http://shinken.wmflabs.org/problems?search=deployment shows alarms for the deployment-mediawikiXX instances that no more respond to HTTP requests. That causes beta to be unavailable.
From deployment-mediawiki04 syslog, it seems that is logrotate that restarted hhvm but an invalid configuration prevent the service from starting.
Nov 16 06:32:45 deployment-mediawiki04 systemd[1]: Stopping HHVM PHP/Hack runtime... Nov 16 06:33:10 deployment-mediawiki04 diamond[455]: Failed to collect metrics Nov 16 06:34:10 deployment-mediawiki04 diamond[455]: HTTPError: HTTP Error 503: Service Unavailable Nov 16 06:34:15 deployment-mediawiki04 systemd[1]: hhvm.service stop-sigterm timed out. Killing. Nov 16 06:34:15 deployment-mediawiki04 systemd[1]: hhvm.service: main process exited, code=killed, status=9/KILL Nov 16 06:34:15 deployment-mediawiki04 systemd[1]: Ignoring invalid environment 'RUN_AS_GROUP=www-data Nov 16 06:34:15 deployment-mediawiki04 systemd[1]: ## Add additional arguments to the hhvm service start up that you can't put in CONFIG_FILE for some reason. Nov 16 06:34:15 deployment-mediawiki04 systemd[1]: # ADDITIONAL_ARGS= Nov 16 06:34:15 deployment-mediawiki04 systemd[1]: ': /etc/default/hhvm Nov 16 06:34:15 deployment-mediawiki04 systemd[1]: Unit hhvm.service entered failed state. Nov 16 06:34:27 deployment-mediawiki04 kernel: [1242254.202289] hhvm[12342]: segfault at 8 ip 0000000000e6dbb1 sp 00007f7646ff9468 error 4 in hhvm[400000+2500000]
Note the segfault.
How to reproduce:
On a deployment-mediawiki instance: furl http://en.wikipedia.beta.wmflabs.org/wiki/Main_Page
See /var/log/hhvm/error.log and you get either one of:
Core dumped: Segmentation fault
Or
Fatal error: Stack overflow in /srv/mediawiki/php-master/includes/libs/objectcache/BagOStuff.php on line 754