Let's just hope nobody spends too much time on changing the error message instead of spending time properly fixing the current mess...

Ralf

Yup - it was out for between an hour and 90 minutes according to the output of a little script I've got that tries to log in to the v3 API once every 15 minutes or so. Partial output for the period in question:

2022-10-24 15:07:50 (UTC) -- starting login test
2022-10-24 15:07:52 (UTC) -- Success!
                             [response took 2105 milliseconds]
2022-10-24 15:22:53 (UTC) -- starting login test
2022-10-24 15:22:53 (UTC) -- 500 Server Error: Internal Server Error
                             [response took 736 milliseconds]
...
2022-10-24 16:25:13 (UTC) -- starting login test
2022-10-24 16:25:19 (UTC) -- 503 Server Error: Service Unavailable
                             [response took 6050 milliseconds]
2022-10-24 16:25:19 (UTC) -- will retry 2 more time(s)
2022-10-24 16:25:29 (UTC) -- 503 Server Error: Service Unavailable
                             [response took 5116 milliseconds]
2022-10-24 16:25:29 (UTC) -- will retry 1 more time(s)
2022-10-24 16:26:40 (UTC) -- 500 Server Error: Internal Server Error
                             [response took 66553 milliseconds]
2022-10-24 16:41:41 (UTC) -- starting login test
2022-10-24 16:41:43 (UTC) -- Success!
                             [response took 2111 milliseconds]

I wonder if it was a planned outage? -- it was certainly short, so either it was deliberate or someone spotted it fairly quickly (unless there's now some sort of script in place to try to cleanly restart whatever within the access control stuff breaks!)

Bearing in mind that it is still not an official production system I personally can live with short outages with no notice -- however, for a proper live system there needs to be somewhere (and not Twitter/Facebook) where such things can be (and will be) announced. (The WCG home page really should not fail to load properly unless the entire infrastructure is down, so information could be added there once that dependency on the authentication mechanism is resolved)

Cheers - Al.

[Edit - there was a textual error in the output code -- those timings were milliseconds (not microseconds!), so I've edited the above to reflect that.]

Makes sense to me.
Cheers