From our meeting yesterday:

• To do progressive deploys of the train, we just have to modify sync-wikiversion and keep track of its completion (by looking at the version of the php file)
• For more general progressive deploys, both of the follwing things must happen:
  • Reorganization of the code we deploy in production with a clear separation of components (mediawiki+extension branches, docroots/static assets, wmf-config)
  • Removal of the sync-file functionality

Some things to consider while implementing this:

• We have a predefined sets of canary hosts in /etc/dsh/group/mediawiki-appserver-canaries and /etc/dsh/group/mediawiki-api-canaries (puppet)
• We have role::mediawiki::canary_appserver and mediawiki::appserver::canary_api; they simply include class::scap::scripts
  • scap runs eg modules/service/logstash_checker.py on a canary server
• There are outdated references and possible code parts (eg references to tin.eqiad.wmnet) which are causing some confusion
• Canary wait time is option canary_wait_time which can be overriden
• Scap2 and scap3 are in the same repo

Additional ideas:

• We could use conftool directly so to access server weights and calculate percentage of traffic
  • This might need changes in development environment (scap-vagrant)