- Remove hive files from old tasks
- Recover from unexpected results from tasks
- Handle "create table if not exists" statements
- Improve mechanism to verify query rerun completeness
- Handle case where table-scan task finishes after resume is done
- Avoid sharing memory context between old and new tasks
- Union sources produce same number of markers
- Exchange operator will be blocked waiting for noMoreSplits to be set
- Made noMoreSplits call happen early by scheduling all tasks with marker 0
- QuerySnapshotManager record snapshot initiated for accurate cleanup of snapshot subfiles
- Fix typo in driver logs
- Update HashBuilder comment
- Consider restored pages in ExchangeOperator isBlocked()
- Remove special handling in MergeOperator isBlocked()
- Fix issue where shared marker is updated with different origins
- Avoid too frequent resumes
- Avoid index exception when renaming snapshot files
- Query may be in rescheduling state when scheduler finishes
- Ensure cancel-to-resume finishes so query can be retried
- Fix issue where hash builder causes "too many components" error
- Do not delete task result if task instance id is not available
- Do not delete files when table-writer is cancelled or restored
- Instead remove files before resume
- Do not move cancelled tasks to aborted, and do not close outputBuffer
- Improve synchronization when marking drivers as finished
- Fix potential NPE when a task error is received
- Improved logging
- This reverts commit a997fdfe
- Added fix for potential NPE when rolling back hive writes
- Fix MarkerSplitSource's sentFinalMarker for resume marker
- Avoid failing query if only 1 worker is left
- Fix issue with updating Hive transactional tables
- When there aren't enough worker available, retry earlier snapshots
- Move state save and load code to TaskSnapshotManager
- Reliably delete snapshots for finished queries
- Recognize restore failures on first error
- Fix issue with MultiChannelGroupByHash capture/restore
- Handle error when cleaning up sub files
- Handle case where no file system profile is available
- Support snapshot in Hive connector (ORC only)
- Update query editor to support new query states
- Update TPC-H and TPC-DS connectors to support rescheduling of splits
rajeevrastogi
i-robot