Users getting disconnected frequently

We have Jitsi meet installed on Ubuntu 18.04.3 LTS
There were around 60 participants in total 2 conferences and they kept getting disconnected.
We have All the services running on same server.
ChannelLastN is set to 2
All users were on Chrome

The JICFO log has a very high count of estimated video streams. Is there something wrong with our configuration?

Jicofo 2020-03-03 19:57:02.796 INFO: [141] org.jitsi.jicofo.JitsiMeetConferenceImpl.log() Expiring channels for: on: Bridge[, relayId=null, region=null]
Jicofo 2020-03-03 19:57:02.796 INFO: [141] org.jitsi.jicofo.JitsiMeetConferenceImpl.log() Removed participant: true,
Jicofo 2020-03-03 19:57:02.796 INFO: [70] org.jitsi.jicofo.Bridge.log() Removing 389 video streams on video streams: 0 diff: 69707 (estimated: 69707)
Jicofo 2020-03-03 19:57:02.902 INFO: [74] org.jitsi.jicofo.xmpp.FocusComponent.handleConferenceIq().401 Focus request for room:

I think you’re right that the video stream count is way off. The diff is reset every time jicofo receives updated stats from the bridge – I suspect it’s not receiving it. Which version of the bridge are you using? Is the machine overloaded?


1 Like

Thanks Boris. Good to know we are thinking in the right direction. We had not enabled Videobridge stats A couple of follow up questions - which might help us understand why we see “no socket available” error in our logs.

  1. This estimated count - does jicofo or jvb provision sockets for this count? And with this high number - will OS run out of sockets and hence the error and disconnects? Or the number is just for the bridge to know and has nothing to do with provisioning or blocking sockets?
  2. We will enable stats with pubsub and XMPP, when we do that - When reset fires - with its estimation correction also release any sockets or anything else its provisoned for earlier?

To answer your question - machine was not overloaded as such when we saw the numbers but sometimes the RAM usage on prosody goes super high and topples the server. We are on JVB version 1


A few comments. The “no socket available” messages are most probably harmless and a red herring (they do NOT refer to any of the jitsi processes being unable to allocate a socket).

The video stream count or diff have no relation to the number of sockets. The bridge generally uses a single UDP socket. It may run out of file descriptors, but this is unlikely (it uses a small number of FDs per participant, they definitely grow linearly).

The video stream count issue only affects load balancing. If you only have one jitsi-videobridge instance, it will have no effect.

We recommend using the MUC mode as opposed to pubsub for the stats. We also recommend updating the bridge to 2.1. If you still see issues with jvb 2.1, we can get a lot of diagnostics output out of it and work from there.


1 Like

Thanks Boris,

This clears up a bunch of queries. However, we do see this “SEVERE” log item in JVB logs which I was trying to connect to this stream count issue.

JVB 2020-03-03 16:18:22.221 SEVERE: [3120959] org.jitsi.impl.neomedia.RTPConnectorOutputStream.log() Failed to send a packet to target / No active socket.

We get a bunch of these almost at the time when the users report disconnect or issues with the conference.

Also, is JVB version 2.1 in stable mode now that we can run it in live?

Yes, it has been running on for a few months now.

Thanks. We will try JVB 2.1 on our QA server.

It would be very helpful if we could get a hint or possible reason for the frequent “No active socket” error. I can share any other log that you would like to see.

hey @amritanshtambi, how is it going? is upgrading to jvb 2.1 solve the problem?

Is this problem solved. if so then can you please help us with some hints ?

We have a new set of issue on new version. Jitsi Videobridge itself stops working after a few minutes.