Tom Parrott
Tom Parrott
We've been tightening up the timesouts recently, or rather adding some, as there was a distinct lack of them previously (which is rather problematic for a network daemon).
> I'd be happy to try increasing timeouts to see if that helps, if they're configurable? They're not at this time, although they are pretty lenient. I'll identify the parts...
OK so looking afresh, I can see that currently `lxc exec` doesn't actually have any keepalive timeouts on it (its certainly another candidate to have them added though). I added...
@stgraber maybe we can step up using your new approach just for exec? I'm also wondering if we should see about the possibility of pushing the PTY allocation into `forkexec`...
@cjwatson the events websocket (used for operations) heartbeat is currently 10s with 3 missed "pongs" required, which does also log if it timeouts: https://github.com/lxc/lxd/blob/34d0d2e0b5c8f3c9f0904adaf8c8609f0fe271c2/lxd/events/common.go#L43-L84
@cjwatson I would be interested to see the debug log output (either from`/var/snap/lxd/common/lxd/logs/lxd.log` or from `lxc monitor --pretty` ) when this next occurs to try and correlate the order that...
@cjwatson also interesting that the error `Error: write unix @->/var/snap/lxd/common/lxd/unix.socket: i/o timeout` suggests that the timeout is occurring writing stdin data to the unix socket (rather than it being abruptly...
Perfect thanks @cjwatson , there's the problem: ``` DBUG[03-11|17:15:33] Failed getting exec control websocket reader, killing command interactive=true project=default PID=4124212 err="websocket: close 1006 (abnormal closure): unexpected EOF" instance=xenial-lptest-596f3e10-6577-4d71-aa34-4b9527c4ecc0 ``` Which...
The question is why is the control socket from the client getting closed.