[Ubuntu 16.04 and official Docker image] System.bundle stops responding
I am having an issue with Plex Media Server, including versions 126.96.36.19937 through current, 188.8.131.5290. I originally had the issue under a native install in Ubuntu 16.04, so I tried moving to docker (where I am now), and the issue persisted.
1) Matching stops working. Automatic matching doesn't match anything, and trying to match manually results in a timeout (There was a problem...)
2) Server settings page cannot be loaded (Results in "Server settings are unavailable.")
3) Content art doesn't display for random selection of items (locally cached works, maybe?) Displays the gray PC monitor look'n icon instead of the art.
This can occur after a couple days of uptime, or a few hours, it seems entirely random. I've had to setup cron to originally restart plex, and now restart the docker container, nightly. Even so, it still can happen within the 24 hour window, sometimes as early as a few hours after startup.
I've determined the System.bundle appears to be to blame. In the logs, the problem is indicated by a lot of curl_easy_perform timeout errors, all against the current port System.bundle is listening on, as shown in the attached log file. The com.plexapp.system.log shows nothing eventful going on, and a strace on the process shows just a bunch of what appear to be busylooping threads. In a normal, non-docker install, the issue can be resolved by killing the pid of the System.bundle python instance, and letting plex re-spawn it. This doesn't seem to work for a docker instance, as it doesn't appear to respawn the process.
I've had this issue since at least 184.108.40.20637 (the latest I made notes from), up to the most current version, 220.127.116.1190. It occurs both under a standard install in Ubuntu 16.04, and the official Plex docker image. The library was preserved during the move to docker, and I have checked the sqlite databases for any corruption, and they all seem to check out.