CodeParkAOF rewriting replaces the long AOF log with the shortest set of commands that rebuilds the current data. It is needed because the AOF grows with every write.
For example, a counter incremented a million times creates a million INCR entries. But its final state is just one value. The rewrite stores that one value instead.
How Redis does it without blocking:
- Redis forks: it creates a child process with a copy of memory.
- The child writes the new base file from memory.
- Meanwhile, the parent keeps serving clients and logging new writes.
Since Redis 7, the AOF is multi-part. It has a base file (in RDB format by default) plus incremental AOF files. A manifest file keeps track of them. During a rewrite, new writes go to a fresh incremental file.
Rewrites start automatically, based on auto-aof-rewrite-percentage and auto-aof-rewrite-min-size. You can also start one manually with BGREWRITEAOF.
- The AOF log grows with redundant commands.
- A rewrite writes the fewest commands needed for the current state.
- A forked child does the rewrite, so clients are not blocked.
- Redis 7 uses a base file, incremental files and a manifest.
- Rewrites start at automatic thresholds or with
BGREWRITEAOF.