dun
error connecting to virsh console: unable to get tty attributes
What steps will reproduce the problem?
Enable a serial console on a kvm virtual machine. Then edit conman.conf to access this console through libvirt:
console name="foo" dev="/usr/bin/virsh console foo"
What is the expected output? What do you see instead?
Instead of successfully spawning a "virsh console" process that the client can interact with, the following output is observed in the console log for foo:
<ConMan> Console [foo] connected to "virsh" (pid 23456) at 06-10 12:34.
12:34:56.789: error : vshRunConsole:88 : unable to get tty attributes: Invalid argument
Connected to domain foo
Escape character is ^]
<ConMan> Console [foo] disconnected from "virsh" (pid 23456) after 0s at 06-10 12:34.
What version of the software are you using? On what operating system?
conman-0.2.5 libvirt-bin-0.8.1-1 (Debian Squeeze)
Please provide any additional information below.
The virsh error is triggered by a failed call to tcgetattr(STDIN_FILENO,&ttyattr) in libvirt-0.8.1/tools/console.c.
Original issue reported on code.google.com by chris.m.dunlap on 10 Jun 2010 at 9:27
So yeah. Any progress on this one? Any work-arounds to give virsh the tty it thinks it wants?
Original comment by [email protected] on 5 Sep 2012 at 3:14
I haven't made any progress on this yet. I'd like to get 0.2.8 released before too much longer. I need to set aside some time to finish that. If there's time, I'll try to work this into 0.2.8.
Original comment by chris.m.dunlap on 5 Sep 2012 at 7:22
I actually managed to get a little something working here. I created a script I call "vcon":
#!/bin/bash
vm=$1
pty=$(virsh ttyconsole $vm)
#socat - $pty,raw,echo=0,crnl
exec socat - $pty,echo=0,ocrnl=1
and then configure conman to use that with the virtual machine name to connect to its console. I am not at all convinced this survives through a virsh destroy ...; virsh start .... I usually just restart conman when I restart a VM (which is almost never with any luck).
Any advice on how to make that more elegant with conman? Should vcon block if the virtual machine is not up or is there some way for vcon to indicate to conman that it cannot connect to the machine such that conman will periodically retry? Or some other way to prod conman?
Original comment by [email protected] on 5 Sep 2012 at 7:26
Oh, and I'm not positive the ocrnl=1 is actually needed. I had it in there debugging something else and just forgot to remove it. Feel free to season according to taste.
Original comment by [email protected] on 5 Sep 2012 at 7:30
As for vcon, the easy thing to do is just exit if the console is not up. conmand will immediately restart external processes that terminate. If the new process immediately terminates, conmand will wait 1min before the next retry, followed by an exponential backoff up to a maximum of 30min between retries until the process survives for at least a minute (at which point the reconnect timer is set back to 0).
Original comment by chris.m.dunlap on 6 Sep 2012 at 12:22