Incident documentation/2021-09-26 appserver latency

document status: in-review

Summary

Increased db load for enwiki (s1) resulted in slower responses, which in turn resulted in overall php-fpm worker limits being reached and thus affecting requests for all wikis. For requests above the limit, the error was "upstream connect error or disconnect/reset before headers. reset reason: overflow".

Impact: For about 15 minutes, backend appservers were slower or unable to respond for all wikis. This mainly affected logged-in users and most bot/API queries. Some page views from unregistered users were affected, for pages that were recently edited or otherwise expired from the CDN cache.

Documentation:

• Public incident task: T291767

• Similar to Incident documentation/2021-09-04 appserver latency and Incident documentation/2021-09-18 appserver latency.

Actionables

• T291767 (restricted)
• T251885 (restricted)