add config to disable upload page data when do checkpoint (#7753)

ref #6827

dbms/src/Interpreters/Settings.h

@@ -234,6 +234,7 @@ struct Settings
     M(SettingInt64, dt_compression_level, 1, "The compression level.")                                                                                                                                                                  \
     \
     M(SettingInt64, remote_checkpoint_interval_seconds, 30, "The interval of uploading checkpoint to the remote store. Unit is second.")                                                                                                \
+    M(SettingBool, remote_checkpoint_only_upload_manifest, true, "Only upload manifest data when uploading checkpoint")                                                                                                                 \
     M(SettingInt64, remote_gc_method, 1, "The method of running GC task on the remote store. 1 - lifecycle, 2 - scan.")                                                                                                                 \
     M(SettingInt64, remote_gc_interval_seconds, 3600, "The interval of running GC task on the remote store. Unit is second.")                                                                                                           \
     M(SettingInt64, remote_gc_verify_consistency, 0, "Verify the consistenct of valid locks when doing GC")                                                                                                                             \

dbms/src/Storages/Page/V3/CheckpointFile/CPFilesWriter.cpp

@@ -19,6 +19,7 @@
 #include <Storages/Page/V3/Universal/UniversalPageIdFormatImpl.h>
 #include <fmt/core.h>
 
+#include <memory>
 #include <unordered_map>
 
 namespace DB::PS::V3
@@ -67,7 +68,8 @@ void CPFilesWriter::writePrefix(const CPFilesWriter::PrefixInfo & info)
 
 CPDataDumpStats CPFilesWriter::writeEditsAndApplyCheckpointInfo(
     universal::PageEntriesEdit & edits,
-    const CPFilesWriter::CompactOptions & options)
+    const CPFilesWriter::CompactOptions & options,
+    bool manifest_only)
 {
     RUNTIME_CHECK_MSG(write_stage == WriteStage::WritingEdits, "unexpected write stage {}", magic_enum::enum_name(write_stage));
 
@@ -132,6 +134,11 @@ CPDataDumpStats CPFilesWriter::writeEditsAndApplyCheckpointInfo(
         }
 
         assert(rec_edit.type == EditRecordType::VAR_ENTRY);
+        // No need to read and write page data for the manifest only checkpoint.
+        if (manifest_only)
+        {
+            continue;
+        }
         bool is_compaction = false;
         if (rec_edit.entry.checkpoint_info.has_value())
         {

dbms/src/Storages/Page/V3/CheckpointFile/CPFilesWriter.h

@@ -96,7 +96,8 @@ class CPFilesWriter : private boost::noncopyable
     };
     CPDataDumpStats writeEditsAndApplyCheckpointInfo(
         universal::PageEntriesEdit & edit,
-        const CompactOptions & options = CompactOptions(false));
+        const CompactOptions & options = CompactOptions(false),
+        bool manifest_only = false);
 
     /**
      * This function must be called, and must be called last, after other `writeXxx`.

dbms/src/Storages/Page/V3/PageEntriesEdit.cpp

@@ -28,11 +28,14 @@ CheckpointProto::EditRecord PageEntriesEdit<UniversalPageId>::EditRecord::toProt
     proto_edit.set_version_epoch(version.epoch);
     if (type == EditRecordType::VAR_ENTRY)
     {
-        RUNTIME_CHECK(entry.checkpoint_info.has_value());
         proto_edit.set_entry_size(entry.size);
         proto_edit.set_entry_tag(entry.tag);
         proto_edit.set_entry_checksum(entry.checksum);
-        proto_edit.mutable_entry_location()->CopyFrom(entry.checkpoint_info.data_location.toProto());
+        // uploading page data may be disabled
+        if (entry.checkpoint_info.has_value())
+        {
+            proto_edit.mutable_entry_location()->CopyFrom(entry.checkpoint_info.data_location.toProto());
+        }
         for (const auto & [offset, checksum] : entry.field_offsets)
         {
             proto_edit.add_entry_fields_offset(offset);

dbms/src/Storages/Page/V3/Universal/UniversalPageStorage.cpp

@@ -449,10 +449,13 @@ PS::V3::CPDataDumpStats UniversalPageStorage::dumpIncrementalCheckpoint(const Un
     if (options.override_sequence)
         sequence = options.override_sequence.value();
 
