In https://gitlab.com/gitlab-com/gl-infra/scalability/-/issues/661, we found that many concurrent calls to the raw blob endpoint using a ref, not a SHA, could cause Gitaly's FindCommit method some pain. In https://gitlab.com/gitlab-com/gl-infra/scalability/-/issues/661#note_450983566, Bob pointed out we don't need to call FindCommit at all for this action: >>> We could probably also remove the `FindCommit` from that endpoint. The `@commit.id` that we're using in [`RawController#show`](https://gitlab.com/gitlab-org/gitlab/blob/ff05e068037944aef1a9bf6bd257b9c74cf56de2/app/controllers/projects/raw_controller.rb#L20) isn't really needed, we can also pass a ref-name to `blob_at`. There's some validation that happens in [`ExtractsPath#assign_ref_vars`](https://gitlab.com/gitlab-org/gitlab/blob/ff05e068037944aef1a9bf6bd257b9c74cf56de2/lib/extracts_path.rb#L31) but I wonder if we could optimize there by only triggering the blob-rpc, since we don't really need any of the other assigned instance variables. >>>