Steps to replicate the issue
This message is a bit strange, I can and did log in with OAuth for several other tools without any problem.
The problem seems to emanate from a single tools proxy (tools-proxy-9.tools.eqiad1.wikimedia.cloud). Adding infrastructure team and setting priority to medium as outage does not seem to be affecting other tools.
Still unresolved and appears to start affecting a limited number of other tools.
@dcaro if you have a second to look into this or redirect to the correct person/group, I'm not really specialized in that type of issues. Thanks.
I see the following error in the proxy logs:
2025/10/17 12:07:59 [error] 1395211#1395211: *185108674 upstream sent too big header while reading response header from upstream
This might be fixed by tweaking the proxy buffer sizes (or disabling it completely, that I think we do somewhere else already).
Weirdly enough it's just the returned data for OAuth, it's not a particularly large data source or input after authentication.
This seems to have been fixed as a side effect of T283948: Merge Toolforge Nginx front proxy into the existing K8s HAProxy setup.