =
                     TFsPath(Config->GetHandleOpsQueuePath()) /
-                    filestoreConfig->GetFileSystemId() /
+                    FileSystemConfig->GetFileSystemId() /
                     response.GetSession().GetSessionId();
                 if (!NFs::MakeDirectoryRecursive(path)) {
                     TString msg = TStringBuilder()
@@ -816,7 +817,7 @@ class TFileSystemLoop final
                 ProfileLog,
                 Scheduler,
                 Timer,
-                filestoreConfig,
+                FileSystemConfig,
                 Session,
                 RequestStats,
                 CompletionQueue,
@@ -834,7 +835,7 @@ class TFileSystemLoop final
                 SessionState.empty() ? "new" : "existing");
 
             TStringStream filestoreConfigDump;
-            filestoreConfig->Dump(filestoreConfigDump);
+            FileSystemConfig->Dump(filestoreConfigDump);
             STORAGE_INFO(
                 "[f:%s][c:%s] new session filestore config: %s",
                 Config->GetFileSystemId().Quote().c_str(),
@@ -900,6 +901,9 @@ class TFileSystemLoop final
         config.SetDirectIoEnabled(features.GetDirectIoEnabled());
         config.SetDirectIoAlign(features.GetDirectIoAlign());
 
+        config.SetGuestWritebackCacheEnabled(
+            features.GetGuestWritebackCacheEnabled());
+
         return std::make_shared<TFileSystemConfig>(config);
     }
 
@@ -931,6 +935,10 @@ class TFileSystemLoop final
         // e.g. left from a crash or smth, paranoid mode
         ResetSessionState(SessionThread->GetSession().Dump());
 
+        if (FileSystemConfig->GetGuestWritebackCacheEnabled()) {
+            conn->want |=  FUSE_CAP_WRITEBACK_CACHE;
+        }
+
         FileSystem->Init();
     }
 

diff --git a/cloud/filestore/public/api/protos/fs.proto b/cloud/filestore/public/api/protos/fs.proto
@@ -26,6 +26,7 @@ message TFileStoreFeatures
     uint32 DirectIoAlign = 11;
     bool TwoStageReadDisabledForHDD = 12;
     bool ThreeStageWriteDisabledForHDD = 13;
+    bool GuestWritebackCacheEnabled = 14;
 }
 
 message TFileStore

diff --git a/contrib/libs/fuse/include/fuse_common.h b/contrib/libs/fuse/include/fuse_common.h
@@ -100,6 +100,7 @@ struct fuse_file_info {
  * FUSE_CAP_SPLICE_MOVE: ability to move data to the fuse device with splice()
  * FUSE_CAP_SPLICE_READ: ability to use splice() to read from the fuse device
  * FUSE_CAP_IOCTL_DIR: ioctl support on directories
+ * FUSE_CAP_WRITEBACK_CACHE: writeback caching should be enabled
  */
 #define FUSE_CAP_ASYNC_READ	(1 << 0)
 #define FUSE_CAP_POSIX_LOCKS	(1 << 1)
@@ -112,6 +113,7 @@ struct fuse_file_info {
 #define FUSE_CAP_SPLICE_READ	(1 << 9)
 #define FUSE_CAP_FLOCK_LOCKS	(1 << 10)
 #define FUSE_CAP_IOCTL_DIR	(1 << 11)
+#define FUSE_CAP_WRITEBACK_CACHE	(1 << 16)
 
 /**
  * Ioctl flags

diff --git a/contrib/libs/fuse/include/fuse_kernel.h b/contrib/libs/fuse/include/fuse_kernel.h
@@ -211,6 +211,7 @@ struct fuse_file_lock {
 #define FUSE_BIG_WRITES		(1 << 5)
 #define FUSE_DONT_MASK		(1 << 6)
 #define FUSE_FLOCK_LOCKS	(1 << 10)
+#define FUSE_WRITEBACK_CACHE	(1 << 16)
 
 /**
  * CUSE INIT request/reply flags

diff --git a/contrib/libs/fuse/lib/fuse_lowlevel.c b/contrib/libs/fuse/lib/fuse_lowlevel.c
@@ -1856,6 +1856,9 @@ static void do_init(fuse_req_t req, fuse_ino_t nodeid, const void *inarg)
 		outarg.flags |= FUSE_DONT_MASK;
 	if (f->conn.want & FUSE_CAP_FLOCK_LOCKS)
 		outarg.flags |= FUSE_FLOCK_LOCKS;
+	if (f->conn.want & FUSE_CAP_WRITEBACK_CACHE)
+		outarg.flags |= FUSE_WRITEBACK_CACHE;
+
 	outarg.max_readahead = f->conn.max_readahead;
 	outarg.max_write = f->conn.max_write;
 	if (f->conn.proto_minor >= 13) {