RFR: 1801: Exclude some GraphQL calls from RestRequestCache rate limiter

Magnus Ihse Bursie ihse at openjdk.org
Tue Jan 24 16:15:37 UTC 2023


On Tue, 24 Jan 2023 14:41:46 GMT, Erik Joelsson <erikj at openjdk.org> wrote:

> In [SKARA-1731](https://bugs.openjdk.org/browse/SKARA-1731), the best practices regarding non GET requests on Github became better enforced. Unfortunately that has lead to some unexpected performance issues. I believe this is mostly caused by GraphQL requests, which are always POST, being treated as a non GET REST requests by the limiter and limited to one per second. This limitation on a subset of our github queries is currently causing a queue buildup for that lock, which stabilizes to about 1m wait time. This delay is slowing down response times from the bots massively.
> 
> Reading the Integrator best practices document again (https://docs.github.com/en/rest/guides/best-practices-for-integrators?apiVersion=2022-11-28) and the GraphQL resource limitations (https://docs.github.com/en/graphql/overview/resource-limitations) I'm pretty sure you aren't supposed to limit GraphQL to one call per second, but rather just make sure you stay within the points limit. The one per second recommendation is for the REST API.
> 
> I want to make it possible to exclude certain calls from this rate limiter to see if this can alleviate the contention we are currently seeing. I have run with this patch for the pr bot since yesterday afternoon (PT) and it seems to be working well.
> 
> Two PRs were unable to integrate yesterday due to this contention:
> https://github.com/openjdk/jdk/pull/12139
> https://github.com/openjdk/jdk/pull/11922

You don't think we need to explicitly track GraphQL "points" to allow the rate limiter in RestRequest allow for future GraphQL requests which are more expensive? Your comment seem to indicate that you have manually analyzed our current GraphQL requests and come to the conclusion that they are worth 1 point each.

-------------

PR: https://git.openjdk.org/skara/pull/1461


More information about the skara-dev mailing list