redacted

[ghost]

redacted

[hawkowl]

@gerg5c42g542g2c54g52c is this the matrix.org debian package or the Debian-backports one?

[richvdh]

This is presumably only true if you have changed macaroon_secret_key or registration_shared_secret?

[richvdh]

I'm not really sure what more we can do if you choose to overwrite your config so that synapse can't find its database any more?

[richvdh]

Warn of the consequences;

replacing the configuration is done by dpkg - we don't have an opportunity to add a warning.

don't automatically restart synapse and warn about this;

Not restarting synapse on package upgrade would be more surprising to most users; we used to get frequent bug reports about it until this was fixed.

provide the new config, but disabled fe. as homeserver.yaml.dpkg-new;

That would mean that synapse wouldn't work out-of-the-box, which again would be more surprising to most users.

including a sufficient mention of this in docs, something like 'if you want to revert to or use a clean config do this, but note that you'll have downtime and users logged out';

I'm a bit unconvinced that anyone would read any documentation like this.

perhaps provide a current default config somewhere to download

This is #1433, and certainly something I want to address.

or info on how to extract it or just say 'don't do this!'?

I don't really know what you mean by this.

I see that the default behavior is moving the original config to homeserver.yaml.dpkg-old and putting the new one as the default, which is the opposite to what people want. That might be a debian thing though.

It is a debian thing.

Basically, on this, I understand your pain, but it's hard to do much about it.

One thing that I think might help is to suggest that people put config overrides in a file other than homeserver.yaml, so that they are preserved on upgrade.

[neilisfragile]

Okay this is bad situation and I'm sorry you lost your keys - the impact of the log outs will be reduced by features like https://github.com/matrix-org/matrix-doc/blob/55ebf7dc3275822ad0126546baef4b1361418baf/proposals/1466-soft-logout.md

In terms of concrete todos #1433 seems to cover it - I don't think there is much else we can do here. Closing this issue.