matrix.org reports failures

[richvdh]

I'm not really sure if this is matrix.org's fault or the voip tester's; if it's the former then it seems pretty bad if the primary public server isn't getting this right:

On Chrome:

On Firefox, it took ages, and eventually gave:

[reivilibre]

Is this on test.voip.librepush.net (which is currently running with #18) or on a local copy?

This is probably the tester's fault, especially if Firefox gives it a clean bill of health. The inconsistency is an issue even if nothing else.
It did actually catch an issue with matrix.org some months ago but I seem to remember it was fixed and monitoring was added to make sure it didn't happen again :) — so yes, probably the tester's fault here.

[reivilibre]

matrix.org works (and quickly) for me on Brave, which is the closest to Chrome that I had, but I downloaded a copy of ungoogled-chromium and it took ages (you really weren't kidding, although you said this about Firefox, which works fine for me).

At the end of it, it came up with this contradiction:

(The overall result probably shouldn't be 'Excellent' if the actual test runs fail even though a promising-looking candidate was found, but that's for another issue.)

I wonder why it's quite so inconsistent (for you, Firefox misbehaves; for me, it's Chromium).

The 'taking ages' part seems to be waiting for ICE to signal that it finished gathering candidates — even though all of them were available within the first second. This could be ameliorated by adding Trickle ICE support (or perhaps even just cutting ICE off, once a candidate that is interesting for our test is received, if this is allowed).

The failure seems to come from an error in the testing service itself, and it comes back seemingly instantly, but the logs don't immediately reveal anything.

[richvdh]

Is this on test.voip.librepush.net (which is currently running with #18) or on a local copy?

yes, this was on test.voip.librepush.net.

Something I have noticed in the past is that Chrome gets upset by TURN servers that expose IPv6 addresses (as turn.matrix.org) does, though I now can't quite remember the details. So part of the difference may be about whether turn.matrix.org is reachable over IPv6 from where you are?

[reivilibre]

Is this on test.voip.librepush.net (which is currently running with #18) or on a local copy?

yes, this was on test.voip.librepush.net.

Something I have noticed in the past is that Chrome gets upset by TURN servers that expose IPv6 addresses (as turn.matrix.org) does, though I now can't quite remember the details. So part of the difference may be about whether turn.matrix.org is reachable over IPv6 from where you are?

My ISP doesn't do IPv6 yet so you're probably right on the mark with that one

[richvdh]

that might also be part of why chromium is slow for you, if the ipv6 bits are timing out?