Skip directory fsync for filesystem btrfs (facebook#8903)

Summary:
Directory fsync might be expensive on btrfs and it may not be needed.
Here are 4 directory fsync cases:
1. creating a new file: dir-fsync is not needed on btrfs, as long as the
   new file itself is synced.
2. renaming a file: dir-fsync is not needed if the renamed file is
   synced. So an API `FsyncAfterFileRename(filename, ...)` is provided
   to sync the file on btrfs. By default, it just calls dir-fsync.
3. deleting files: dir-fsync is forced by set
   `IOOptions.force_dir_fsync = true`
4. renaming multiple files (like backup and checkpoint): dir-fsync is
   forced, the same as above.

Pull Request resolved: facebook#8903

Test Plan: run tests on btrfs and non btrfs

Reviewed By: ajkr

Differential Revision: D30885059

Pulled By: jay-zhuang

fbshipit-source-id: dd2730b31580b0bcaedffc318a762d7dbf25de4a

HISTORY.md

@@ -20,6 +20,8 @@
 * Made FileSystem extend the Customizable class and added a CreateFromString method.  Implementations need to be registered with the ObjectRegistry and to implement a Name() method in order to be created via this method.
 * Clarified in API comments that RocksDB is not exception safe for callbacks and custom extensions. An exception propagating into RocksDB can lead to undefined behavior, including data loss, unreported corruption, deadlocks, and more.
 * Marked `WriteBufferManager` as `final` because it is not intended for extension.
+* Add API `FSDirectory::FsyncWithDirOptions()`, which provides extra information like directory fsync reason in `DirFsyncOptions`. File system like btrfs is using that to skip directory fsync for creating a new file, or when renaming a file, fsync the target file instead of the directory, which improves the `DB::Open()` speed by ~20%.
+* `DB::Open()` is not going be blocked by obsolete file purge if `DBOptions::avoid_unnecessary_blocking_io` is set to true.
 
 ## 6.26.0 (2021-10-20)
 ### Bug Fixes

Makefile

@@ -1926,6 +1926,9 @@ clipping_iterator_test: $(OBJ_DIR)/db/compaction/clipping_iterator_test.o $(TEST
 ribbon_bench: $(OBJ_DIR)/microbench/ribbon_bench.o $(LIBRARY)
 	$(AM_LINK)
 
+db_basic_bench: $(OBJ_DIR)/microbench/db_basic_bench.o $(LIBRARY)
+	$(AM_LINK)
+
 cache_reservation_manager_test: $(OBJ_DIR)/cache/cache_reservation_manager_test.o $(TEST_LIBRARY) $(LIBRARY)
 	$(AM_LINK)
 #-------------------------------------------------

db/compaction/compaction_job.cc

@@ -763,12 +763,16 @@ Status CompactionJob::Run() {
     constexpr IODebugContext* dbg = nullptr;
 
     if (output_directory_) {
-      io_s = output_directory_->Fsync(IOOptions(), dbg);
+      io_s = output_directory_->FsyncWithDirOptions(
+          IOOptions(), dbg,
+          DirFsyncOptions(DirFsyncOptions::FsyncReason::kNewFileSynced));
     }
 
     if (io_s.ok() && wrote_new_blob_files && blob_output_directory_ &&
         blob_output_directory_ != output_directory_) {
-      io_s = blob_output_directory_->Fsync(IOOptions(), dbg);
+      io_s = blob_output_directory_->FsyncWithDirOptions(
+          IOOptions(), dbg,
+          DirFsyncOptions(DirFsyncOptions::FsyncReason::kNewFileSynced));
     }
   }
   if (io_status_.ok()) {
@@ -2442,7 +2446,8 @@ Status CompactionServiceCompactionJob::Run() {
     constexpr IODebugContext* dbg = nullptr;
 
     if (output_directory_) {
-      io_s = output_directory_->Fsync(IOOptions(), dbg);
+      io_s = output_directory_->FsyncWithDirOptions(IOOptions(), dbg,
+                                                    DirFsyncOptions());
     }
   }
   if (io_status_.ok()) {

db/db_impl/db_impl.cc

@@ -1389,7 +1389,9 @@ Status DBImpl::SyncWAL() {
     IOStatusCheck(io_s);
   }
   if (status.ok() && need_log_dir_sync) {
-    status = directories_.GetWalDir()->Fsync(IOOptions(), nullptr);
+    status = directories_.GetWalDir()->FsyncWithDirOptions(
+        IOOptions(), nullptr,
+        DirFsyncOptions(DirFsyncOptions::FsyncReason::kNewFileSynced));
   }
   TEST_SYNC_POINT("DBWALTest::SyncWALNotWaitWrite:2");
 
