This MR is to skip the redundant persistent ref existence check. This pressurizes Gitaly nodes and if timeout happens, it could end up fetching all refspecs in runner. The assumption was described by @stanhu at gitlab-com/gl-infra/production#2143 (comment 343690877)
Ok, maybe this is what happened based on https://log.gprd.gitlab.net/goto/15e8e5a9084502a3b17113fa0e0c1e60:
- The ref did, in fact, exist. It seems like there are many chances to create the ref since every build attempts to do so.
- However, the
RefExists
call timed out.- As a result, https://gitlab.com/gitlab-org/gitlab/blob/2bfe8f06724470576e408f3cc282a30f0c271880/app/models/ci/persistent_ref.rb#L18-19 returned
false
, and the CI pipeline got an empty refspec.I still wonder why so many RPCs were stalling out, but perhaps we should cache the existence of this ref?
If this MR contains changes to processing or storing of credentials or tokens, authorization and authentication methods and other items described in the security review guidelines:
@gitlab-com/gl-security/appsec