Tracking Issue for StorageEngine::open_region()

[evenyag]

This is a tracking issue for implementing the StorageEngine::open_region() method.

Tasks

• Adds RegionImpl::open() to open and recover a region
  • Code skeleton @evenyag feat: Engine::open_region code skeleton #120
  • RegionImpl::recover_from_manifest() Loads metadata from manifest and construct Version @killme2008 feat: impl recovering version from manifest for region #127
    • RegionManifest::load() should also load VersionEdit
    • RegionManifest::update() should persist metadata of Version(flushed_sequence)
  • Replay data from wal
    • Reads logstore and decode wal entry, implement Wal::decode_entry() @fengjiachun feat: decode WAL entry #123
    • RegionWriter::replay() replay batch into memtable, maintains Version @evenyag feat: Implements replay #135
  • Adds an option to OpenOptions to create region if it is absent (optional feature)
• Manifest uses prev_version to construct the metadata change list (optional feature) @killme2008
• Remove id from Namespace of logstore feat: Engine::open_region code skeleton #120
  • Also remove region_id from Wal::new() feat: Engine::open_region code skeleton #120
• Concurrent control of opening/creating region feat: Engine::open_region code skeleton #120
  • Related issue is potential concurrency hazard in create_region #113
  • Maybe we should introduce state (enum) for region
• Implements LogStore::read() @v0y4g3r feat: impl Logstore::read by LogFile::create_stream #124
• Add tests for open @evenyag feat: Implements replay #135

Related Tasks

• We may also need to add a TableEngine::open_table() method to open all regions of this table MitoEngine needs an open_table API #117
• Support dynamic polymorphism for TableEngine refactor: Make TableEngine object safe #119