I think if startTenantServerInternal returns both a server and an error, which can happen at least here, then we'll ignore the server here and then we retry the server creation at a higher level, here. I think we're supposed to stop the new server, as this comment says. Generally, I'm not sure that attempting to recover from a tenant server failing to start is possible / sane.

I may well be missing something.

Jira issue: CRDB-24807
Epic: CRDB-23559