[NEW] Save RDB in the replica Using a separate thread instead of the mainthread to speed up disk-based full sync

[xbasel]

The problem/use-case that the feature addresses

Offload RDB saving to disk into a separate thread (currently it is done in the main threads), this speeds up disk-based full synchronization, especially when the replica is busy processing client requests.

Description of the feature
Speeds up disk-based full sync.

Alternatives you've considered

An alternative could to prioritize RDB traffic in the replica over normal client requests. But I think a separate thread doing this:

while (!eof)
  buf=read(primaryFd)
  write(rdbFileFd, buf)

is simpler.

[madolson]

@xbasel I'm not sure there is that much value in prioritizing disk based full sync when it is already much slower than directly serializing the data into memory.

[xbasel]

@madolson IMHO there is still value in disk-based fullsync on the replica side if administrators prefer the replica to continue serving clients with stale data while downloading the RDB (as it improves replica availability).

[madolson]

Sounds good.