+
+    // The output of `PageDirectory::dumpSnapshotToEdit` may contain page ids which are logically deleted but have not been gced yet.
+    // These page ids may be GC-ed when dumping snapshot, so we cannot read data of these page ids.
+    // So we create a clean temp page_directory here and use it to dump edits with all visible page ids for `snap`.
+    // But if we just upload manifest without reading page data, we can skip this step.
+    if (!options.only_upload_manifest)
     {
-        // The output of `PageDirectory::dumpSnapshotToEdit` may contain page ids which are logically deleted but have not been gced yet.
-        // These page ids may be GC-ed when dumping snapshot, so we cannot read data of these page ids.
-        // So we create a clean temp page_directory here and use it to dump edits with all visible page ids for `snap`.
         PS::V3::universal::PageDirectoryFactory factory;
         auto temp_page_directory = factory.dangerouslyCreateFromEditWithoutWAL(fmt::format("{}_{}", storage_name, sequence), edit_from_mem);
         edit_from_mem = temp_page_directory->dumpSnapshotToEdit();
@@ -494,7 +497,8 @@ PS::V3::CPDataDumpStats UniversalPageStorage::dumpIncrementalCheckpoint(const Un
     // get the remote file ids that need to be compacted
     const auto checkpoint_dump_stats = writer->writeEditsAndApplyCheckpointInfo(
         edit_from_mem,
-        compact_opts);
+        compact_opts,
+        options.only_upload_manifest);
     auto data_file_paths = writer->writeSuffix();
     writer.reset();
     auto dump_data_seconds = sw.elapsedMillisecondsFromLastTime() / 1000.0;
@@ -521,6 +525,7 @@ PS::V3::CPDataDumpStats UniversalPageStorage::dumpIncrementalCheckpoint(const Un
     {
         // Copy back the checkpoint info to the current PageStorage.
         // New checkpoint infos are attached in `writeEditsAndApplyCheckpointInfo`.
+        RUNTIME_CHECK(!options.only_upload_manifest);
         page_directory->copyCheckpointInfoFromEdit(edit_from_mem);
     }
     auto copy_checkpoint_info_seconds = sw.elapsedMillisecondsFromLastTime() / 1000.0;

dbms/src/Storages/Page/V3/Universal/UniversalPageStorage.h

@@ -203,6 +203,11 @@ class UniversalPageStorage final
          */
         const std::function<std::unordered_set<String>()> compact_getter = nullptr;
 
+        /**
+         * Only upload the manifest file.
+         */
+        bool only_upload_manifest = false;
+
         UInt64 max_data_file_size = 256 * 1024 * 1024; // 256MB
         UInt64 max_edit_records_per_part = 100000;
     };

dbms/src/Storages/Page/V3/Universal/UniversalPageStorageService.cpp

@@ -273,6 +273,7 @@ bool UniversalPageStorageService::uploadCheckpointImpl(
             .remote_store = remote_store,
             .log = log,
         },
+        .only_upload_manifest = settings.remote_checkpoint_only_upload_manifest,
     };
 
     const auto write_stats = uni_page_storage->dumpIncrementalCheckpoint(opts);

dbms/src/Storages/Page/V3/Universal/tests/gtest_checkpoint.cpp

@@ -953,12 +953,23 @@ class UniversalPageStorageServiceCheckpointTest : public DB::base::TiFlashStorag
     void SetUp() override
     {
         TiFlashStorageTestBasic::SetUp();
+        auto & global_context = DB::tests::TiFlashTestEnv::getGlobalContext();
+        auto & settings = global_context.getSettingsRef();
+        old_remote_checkpoint_only_upload_manifest = settings.remote_checkpoint_only_upload_manifest;
+        settings.remote_checkpoint_only_upload_manifest = false;
         uni_ps_service = newService();
         log = Logger::get("UniversalPageStorageServiceCheckpointTest");
         s3_client = S3::ClientFactory::instance().sharedTiFlashClient();
         ASSERT_TRUE(::DB::tests::TiFlashTestEnv::createBucketIfNotExist(*s3_client));
     }
 
+    void TearDown() override
+    {
+        auto & global_context = DB::tests::TiFlashTestEnv::getGlobalContext();
+        auto & settings = global_context.getSettingsRef();
+        settings.remote_checkpoint_only_upload_manifest = old_remote_checkpoint_only_upload_manifest;
+    }
+
     static UniversalPageStorageServicePtr newService()
     {
         auto path = getTemporaryPath();
@@ -1003,6 +1014,8 @@ class UniversalPageStorageServiceCheckpointTest : public DB::base::TiFlashStorag
     UInt64 tag = 0;
     UInt64 store_id = 2;
 
+    bool old_remote_checkpoint_only_upload_manifest;
+
     LoggerPtr log;
 };