In HELP-65140, we found that requests from secondary to primary were being sent every 500ms, but none of them succeeded, so they piled up. We should investigate using a backoff algorithm to avoid sending so many requests when they're all failing or going very slowly.