Make a roachtest with a sink that can keep up with regular traffic but cant keep up with catchup scans. Introduce a transient error and watch the feed never recover. This is something we think we've seen a few times during escalations and the scale test, and it would be nice to have a repro for it.
Make a roachtest with a sink that can keep up with regular traffic but cant keep up with catchup scans. Introduce a transient error and watch the feed never recover. This is something we think we've seen a few times during escalations and the scale test, and it would be nice to have a repro for it.
Jira issue: CRDB-45147