@@ -4288,6 +4290,14 @@ Status DBImpl::RenameTempFileToOptionsFile(const std::string& file_name) {
   if (s.ok()) {
     // Retry if the file name happen to conflict with an existing one.
     s = GetEnv()->RenameFile(file_name, options_file_name);
+    std::unique_ptr<FSDirectory> dir_obj;
+    if (s.ok()) {
+      s = fs_->NewDirectory(GetName(), IOOptions(), &dir_obj, nullptr);
+    }
+    if (s.ok()) {
+      s = dir_obj->FsyncWithDirOptions(IOOptions(), nullptr,
+                                       DirFsyncOptions(options_file_name));
+    }
   }
   if (s.ok()) {
     InstrumentedMutexLock l(&mutex_);

db/db_impl/db_impl.h

@@ -974,6 +974,9 @@ class DBImpl : public DB {
   // is only for the special test of CancelledCompactions
   Status TEST_WaitForCompact(bool waitUnscheduled = false);
 
+  // Wait for any background purge
+  Status TEST_WaitForPurge();
+
   // Get the background error status
   Status TEST_GetBGError();
 

db/db_impl/db_impl_compaction_flush.cc

@@ -120,7 +120,9 @@ IOStatus DBImpl::SyncClosedLogs(JobContext* job_context) {
       }
     }
     if (io_s.ok()) {
-      io_s = directories_.GetWalDir()->Fsync(IOOptions(), nullptr);
+      io_s = directories_.GetWalDir()->FsyncWithDirOptions(
+          IOOptions(), nullptr,
+          DirFsyncOptions(DirFsyncOptions::FsyncReason::kNewFileSynced));
     }
 
     mutex_.Lock();
@@ -532,7 +534,9 @@ Status DBImpl::AtomicFlushMemTablesToOutputFiles(
     // Sync on all distinct output directories.
     for (auto dir : distinct_output_dirs) {
       if (dir != nullptr) {
-        Status error_status = dir->Fsync(IOOptions(), nullptr);
+        Status error_status = dir->FsyncWithDirOptions(
+            IOOptions(), nullptr,
+            DirFsyncOptions(DirFsyncOptions::FsyncReason::kNewFileSynced));
         if (!error_status.ok()) {
           s = error_status;
           break;

db/db_impl/db_impl_debug.cc

@@ -184,6 +184,14 @@ Status DBImpl::TEST_WaitForCompact(bool wait_unscheduled) {
   return error_handler_.GetBGError();
 }
 
+Status DBImpl::TEST_WaitForPurge() {
+  InstrumentedMutexLock l(&mutex_);
+  while (bg_purge_scheduled_ && error_handler_.GetBGError().ok()) {
+    bg_cv_.Wait();
+  }
+  return error_handler_.GetBGError();
+}
+
 Status DBImpl::TEST_GetBGError() {
   InstrumentedMutexLock l(&mutex_);
   return error_handler_.GetBGError();

db/db_impl/db_impl_files.cc

@@ -655,7 +655,13 @@ void DBImpl::DeleteObsoleteFiles() {
 
   mutex_.Unlock();
   if (job_context.HaveSomethingToDelete()) {
-    PurgeObsoleteFiles(job_context);
+    bool defer_purge = immutable_db_options_.avoid_unnecessary_blocking_io;
+    PurgeObsoleteFiles(job_context, defer_purge);
+    if (defer_purge) {
+      mutex_.Lock();
+      SchedulePurge();
+      mutex_.Unlock();
+    }
   }
   job_context.Clean();
   mutex_.Lock();

db/db_impl/db_impl_open.cc

@@ -1694,10 +1694,6 @@ Status DBImpl::Open(const DBOptions& db_options, const std::string& dbname,
       if (impl->two_write_queues_) {
         impl->log_write_mutex_.Unlock();
       }
-
-      impl->DeleteObsoleteFiles();
-      s = impl->directories_.GetDbDir()->Fsync(IOOptions(), nullptr);
-      TEST_SYNC_POINT("DBImpl::Open:AfterDeleteFilesAndSyncDir");
     }
     if (s.ok()) {
       // In WritePrepared there could be gap in sequence numbers. This breaks
@@ -1770,6 +1766,8 @@ Status DBImpl::Open(const DBOptions& db_options, const std::string& dbname,
 
