Last modified: 2014-07-10 22:39:34 UTC
I've got multiple requests in the logs of the form http://commons.m.wikimedia.org/w/api.php?useformat=mobile&r=[redacted]&origin=https://en.m.wikipedia.org with the MIME type text/html. Yuvi informs me these are upload requests from mobile web, which is useful but are they actually getting text/html, or what? Because MIME type filtering is important in how we understand traffic-related metrics, and if they're not asking for text/html content they shouldn't pretend that they are.
Prioritization and scheduling of this bug is tracked on Trello card https://trello.com/c/oGal5O2C
Oliver, anything interesting about their user-agents?
Adding Ezachte; Erik, MaxSem asked about which header we're taking the MIME type found in the sampled logs from. Any chance you know?
Sorry, no idea, I take the mime types in the log at face value. They might be from a footer instead of a header. Christian might know better?
For sampled-1000 logs underneath /a/squid/archive on stat1002, it is [1] %{Content-Type}o So Content-Type header of response to client. [1] https://git.wikimedia.org/blob/operations%2Fpuppet.git/ebcbef50568960d424fcb95fc79ba3be945a905e/modules%2Fvarnish%2Ffiles%2Fvarnishncsa.default#L9
It looks like the bug is in fact not MobileFrontend, but instead PHP. We all knew the language was a big collection of bugs and now we have validated it ;p. Luckily it is a CONSISTENT big collection of bugs and so I can modify the PV heuristics slightly to solve for this.