Details
-
Improvement
-
Status: Open
-
Major
-
Resolution: Unresolved
-
None
-
None
Description
We may improve write latency and throughput by allowing multiple messages to be in-flight from a Raft leader to its followers. Secondarily, it may also help with MTTR (for re-replication) because after tablet copy a node has to catch up via normal streaming means.
Note that pipelining writes could exacerbate the RPC “stacking problem” that we already have under certain conditions.