- Reviving PR 191, to make FileSystem creation off actual path
- Streamline all filesystem access to HoodieTableMetaClient
- Hadoop Conf from Spark Context serialized & passed to executor code too
- Pick up env vars prefixed with HOODIE_ENV_ into Configuration object
- Cleanup usage of FSUtils.getFS, piggybacking off HoodieTableMetaClient.getFS
- Adding s3a to supported schemes & support escaping "." in env vars
- Tests use HoodieTestUtils.getDefaultHadoopConf
- Write with COW/MOR paths work fully
- Read with RO view works on both storages*
- Incremental view supported on COW
- Refactored out HoodieReadClient methods, to just contain key based access
- HoodieDataSourceHelpers class can be now used to construct inputs to datasource
- Tests in hoodie-client using new helpers and mechanisms
- Basic tests around save modes & insert/upserts (more to follow)
- Bumped up scala to 2.11, since 2.10 is deprecated & complains with scalatest
- Updated documentation to describe usage
- New sample app written using the DataSource API
- Merged all filter* and get* methods
- new constructor takes filestatus[]
- All existing tests pass
- FileGroup is all files that belong to a fileID within a partition
- FileSlice is a generation of data and log files, starting at a base commit
- clean/rollback/write paths covered by existing tests
- Snapshot copier fixed to copy metadata file also, and test fixed
- Existing tables need to be repaired by addition of metadata, before this can be rolled out
- Concurreny handled via taskID, failure recovery handled via renames
- Falls back to search 3 levels up
- Cli tool has command to add this to existing tables
- Introduce avro to save clean metadata with details about the last commit that was retained
- Save rollback metadata in the meta timeline
- Create savepoint metadata and add API to createSavepoint, deleteSavepoint and rollbackToSavepoint
- Savepointed commit should not be rolledback or cleaned or archived
- introduce cli commands to show, create and rollback to savepoints
- Write unit tests to test savepoints and rollbackToSavepoints
1. Create HoodieTable abstraction for commits and fileSystemView
2. HoodieMergeOnReadTable created
3. View is now always obtained from the table and the correct view based on the table type is returned
- Refactored timelines to be a single timeline for all active events and one for archived events. CommitTimeline and other timelines can be inferred by applying a filter on the activeTimelime
- Introduced HoodieInstant to abstract different types of action, commit time and if isInFlight
- Implemented other review comments
The following is the gist of changes done
- All low-level operation of creating a commit code was in HoodieClient which made it hard to share code if there was a compaction commit.
- HoodieTableMetadata contained a mix of metadata and filtering files. (Also few operations required FileSystem to be passed in because those were called from TaskExecutors and others had FileSystem as a global variable). Since merge-on-read requires a lot of that code, but will have to change slightly on how it operates on the metadata and how it filters the files. The two set of operation are split into HoodieTableMetaClient and TableFileSystemView.
- Everything (active commits, archived commits, cleaner log, save point log and in future delta and compaction commits) in HoodieTableMetaClient is a HoodieTimeline. Timeline is a series of instants, which has an in-built concept of inflight and completed commit markers.
- A timeline can be queries for ranges, contains and also use to create new datapoint (create a new commit etc). Commit (and all the above metadata) creation/deletion is streamlined in a timeline
- Multiple timelines can be merged into a single timeline, giving us an audit timeline to whatever happened in a hoodie dataset. This also helps with #55.
- Move to java 8 and introduce java 8 succinct syntax in refactored code
