Page MenuHomePhabricator

OAuth error flood
Closed, DuplicatePublic

Description

OAuth has become a gamble, and users are losing. An (incomplete!) list of errors I get:

  • NULL (API breakage; filed bug reports for that in the past)
  • Not logged in (just keep reloading https://tools.wmflabs.org/widar/ , after 2-3 tries it will tell you you haven't authorized OAuth, even if you just did)
  • "The authorization headers in your request are not valid: Invalid signature" (this same error comes in two variants, for some reason)
  • "Invalid token"

All this with the same code that ran fine a week or so ago.

PLEASE MAKE OAUTH WORK AGAIN RELIABLY!!!

(and please, for once, don't try "it's your fault" as default response...)

Related Objects

Event Timeline

Magnus raised the priority of this task from to Needs Triage.
Magnus updated the task description. (Show Details)
Magnus changed Security from none to None.
Magnus subscribed.

Was reminded of another "silent" error: Occasionally, the OAuth edit is attributed to an IP address. Reported this before as well.

By now, most of my code hard-ignores "invalid token" and not-logged-in error, and just tries again; otherwise, work would have become impossible weeks ago. As a result, one occasionally gets "blocked" by some anti-spam measure. Seems that's the only part of OAuth that is currently working...

I get the invalid token error pretty frequently especially in mobile chrome on iOS. Usually going through the process a second time fixes it, but it happens often enough that it is annoying.

@Magnus: do you know when these issues started appearing? The OAuth issues may be related to the HHVM roll-out, as @TheDJ mentioned.

"invalid token" and not-logged-in have been going on for weeks (probably HHVM rollout).

Most of the other ones appeared today for me. Ignoring "invalid token" and not-logged-in, it now takes about ~3-5 attempts for an action to go through.

Update: Seems to have calmed down now. Did anyone fix it?

There is another ticket somewhere this should be merged into and yes.

Ragesoss claimed this task.

This may have been the same issue as T75968, fixed several hours ago, so I'll close. Re-open if those issues (which it seems were HHVM-related, and ramped up yesterday when more servers were switched to HHVM) haven't really gone away.