Incidents/20140121-BitsApplicationServers

From Wikitech
Jump to navigation Jump to search

Summary

A change to the default JavaScript and CSS payload of pages across Wikimedia projects caused the Bits application servers to overload, resulting in some users being served pages with no CSS or JavaScript for a period lasting about eighteen minutes.

Timeline

  • 11:24 - Change I71b70d8ee deployed to 1.23wmf11
  • 11:25 - Change deployed to 1.23wmf10
  • 11:28 - mw1149, mw1150, mw1151 and mw1152 begin flapping, with icinga-wm reporting alternating socket timeouts / recovery. Users on IRC were now reporting that some pages were loading without CSS or JavaScript. Ganglia shows a spike in busy Apache workers.
  • 11:32 - Load peaks; hosts begin to recover.
  • 11:35 - Smaller second peak
  • 11:46 - All hosts recovered.

Actionables

Already on RT according to mark:

  • Ensure changes to default JS / CSS payload are deployed in a staggered fashion.
  • Add redundancy to Bits application cluster or dissolve into broader app server pool.