Introduction

One of the important recovery operations is to replay uncommitted records in WAL. The different WAL recovery modes define the behavior of WAL replay.

WAL Recovery Modes

This mode is acceptable for most application since this provides a reasonable tradeoff between starting RocksDB after an unclean shutdown and consistency.

In this mode, the WAL replay is stopped after encountering an IO error. The system is recovered to a point-in-time where it is consistent. This is ideal for systems with replicas. Data from another replica can be used to replay past the “point-in-time” where the system is recovered to. (This is the default as of version 6.6.)