In the method HostedRepositoryPool#materializeClone, when the bot is trying to reuse a local repo instance, the bot will always check whether the repo is still good by processing command "git fsck --connectivity-only".

Sometimes this command would be slow and Erik said that we should believe the bots are doing right things, so we should assume the local repos are good, so the health check is not always needed. But shutdown of the bot could make a local repo instance unhealthy, so we should at least do the health check once for each local repo instance after the bot is restarted.

To solve this issue, Erik suggested to maintain a static map for keeping track of paths of known checked/good repositories.