     *dbptr = impl;
     impl->opened_successfully_ = true;
+    impl->DeleteObsoleteFiles();
+    TEST_SYNC_POINT("DBImpl::Open:AfterDeleteFiles");
     impl->MaybeScheduleFlushOrCompaction();
   } else {
     persist_options_status.PermitUncheckedError();

db/db_impl/db_impl_write.cc

@@ -1142,7 +1142,9 @@ IOStatus DBImpl::WriteToWAL(const WriteThread::WriteGroup& write_group,
       // We only sync WAL directory the first time WAL syncing is
       // requested, so that in case users never turn on WAL sync,
       // we can avoid the disk I/O in the write code path.
-      io_s = directories_.GetWalDir()->Fsync(IOOptions(), nullptr);
+      io_s = directories_.GetWalDir()->FsyncWithDirOptions(
+          IOOptions(), nullptr,
+          DirFsyncOptions(DirFsyncOptions::FsyncReason::kNewFileSynced));
     }
   }
 

db/db_secondary_test.cc

@@ -690,7 +690,7 @@ TEST_F(DBSecondaryTest, SwitchToNewManifestDuringOpen) {
   SyncPoint::GetInstance()->LoadDependency(
       {{"ReactiveVersionSet::MaybeSwitchManifest:AfterGetCurrentManifestPath:0",
         "VersionSet::ProcessManifestWrites:BeforeNewManifest"},
-       {"DBImpl::Open:AfterDeleteFilesAndSyncDir",
+       {"DBImpl::Open:AfterDeleteFiles",
         "ReactiveVersionSet::MaybeSwitchManifest:AfterGetCurrentManifestPath:"
         "1"}});
   SyncPoint::GetInstance()->EnableProcessing();

db/deletefile_test.cc

@@ -117,9 +117,15 @@ class DeleteFileTest : public DBTestBase {
         manifest_cnt += (type == kDescriptorFile);
       }
     }
-    ASSERT_EQ(required_log, log_cnt);
-    ASSERT_EQ(required_sst, sst_cnt);
-    ASSERT_EQ(required_manifest, manifest_cnt);
+    if (required_log >= 0) {
+      ASSERT_EQ(required_log, log_cnt);
+    }
+    if (required_sst >= 0) {
+      ASSERT_EQ(required_sst, sst_cnt);
+    }
+    if (required_manifest >= 0) {
+      ASSERT_EQ(required_manifest, manifest_cnt);
+    }
   }
 
   static void DoSleep(void* arg) {
@@ -264,6 +270,41 @@ TEST_F(DeleteFileTest, BackgroundPurgeIteratorTest) {
   CheckFileTypeCounts(dbname_, 0, 1, 1);
 }
 
