Often when running long-running tasks such as schema changes we need to depool the hosts and wait for all the connections (from MW servers) to clean up.
This normally involves wikiuser and wikiadmin, the first one is used from production and wikiadmin is mostly used from mwmaint* hosts for maintenance tasks.
It is pretty common to find maintenance scripts connecting to a single host and not reloading the config and hence messing up with general depooling/pooling actions. While these are annoying, there are other critical operations where this can be a massive issue (ie: master swaps).
Lastly, as we are now in process of running automatic schema changes, the schema change scripts does wait for all the connections to be gone after the depooling, so this can mess up and make the script take hours before it can proceed.
Scripts should reload the config after each change, or after X minutes.
See also: