Xtools will go through multiple periods of returning 503 or 500 error codes throughout the day. Each period lasts less than a minute.
Originally noticed using an external process that pings for status codes every 5 minutes. URL used was http://tools.wmflabs.org/xtools/sc
This is the web watcher I believe. It detected the service did not respond (at the time) with a 200, so it restarted it. This very brief period where it restarts will throw a 500, and I guess a 503 (don't remember), during stages of it rebooting. As far as the restart process goes, I see the same behavior with other tools running on Lighttpd and the grid engine. The new kubernetes system seems much smoother and apparently eliminates the need for the web watcher script. This however doesn't play nice with the base xTools suite, something we should look into.
I am closing this, T153112: Epic: Rewriting XTools should fix this by eliminating the need for the web watcher at all. T167217: Planning for Xtools beta is the plan for rolling that out.