sstables: delete_with_pending_deletion_log: batch sync_directory

When deleting multiple sstables with the same prefix
the deletion atomicity is ensured by the pending_delete_log file,
so if scylla crashes in the middle, deletions will be replyed on
restart.

Therefore, we don't have to ensure atomicity of each individual
`unlink`.  We just need to sync the directory once, before
removing the pending_delete_log file.

Signed-off-by: Benny Halevy <bhalevy@scylladb.com>

Closes #14967

sstables/sstable_directory.cc

@@ -527,9 +527,11 @@ future<> sstable_directory::delete_with_pending_deletion_log(std::vector<shared_
         }
 
         parallel_for_each(ssts, [] (shared_sstable sst) {
-            return sst->unlink();
+            return sst->unlink(sstables::storage::sync_dir::no);
         }).get();
 
+        sync_directory(first->_storage->prefix()).get();
+
         // Once all sstables are deleted, the log file can be removed.
         // Note: the log file will be removed also if unlink failed to remove
         // any sstable and ignored the error.

sstables/sstables.cc

@@ -2609,7 +2609,7 @@ std::optional<std::pair<uint64_t, uint64_t>> sstable::get_sample_indexes_for_ran
     return std::nullopt;
 }
 
-future<> remove_by_toc_name(sstring sstable_toc_name) {
+future<> remove_by_toc_name(sstring sstable_toc_name, storage::sync_dir sync) {
     sstring prefix = sstable_toc_name.substr(0, sstable_toc_name.size() - sstable_version_constants::TOC_SUFFIX.size());
     sstring new_toc_name = prefix + sstable_version_constants::TEMPORARY_TOC_SUFFIX;
 
@@ -2617,7 +2617,9 @@ future<> remove_by_toc_name(sstring sstable_toc_name) {
     if (co_await sstable_io_check(sstable_write_error_handler, file_exists, sstable_toc_name)) {
         // If new_toc_name exists it will be atomically replaced.  See rename(2)
         co_await sstable_io_check(sstable_write_error_handler, rename_file, sstable_toc_name, new_toc_name);
-        co_await sstable_io_check(sstable_write_error_handler, sync_directory, parent_path(new_toc_name));
+        if (sync) {
+            co_await sstable_io_check(sstable_write_error_handler, sync_directory, parent_path(new_toc_name));
+        }
     } else {
         if (!co_await sstable_io_check(sstable_write_error_handler, file_exists, new_toc_name)) {
             sstlog.warn("Unable to delete {} because it doesn't exist.", sstable_toc_name);
@@ -2653,7 +2655,9 @@ future<> remove_by_toc_name(sstring sstable_toc_name) {
             sstlog.debug("Forgiving ENOENT when deleting file {}", fname);
         }
     });
-    co_await sstable_io_check(sstable_write_error_handler, sync_directory, parent_path(new_toc_name));
+    if (sync) {
+        co_await sstable_io_check(sstable_write_error_handler, sync_directory, parent_path(new_toc_name));
+    }
     co_await sstable_io_check(sstable_write_error_handler, remove_file, new_toc_name);
 }
 
@@ -2791,8 +2795,8 @@ utils::hashed_key sstable::make_hashed_key(const schema& s, const partition_key&
 }
 
 future<>
-sstable::unlink() noexcept {
-    auto remove_fut = _storage->wipe(*this);
+sstable::unlink(storage::sync_dir sync) noexcept {
+    auto remove_fut = _storage->wipe(*this, sync);
 
     try {
         co_await get_large_data_handler().maybe_delete_large_data_entries(shared_from_this());

sstables/sstables.hh

@@ -388,7 +388,9 @@ public:
 
     // Delete the sstable by unlinking all sstable files
     // Ignores all errors.
-    future<> unlink() noexcept;
+    // Caller may pass sync_dir::no for batching multiple deletes in the same directory,
+    // and make sure the directory is sync'ed on or after the last call.
+    future<> unlink(storage::sync_dir sync = storage::sync_dir::yes) noexcept;
 
     db::large_data_handler& get_large_data_handler() {
         return _large_data_handler;
@@ -958,6 +960,8 @@ public:
 future<> remove_table_directory_if_has_no_snapshots(fs::path table_dir);
 
 // similar to sstable::unlink, but works on a TOC file name
-future<> remove_by_toc_name(sstring sstable_toc_name);
+// Caller may pass sync_dir::no for batching multiple deletes in the same directory,
+// and make sure the directory is sync'ed on or after the last call.
+future<> remove_by_toc_name(sstring sstable_toc_name, storage::sync_dir sync = storage::sync_dir::yes);
 
 } // namespace sstables

sstables/storage.cc

@@ -62,7 +62,7 @@ public:
     virtual future<> change_state(const sstable& sst, sstring to, generation_type generation, delayed_commit_changes* delay) override;
     // runs in async context
     virtual void open(sstable& sst) override;
-    virtual future<> wipe(const sstable& sst) noexcept override;
+    virtual future<> wipe(const sstable& sst, sync_dir) noexcept override;
     virtual future<file> open_component(const sstable& sst, component_type type, open_flags flags, file_open_options options, bool check_integrity) override;
     virtual future<data_sink> make_data_or_index_sink(sstable& sst, component_type type) override;
     virtual future<data_sink> make_component_sink(sstable& sst, component_type type, open_flags oflags, file_output_stream_options options) override;
@@ -387,7 +387,7 @@ future<> filesystem_storage::change_state(const sstable& sst, sstring to, genera
     co_await move(sst, path.native(), std::move(new_generation), delay_commit);
 }
 
-future<> filesystem_storage::wipe(const sstable& sst) noexcept {
+future<> filesystem_storage::wipe(const sstable& sst, sync_dir sync) noexcept {
     // We must be able to generate toc_filename()
     // in order to delete the sstable.
     // Running out of memory here will terminate.
@@ -397,7 +397,7 @@ future<> filesystem_storage::wipe(const sstable& sst) noexcept {
     }();
 
     try {
-        co_await remove_by_toc_name(name);
+        co_await remove_by_toc_name(name, sync);
     } catch (...) {
         // Log and ignore the failure since there is nothing much we can do about it at this point.
         // a. Compaction will retry deleting the sstable in the next pass, and
@@ -446,7 +446,7 @@ public:
     virtual future<> change_state(const sstable& sst, sstring to, generation_type generation, delayed_commit_changes* delay) override;
     // runs in async context
     virtual void open(sstable& sst) override;
-    virtual future<> wipe(const sstable& sst) noexcept override;
+    virtual future<> wipe(const sstable& sst, sync_dir) noexcept override;
     virtual future<file> open_component(const sstable& sst, component_type type, open_flags flags, file_open_options options, bool check_integrity) override;
     virtual future<data_sink> make_data_or_index_sink(sstable& sst, component_type type) override;
     virtual future<data_sink> make_component_sink(sstable& sst, component_type type, open_flags oflags, file_output_stream_options options) override;
@@ -518,7 +518,7 @@ future<> s3_storage::change_state(const sstable& sst, sstring to, generation_typ
     co_await coroutine::return_exception(std::runtime_error("Moving S3 objects not implemented"));
 }
 
-future<> s3_storage::wipe(const sstable& sst) noexcept {
+future<> s3_storage::wipe(const sstable& sst, sync_dir) noexcept {
     auto& sys_ks = sst.manager().system_keyspace();
 
     co_await sys_ks.sstables_registry_update_entry_status(_location, sst.generation(), status_removing);

sstables/storage.hh

@@ -48,13 +48,14 @@ public:
     virtual ~storage() {}
 
     using absolute_path = bool_class<class absolute_path_tag>; // FIXME -- should go away eventually
+    using sync_dir = bool_class<struct sync_dir_tag>; // meaningful only to filesystem storage
 
     virtual future<> seal(const sstable& sst) = 0;
     virtual future<> snapshot(const sstable& sst, sstring dir, absolute_path abs) const = 0;
     virtual future<> change_state(const sstable& sst, sstring to, generation_type generation, delayed_commit_changes* delay) = 0;
     // runs in async context
     virtual void open(sstable& sst) = 0;
-    virtual future<> wipe(const sstable& sst) noexcept = 0;
+    virtual future<> wipe(const sstable& sst, sync_dir) noexcept = 0;
     virtual future<file> open_component(const sstable& sst, component_type type, open_flags flags, file_open_options options, bool check_integrity) = 0;
     virtual future<data_sink> make_data_or_index_sink(sstable& sst, component_type type) = 0;
     virtual future<data_sink> make_component_sink(sstable& sst, component_type type, open_flags oflags, file_output_stream_options options) = 0;