+TEST_F(DeleteFileTest, PurgeDuringOpen) {
+  Options options = CurrentOptions();
+  CheckFileTypeCounts(dbname_, -1, 0, -1);
+  Close();
+  std::unique_ptr<WritableFile> file;
+  ASSERT_OK(options.env->NewWritableFile(dbname_ + "/000002.sst", &file,
+                                         EnvOptions()));
+  ASSERT_OK(file->Close());
+  CheckFileTypeCounts(dbname_, -1, 1, -1);
+  options.avoid_unnecessary_blocking_io = false;
+  options.create_if_missing = false;
+  Reopen(options);
+  CheckFileTypeCounts(dbname_, -1, 0, -1);
+  Close();
+
+  // test background purge
+  options.avoid_unnecessary_blocking_io = true;
+  options.create_if_missing = false;
+  ASSERT_OK(options.env->NewWritableFile(dbname_ + "/000002.sst", &file,
+                                         EnvOptions()));
+  ASSERT_OK(file->Close());
+  CheckFileTypeCounts(dbname_, -1, 1, -1);
+  SyncPoint::GetInstance()->DisableProcessing();
+  SyncPoint::GetInstance()->ClearAllCallBacks();
+  SyncPoint::GetInstance()->LoadDependency(
+      {{"DeleteFileTest::PurgeDuringOpen:1", "DBImpl::BGWorkPurge:start"}});
+  SyncPoint::GetInstance()->EnableProcessing();
+  Reopen(options);
+  // the obsolete file is not deleted until the background purge job is ran
+  CheckFileTypeCounts(dbname_, -1, 1, -1);
+  TEST_SYNC_POINT("DeleteFileTest::PurgeDuringOpen:1");
+  ASSERT_OK(dbfull()->TEST_WaitForPurge());
+  CheckFileTypeCounts(dbname_, -1, 0, -1);
+}
+
 TEST_F(DeleteFileTest, BackgroundPurgeCFDropTest) {
   Options options = CurrentOptions();
   SetOptions(&options);
@@ -310,16 +351,18 @@ TEST_F(DeleteFileTest, BackgroundPurgeCFDropTest) {
     do_test(false);
   }
 
+  options.avoid_unnecessary_blocking_io = true;
+  options.create_if_missing = false;
+  Reopen(options);
+  ASSERT_OK(dbfull()->TEST_WaitForPurge());
+
   SyncPoint::GetInstance()->DisableProcessing();
   SyncPoint::GetInstance()->ClearAllCallBacks();
   SyncPoint::GetInstance()->LoadDependency(
       {{"DeleteFileTest::BackgroundPurgeCFDropTest:1",
         "DBImpl::BGWorkPurge:start"}});
   SyncPoint::GetInstance()->EnableProcessing();
 
-  options.avoid_unnecessary_blocking_io = true;
-  options.create_if_missing = false;
-  Reopen(options);
   {
     SCOPED_TRACE("avoid_unnecessary_blocking_io = true");
     do_test(true);

db/external_sst_file_ingestion_job.cc

@@ -168,7 +168,9 @@ Status ExternalSstFileIngestionJob::Prepare(
   TEST_SYNC_POINT("ExternalSstFileIngestionJob::BeforeSyncDir");
   if (status.ok()) {
     for (auto path_id : ingestion_path_ids) {
-      status = directories_->GetDataDir(path_id)->Fsync(IOOptions(), nullptr);
+      status = directories_->GetDataDir(path_id)->FsyncWithDirOptions(
+          IOOptions(), nullptr,
+          DirFsyncOptions(DirFsyncOptions::FsyncReason::kNewFileSynced));
       if (!status.ok()) {
         ROCKS_LOG_WARN(db_options_.info_log,
                        "Failed to sync directory %" ROCKSDB_PRIszt

db/flush_job.cc

@@ -937,7 +937,9 @@ Status FlushJob::WriteLevel0Table() {
                    meta_.marked_for_compaction ? " (needs compaction)" : "");
 
     if (s.ok() && output_file_directory_ != nullptr && sync_output_directory_) {
-      s = output_file_directory_->Fsync(IOOptions(), nullptr);
+      s = output_file_directory_->FsyncWithDirOptions(
+          IOOptions(), nullptr,
+          DirFsyncOptions(DirFsyncOptions::FsyncReason::kNewFileSynced));
     }
     TEST_SYNC_POINT_CALLBACK("FlushJob::WriteLevel0Table", &mems_);
     db_mutex_->Lock();

env/composite_env.cc

@@ -272,7 +272,7 @@ class CompositeDirectoryWrapper : public Directory {
   Status Fsync() override {
     IOOptions io_opts;
     IODebugContext dbg;
-    return target_->Fsync(io_opts, &dbg);
+    return target_->FsyncWithDirOptions(io_opts, &dbg, DirFsyncOptions());
   }
   size_t GetUniqueId(char* id, size_t max_size) const override {
     return target_->GetUniqueId(id, max_size);

