fix increased registry error rates

Is this benign? no! We should investigate this further. There is an ongoing incident in GCP related to GKE, but if you zoom out in grafana you'll see registry errors were also slightly elevated last week, e.g. https://dashboards.gitlab.net/d/general-service/general-service-platform-metrics?orgId=1&from=1573257602000&to=1573344001000&panelId=8&tz=UTC&var-environment=gprd&var-type=registry&var-stage=main&fullscreen

So even though the error rate exceeding SLO today is caused by GKE issue, we had a higher then expected error rate for some time now that did not exceed the SLO threshold