Page MenuHomePhabricator

rel=canonical of https pages should point to http
Closed, ResolvedPublic

Description

According to Google, HTTPS versions of our pages should have the rel=canonical head tag that points to the http version. See http://support.google.com/webmasters/bin/answer.py?hl=en&answer=139394

I can't remember that we ever had it like this, but if Google suddenly started enforcing this, then it might explain the http-https google split and google rank penalty that Wikid77 has been discussing on WP:VP/T.


Version: wmf-deployment
Severity: normal
URL: http://support.google.com/webmasters/bin/answer.py?hl=en&answer=139394
See Also:
https://bugzilla.wikimedia.org/show_bug.cgi?id=25882
https://bugzilla.wikimedia.org/show_bug.cgi?id=28602
https://bugzilla.wikimedia.org/show_bug.cgi?id=51002

Details

Reference
bz48402

Event Timeline

bzimport raised the priority of this task from to High.Nov 22 2014, 1:31 AM
bzimport added a project: HTTPS.
bzimport set Reference to bz48402.
bzimport added a subscriber: Unknown Object (MLST).
TheDJ created this task.May 13 2013, 1:13 PM

(In reply to comment #0)

it might explain the http-https google split and google
rank penalty that Wikid77 has been discussing on WP:VP/T.

Link: https://en.wikipedia.org/w/index.php?title=Wikipedia:Village_pump_%28technical%29&oldid=554883644#Relinking_Google_for_SSL_https

I would strongly suggest WONTFIXing this bug in favour of 51002, which feels like a better solution given our long terms goals (see bug 47832).

Until we're ready to switch to HTTPS for anons by default it makes sense to point HTTPS versions of pages to HTTP via rel=canonical. We should be setting one or the other, rather than having a split.

The load on the SSL cluster has been doubling month by month and this split may be the reason why. Unfortunately we don't have conclusive metrics to prove this, but making this setting explicit rules that out as a possibility.

(In reply to comment #3)

Until we're ready to switch to HTTPS for anons by default it makes sense to
point HTTPS versions of pages to HTTP via rel=canonical. We should be setting
one or the other, rather than having a split.
The load on the SSL cluster has been doubling month by month and this split
may be the reason why. Unfortunately we don't have conclusive metrics to prove
this, but making this setting explicit rules that out as a possibility.

So we should do this one now and bug 51002 next month/quarter/year/decade when we get around to it?

https://gerrit.wikimedia.org/r/76856 has been submitted and merged. I imagine it'll be deployed shortly.

Created attachment 13066
SSL cluster eqiad network_report

Also for the records, from #wikimedia-operations:

18.15 < Ryan_Lane> looks like the rel=canonical change helped
18.15 < Ryan_Lane> https://ganglia.wikimedia.org/latest/graph.php?r=week&z=xlarge&c=SSL+cluster+eqiad&m=cpu_report&s=by+name&mc=2&g=network_report
18.15 < Ryan_Lane> I had a feeling that was causing the surge
18.16 < Ryan_Lane> I'm surprised how quickly it helped, though

Attached:

Restricted Application added a project: Traffic. · View Herald TranscriptAug 5 2016, 11:40 AM