If the WAL is enabled, these updates are then added to the WALEdit and attempted to be written to the WAL after the WALEdit has been saved. If any of the index updates fail, this server is immediately terminated and we rely on WAL replay to attempt the index updates again (see {@link #preWALRestore(ObserverContext,HRegionInfo,HLogKey,WALEdit)}).
If the WAL is disabled, the updates are attempted immediately. No consistency guarantees are made if the WAL is disabled - some or none of the index updates may be successful. All updates in a single batch must have the same durability level - either everything gets written to the WAL or nothing does. Currently, we do not support mixed-durability updates within a single batch. If you want to have different durability levels, you only need to split the updates into two different batches.
|
|