If default_server_name is set and that server is down, Riot spins forever on startup

[jryans]

1. Set default_server_name to some server in Riot's config.json
2. Ensure that the named server is down and not responding
3. Reload Riot
4. Riot will hang forever / a long time trying to query .well-known from that server, even if you are already logged in

Local build of https://riot.im/develop

[jryans]

This is currently blocking Modular's usage of default_server_name.

[lampholder]

What actually causes the problem? Is it:

• the matrix server is down, or
• the .well-known is not being served?

Or is it both?

What behaviour do we want here? What error messages do we want to report to the user?

default_server_name == cooldomain.horse

• if https://cooldomain.horse/.well-known/matrix/client 404s then
  • we could try and log in to https://cooldomain.horse/_matrix/
  • if that works, 👍
  • if that doesn't work, report to the user "cooldomain.horse (https://cooldomain.horse/) is not responding, please contact the homeserver administrator or choose a different homeserver
• if https://cooldomain.horse/.well-known/matrix/client resolves to a valid json file
  • try and log in to https://domain-from-well-known/path-from-well-known/_matrix
  • if that doesn't work, report to the user "cooldomain.horse (https://domain-from-well-known/path-from-well-known/) is not responding, please contact the homeserver administrator or choose a different homeserver"

And, we can actually do better with our error messages, too, depending on whether the server that is not responding == default_server_name. There should be a stronger emphasis on contacting the HS admin if default_server_name isn't working - if a random user-entered thing isn't working we should be encouraging the user to check their input.

[jryans]

What actually causes the problem? Is it:

• the matrix server is down, or
• the .well-known is not being served?

This specific issue only happens if the server at default_server_name is down / non-responsive. Riot will wait until the request times out before showing any UI at all (beyond a spinner). We could potentially change behaviour to show the overall auth UI but incorporate a spinner into the auth flow instead of blocking the whole app. It's notable that this current issue blocks you from using Riot even if you are already logged in.

If default_server_name is up but .well-known/matrix/client 404s, we don't hit this issue, but instead a more subtle thing currently happens: Riot will say "Sign in to your Matrix account on <default_server_name>", but it's actually using default_hs_url, which may not be what you want. (For Modular, it would probably be okay, since they plan to define both default_server_name and default_hs_url.)

[jryans]

If default_server_name is up but .well-known/matrix/client 404s, we don't hit this issue, but instead a more subtle thing currently happens: Riot will say "Sign in to your Matrix account on <default_server_name>", but it's actually using default_hs_url, which may not be what you want.

Isn't this #8818?

Yes, that seems to match the 404 case.

[jryans]

We're hoping to rework default_server_name a bit so that more of the config can be available locally and there are less traps. Removing the fire status for now.

[turt2live]

As per @jryans' comment, this is blocked on #9290

[turt2live]

Fixed by matrix-org/matrix-react-sdk#3001 - it won't make it into 1.2.0, but should be in the next regularly scheduled release.