env/file_system.cc

@@ -241,4 +241,16 @@ std::string FileSystemWrapper::SerializeOptions(
   }
 }
 #endif  // ROCKSDB_LITE
+
+DirFsyncOptions::DirFsyncOptions() { reason = kDefault; }
+
+DirFsyncOptions::DirFsyncOptions(std::string file_renamed_new_name) {
+  reason = kFileRenamed;
+  renamed_new_name = file_renamed_new_name;
+}
+
+DirFsyncOptions::DirFsyncOptions(FsyncReason fsync_reason) {
+  assert(fsync_reason != kFileRenamed);
+  reason = fsync_reason;
+}
 }  // namespace ROCKSDB_NAMESPACE

env/io_posix.cc

@@ -1534,17 +1534,61 @@ PosixMemoryMappedFileBuffer::~PosixMemoryMappedFileBuffer() {
 /*
  * PosixDirectory
  */
+#if !defined(BTRFS_SUPER_MAGIC)
+// The magic number for BTRFS is fixed, if it's not defined, define it here
+#define BTRFS_SUPER_MAGIC 0x9123683E
+#endif
+PosixDirectory::PosixDirectory(int fd) : fd_(fd) {
+  is_btrfs_ = false;
+#ifdef OS_LINUX
+  struct statfs buf;
+  int ret = fstatfs(fd, &buf);
+  is_btrfs_ = (ret == 0 && buf.f_type == BTRFS_SUPER_MAGIC);
+#endif
+}
 
 PosixDirectory::~PosixDirectory() { close(fd_); }
 
-IOStatus PosixDirectory::Fsync(const IOOptions& /*opts*/,
-                               IODebugContext* /*dbg*/) {
+IOStatus PosixDirectory::Fsync(const IOOptions& opts, IODebugContext* dbg) {
+  return FsyncWithDirOptions(opts, dbg, DirFsyncOptions());
+}
+
+IOStatus PosixDirectory::FsyncWithDirOptions(
+    const IOOptions& /*opts*/, IODebugContext* /*dbg*/,
+    const DirFsyncOptions& dir_fsync_options) {
+  IOStatus s = IOStatus::OK();
 #ifndef OS_AIX
+  if (is_btrfs_) {
+    // skip dir fsync for new file creation, which is not needed for btrfs
+    if (dir_fsync_options.reason == DirFsyncOptions::kNewFileSynced) {
+      return s;
+    }
+    // skip dir fsync for renaming file, only need to sync new file
+    if (dir_fsync_options.reason == DirFsyncOptions::kFileRenamed) {
+      std::string new_name = dir_fsync_options.renamed_new_name;
+      assert(!new_name.empty());
+      int fd;
+      do {
+        IOSTATS_TIMER_GUARD(open_nanos);
+        fd = open(new_name.c_str(), O_RDONLY);
+      } while (fd < 0 && errno == EINTR);
+      if (fd < 0) {
+        s = IOError("While open renaming file", new_name, errno);
+      } else if (fsync(fd) < 0) {
+        s = IOError("While fsync renaming file", new_name, errno);
+      }
+      if (close(fd) < 0) {
+        s = IOError("While closing file after fsync", new_name, errno);
+      }
+      return s;
+    }
+    // fallback to dir-fsync for kDefault, kDirRenamed and kFileDeleted
+  }
   if (fsync(fd_) == -1) {
-    return IOError("While fsync", "a directory", errno);
+    s = IOError("While fsync", "a directory", errno);
   }
 #endif
-  return IOStatus::OK();
+  return s;
 }
 }  // namespace ROCKSDB_NAMESPACE
 #endif

env/io_posix.h

@@ -391,12 +391,17 @@ struct PosixMemoryMappedFileBuffer : public MemoryMappedFileBuffer {
 
 class PosixDirectory : public FSDirectory {
  public:
-  explicit PosixDirectory(int fd) : fd_(fd) {}
+  explicit PosixDirectory(int fd);
   ~PosixDirectory();
   virtual IOStatus Fsync(const IOOptions& opts, IODebugContext* dbg) override;
 
+  virtual IOStatus FsyncWithDirOptions(
+      const IOOptions&, IODebugContext*,
+      const DirFsyncOptions& dir_fsync_options) override;
+
  private:
   int fd_;
+  bool is_btrfs_;
 };
 
 }  // namespace ROCKSDB_NAMESPACE