Jump to navigation Jump to search
document status: final
|Incident ID||2022-05-25 de.wikipedia.org||Start||20:08|
|People paged||26||Responder count||8|
|Impact||For 6 minutes, a portion of logged-in users and non-cached pages experienced a slower response or an error. This was due to increased load on one of the databases.|
An increase in POST requests to de.wikipedia.org caused an increase in load on one of the DB servers resulting in an increase in 503 responses and increased response time
All times in UTC.
- 20:04 OUTAGE BEGINS
- 20:04 Received page "Service text-https:443 has failed probes"
- 20:08 rzl starts investigation
- 20:08 Received page "(FrontendUnavailable) firing: HAProxy (cache_text) has reduced HTTP availability"
- 20:08 rzl asked cjming to halt deploying
- 20:09 Recived page "(FrontendUnavailable) firing: varnish-text has reduced HTTP availability"
- 20:09 jbond takes IC
- 20:10 < rzl> looks like a spike of DB queries to s5 that saturated php-fpm workers, seems like it's already cleared
- 20:11 Received recovery "RECOVERY - High average GET latency for mw requests on appserver"
- 20:11 < cwhite> Lots of POST to https://de.wikipedia.org
- 20:12 < rzl> s5 did see a traffic spike but recovered, still digging
- 20:13 Received recovery "resolved: (8) Service text-https:443 has failed probes"
- 20:13 Received recovery "resolved: HAProxy (cache_text) has reduced HTTP availability"
- 20:14 Received recovery "resolved: varnish-text has reduced HTTP availability"
- 20:14 OUTAGE ENDS
- 20:14 < cwhite> 2217 unique ips (according to logstash)
- 20:18 < bblack> identified traffic as "a bunch of dewiki root URLs"
- 20:22 < _joe_> php slowlogs mostly showed query() or curl_exec()
- 20:30 < _joe_> someone was calling radompage repeatedly?
- 20:31 <rzl> looks like it was all appservers pretty equally
- 20:40 Discuss remediation strategy
- 20:48 Incident officially closed
- 20:51 < rzl> gave cjming all clear to continue with deploy
- 21:29 requestctl rule put in place
Error was detected by alert manager monitoring
20:08 <+jinxer-wm> (ProbeDown) firing: (8) Service text-https:443 has failed probes (http_text-https_ip4) #page - https://wikitech.wikimedia.org/wiki/Network_monitoring#ProbeDown - https://grafana.wikimedia.org/d/O0nHhdhnz/network-probes-overview?var-job=probes/service&var-module=http - https://alerts.wikimedia.org/?q=alertname%3DProbeDown 20:08 <+jinxer-wm> (FrontendUnavailable) firing: HAProxy (cache_text) has reduced HTTP availability #page - TODO - https://grafana.wikimedia.org/d/000000479/frontend-traffic?viewPanel=13 - https://alerts.wikimedia.org/?q=alertname%3DFrontendUnavailable 20:09 <+jinxer-wm> (FrontendUnavailable) firing: varnish-text has reduced HTTP availability #page - https://wikitech.wikimedia.org/wiki/Varnish#Diagnosing_Varnish_alerts - https://grafana.wikimedia.org/d/000000479/frontend-traffic?viewPanel=3 - https://alerts.wikimedia.org/?q=alertname%3DFrontendUnavailable
Understanding of legitimate backed traffic would enable us to better sanitize bad traffic at the front end
What went well?
- automated monitoring detected the incident
- Had a good amount of incident responders
What went poorly?
- Was difficult to get a signature of the post traffic
Where did we get lucky?
- Incident ended quickly on its own
How many people were involved in the remediation?
Links to relevant documentation
- T309147 any POST that doesn't go to /w/*.php or /wiki/.* should become a 301 to the same url
- T309186 Created sampled log of post data
- T310009 Make it easier to create a new requestctl object
|People||Were the people responding to this incident sufficiently different than the previous five incidents?||yes|
|Were the people who responded prepared enough to respond effectively||yes|
|Were fewer than five people paged?||no|
|Were pages routed to the correct sub-team(s)?||no|
|Were pages routed to online (business hours) engineers? Answer “no” if engineers were paged after business hours.||no|
|Process||Was the incident status section actively updated during the incident?||yes|
|Was the public status page updated?||no|
|Is there a phabricator task for the incident?||yes||(created retrospectively)|
|Are the documented action items assigned?|
|Is this incident sufficiently different from earlier incidents so as not to be a repeat occurrence?||no||(similar to "2022-05-20 Database slow / appserver")|
|Tooling||To the best of your knowledge was the open task queue free of any tasks that would have prevented this incident? Answer “no” if there are
open tasks that would prevent this incident or make mitigation easier if implemented.
|Were the people responding able to communicate effectively during the incident with the existing tooling?||yes|
|Did existing monitoring notify the initial responders?||yes|
|Were all engineering tools required available and in service?||yes|
|Was there a runbook for all known issues present?||no||Setting to no as we need to update the DDoS playbook. We have also updated the question from now on to reflect that.|
|Total score (count of all “yes” answers above)||7|