A Redis slave which persists to disk without loading the data set into memory.
Development
A problem with offloading persistence to a slave is that it requires the entire data set to be held in memory. This results in persistence being memory-bound. You cannot, for example, have one or few machines dedicated to persisting a cluster of Redis masters without having sufficient memory.
On startup, a sync command is sent to the configured master. The master first replied with a full dump (.rdb), which redisync saves in a standalone file. The master continues to send a replication stream, which redisync saves in an append only file (.aof).
To limit the size of the append only file, a new worker is start as a configurable interval (the old worker doesn't stop until the new worker is properly established).
Restoration is accomplished in two steps.
First, the desired .rdb file is loaded into the new master server (the same way you restore any redis rdb file).
Next, the corresponding .aof file is piped into the master:
cat 20131124_135146.aof | redis-cli --pipe
Configuration is done via the config.json file (or, optionally by specifying a path via the -config=PATH parameter).
- ttl the time, in seconds, that a worker will live for
- network tcp/unix
- address the address:port for the master
- storage the path to store files