Application servers

Wikimedia infrastructure

Data centers

Networking

Global traffic routing

MediaWiki

• Application servers
• PHP 7 and php-fpm
• Dumps
• Memcached
• Mcrouter runbook
• Kafka Job Queue
• Nutcracker
• Envoy
• MediaWiki HTTP cache headers
• Redis
• Parser cache
• Statsd
• MediaWiki On Kubernetes
• MediaWiki and EtcdConfig
• Citoid
• Mathoid

Multimedia

Data Engineering

Search

Data Persistence

Infra Foundations

Observability

Performance Team

[]

See also Application servers/Runbook for how to perform common tasks, or diagnose issues.

The Application servers (or app servers) are the several hundred Apache servers that run the MediaWiki backend software (written in PHP).

Service

Puppet roles:

• mediawiki::appserver, mediawiki::canary_appserver
• mediawiki::appserver::api, mediawiki::appserver::canary_api
• mediawiki::maintenance
• mediawiki::jobrunner

Relevant puppet classes:

• profile::mediawiki::webserver, this provisions Apache, and any other packages or resources needed by MediaWiki on app servers.
  • profile::mediawiki::httpd, the Apache service.
  • mediawiki::web::prod_sites, the Apache configuration for all production websites (including wikipedia.org).
  • Additional Apache configurations are at modules/mediawiki/files/apache/sites/. Prior to 2012, Apache configuration were in a Subversion repository.

Architecture

See also: MediaWiki at WMF and HTTP timeouts#App servers

The application servers are load-balanced via LVS. Connections between our CDN (HTTP cache proxies) and app servers are encrypted with TLS, which is terminated locally on the app server using Envoy. Envoy then hands the request off to the local Apache.

Apache is in charge of handling redirects, rewrite rules, and determining the document root. It then uses php-fpm to invoke the MediaWiki software.

The Apache MPM we use is mod_worker, which decides how php-fpm processes are spawned.

Logging

Apache errors are logged to /srv/mw-log/apache2.log on mwlog1001.

Apache access logs are mostly disabled. Statistics are drawn from Varnish front ends instead.

Hardware repair

When taking down application servers (running mediawiki) for things like disk replacement or other hardware repair, _do not forget to_:

• before: remove from dsh group

These are in puppet, operations/puppet repo, in modules/dsh/files/group. The important one for Mediawiki sync is "mediawiki-installation".

• before: de-pool in pybal
• TODO: Document what to do if it's a scap proxy (see hieradata/common/dsh/config.yaml)

See pybal. You can just grep for the server name and set 'enabled': False and save.

• before: check nobody is scapping right now (best: announce with a !log line in IRC)

This is an IRC thing on libera.chat in #wikimedia-dev ^connect/#wikimedia-tech ^connect/#wikimedia-operations ^connect

• during: acknowledge Icinga monitoring checks (best: with related ticket number as comment)

Do this by logging in via browser on icinga.wikimedia.org. search for the hostname, check all services and use the "acknowledge" option. You'll see the IRC bots outputting this as well and they will stop repeating things over and over in the channels.

• after: re-add to dsh groups

Revert the above.

• after: re-pool in pybal

Revert the above.

See also

• Apache log format
• UID