Binglin Chang has posted comments on this change.

Change subject: Add per tablet write RPC throttling
......................................................................


Patch Set 2:

> have you tried running this in a cluster yet? curious to hear a report if it 
> works in practice.
Haven't done on a cluster, but the integration tests is working.

> since this is more of a short-term solution for multitenancy, I think we 
> should tag the configuration flags as 'experimental'. This tag means that we 
> are free to break the configuration in a future version (not preserve 
> backwards-compatibility). Are you OK with this? I think we may want to add 
> something more advanced like global fair sharing, or time-slice based 
> fairness rather than explicit ops/second throttles later.
OK

> would be nice to add an integration test which uses the client against a 
> single-server cluster where you've throttled to 100 ops/second or something, 
> and verify on the client side that the limit is applied correctly. That way 
> we can verify that the back-pressure mechanisms work properly, etc.
Done.

Also I see the client retry sleep time is  retryTime + rand(5), this may not be 
long enough for some configurations, so the client may retry multiple times 
before success, we may need a way to return backoff time to clients.

-- 
To view, visit http://gerrit.cloudera.org:8080/2327
To unsubscribe, visit http://gerrit.cloudera.org:8080/settings

Gerrit-MessageType: comment
Gerrit-Change-Id: I33cb6934d27b883a783682cef1e0723100637d45
Gerrit-PatchSet: 2
Gerrit-Project: kudu
Gerrit-Branch: master
Gerrit-Owner: Binglin Chang <[email protected]>
Gerrit-Reviewer: Binglin Chang <[email protected]>
Gerrit-Reviewer: Kudu Jenkins
Gerrit-Reviewer: Todd Lipcon <[email protected]>
Gerrit-HasComments: No

Reply via email to