Page MenuHomePhabricator

Detect image loading from production, pull those images into beta cluster
Closed, DeclinedPublic

Description

Specifically because of this:
http://en.wikipedia.beta.wmflabs.org/wiki/Template:Wikipedia%27s_sister_projects

The icons come from upload.wikimedia.org

Same issue other places as well, eg:
http://en.wikipedia.beta.wmflabs.org/wiki/Hurricane_Celeste_%281972%29
which loads the acrobat.gif from upload.wikimedia.org


Version: unspecified
Severity: enhancement
See Also:
https://bugzilla.wikimedia.org/show_bug.cgi?id=37245

Details

Reference
bz61784

Event Timeline

bzimport raised the priority of this task from to Low.Nov 22 2014, 3:03 AM
bzimport set Reference to bz61784.
bzimport added a subscriber: Unknown Object (MLST).

See also bug 37245 (previous version of this bug)

Oh, that's smarter than I thought...

So, if the file is available on upload.beta, the it's served from there. If it isn't, it fails over to upload.wikimedia? Not bad.

Renaming bug and resetting priority to clarify the new goal :)

The ideal (something like):

  1. Define a set of pages that are high use in (auto/manual) testing
  2. Detect (daily?) any image loads from those pages that are from production
  3. Fetch them from production and put them into beta cluster

We do have instant commons and the foreignApi backend file system of doom. So in theory all images are looked locally on beta cluster, then on beta commons and finally on the production commons.

When production goes down (like this morning) Beta Cluster can't get images that are hardcoded, eg the main page:

Selection_016.png (289×1 px, 38 KB)

It still feels icky to depend on prod in any way.