-
-
Notifications
You must be signed in to change notification settings - Fork 2.1k
Synapse users lose access to all rooms after logging out and in again #11568
Comments
I suggest you share your homeserver logs, including the full exception - it's unlikely we'll be able to shed much light without them. |
Thanks for your reply! Please find the server log at info level (w/ some IP addresses redacted) at: https://gist.github.com/ginkel/d954e86015f514df9a49a19dc0178806. The login operation (that ends up with a basically empty client) starts at time index Please let me know if you need additional details / more verbose logs. |
I can easily create a log at |
Thanks @ginkel. I'm surprised to read that restoring the DB from backup didn't resolve the problem. Not much to go on in those logs, I'm afraid. No logging in the /sync responses, and they all ended up returning with status 200 OK. If possible, could you try logging out and in again on a client and capturing the server logs at DEBUG level? I'm particularly interested in the logs for the first call to /sync after a login (which should have no
Don't have an authoritative answer for you here. You could try emailing them to me ( |
I also wanted to note that this looks like |
Me too. ;-) I have run the Element Android client in a debugger and AFAICS room membership info gets synced to the client, but somehow has no effect:
I sent you the debug log via Matrix. |
That room is actually version 1. But v6 rooms (that's the newest I have) also have the homeserver name on the room id. |
Thanks for this. I don't see an invite as such yet, perhaps your homeserver isn't federating due to this problem? I'll check again tomorrow, but it might be worth forwarding a copy over email too as a fallback. |
@DMRobertson Weird... Anyway, I just sent you a copy via email. Thanks for your support! :-) |
The logs you've sent to @DMRobertson might shed more light, but in the meantime it might be worth a quick sanity-check on your database. Could you try this query? SELECT room_id
FROM current_state_events AS c
INNER JOIN events AS e USING (room_id, event_id)
WHERE
c.type = 'm.room.member'
AND c.state_key = '@tg:tgbyte.de'
AND c.membership = 'join'; It should return a list of rooms your server thinks your user is a member of. |
Interesting. That query comes back empty for my user as well as any other user I tried...
|
iiinteresting. are you able to DM me? I'm |
As it turned out, a database locale problem caused |
related: #6696 |
Description
I was somewhat uncertain whether this may quailify as a support issue, but considering it involves catastrophic data loss, where the synapse instance is unable to locate rooms despite them being present in the database, I am inclined to open a bug report.
I am operating a Synapse (currently v1.48) instance along with an instance of mxisd. The trouble started when I migrated mxisd to ma1sd and - due to a misconfigured URL - Synapse could no longer reach ma1sd. This caused Element (Android) clients to be logged out.
Upon fixing the configuration error and logging in the client again, I was greeted with a room list that just contains one "Empty room" entry.
I do have a backup of the database and file system of the Matrix instance that pre-dates the configuration glitch.
Unfortunately, restoring it to that state does not remediate the situation, i.e., rooms are still gone.
I verified in the database that rooms are present in the
rooms
table and that thecurrent_state_events
table contains anm.room.create
event for those rooms.In the log I see messages like:
But what I wrote above about the room being present on the DB holds true for the room mentioned in the error message.
I can't rule out that the underlying problem has been lingering for a while and was only uncovered by the auth issues that logged out some clients. The Synapse instance is > 2 years old, so it has undergone all database migrations that occurred over time as the software evolved.
I'd appreciate your assistance to figure out what may be causing this and to eventually restore access for all users' rooms.
Thanks,
Thilo
P.S.: Most channels were encrypted if that makes any difference
Steps to reproduce
I am afraid that I do not have clear clear instructions how to reproduce the problem except from what I wrote above.
Version information
If not matrix.org:
Version:
{"server_version":"1.48.0","python_version":"3.8.12"}
Install method: Docker
The text was updated successfully, but these errors were encountered: