We've been removing wmf branches regularly, but I don't feel very confident in how we're doing this right now. Even if not on a regular basis, could should put some monitoring in place to verify how long it takes for traffic to an old wmf branch to train after having rolled out a new branch to all wikis.
It should take about 30 days (given there are hardcoded references in the cached html).
Get data to answer:
- How fast do we roll over? Does most the majority of requests hit fresh versions within the first few days? Or does the majority of requests go to old versions until the cache expires?
- How long the tail of left overs? (presumably only from third-party html caches)
- Do all our own caches really expire within 30 days, or does the post-30-day tail contain requests with referrers from our own domains as well?