Details
-
Sub-task
-
Status: Resolved
-
Major
-
Resolution: Fixed
-
M3
-
None
-
None
Description
In a load test, I noticed that it's possible for the tablet to "fall behind" the RPCs, now that the RPCs are being responded to asynchronously. The various executors inside TabletPeer probably need to have their queues become bounded, and return a "come back later" type response to clients if they get backed up too much. Otherwise, the TS ends up growing larger and larger in memory as it gets backed up.