Decide what to do with `/-/readiness`
The following discussion from !18665 (merged) should be addressed:
-
@andrewn started a discussion: (+1 comment) @ayufan seems good to me, although I would still like to steer clear of
/-/readiness
in our health checks on GitLab.com for now.A future improvement: healthchecks happen a lot, and at each tier of your application, they tend to get amplified.
For example, with 1
/-/readiness
check from the load-balancer to each of your web tier servers, which in turn checks each Gitaly node. This results in (say) 40 web nodes, each making 40 calls to Gitaly servers, or 1600 health check calls in total. If the load balancer is configured to do this once a second, it generates a lot of traffic.One improvement is to schedule the healthcheck at a fixed interval in the service process. When a request is made to the health check it only returns the last result from the scheduled execution, not a synchronous result.
That's out of scope for this MR though, so
👍 approved from me.