"internal error character device (null) is not using a PTY" when using virsh console

[Tom Ellis]

After setting up nova and launching an instance, I can't ssh to it and running virsh console shows:

Connected to domain instance-0000000f
Escape character is ^]
error: internal error character device (null) is not using a PTY

Instance is an oneiric tarball:
http://uec-images.ubuntu.com/releases/oneiric/alpha-2a/oneiric-server-uec-amd64.tar.gz

ProblemType: Bug
DistroRelease: Ubuntu 11.04
Package: nova-compute 2011.2-0ubuntu1
ProcVersionSignature: Ubuntu 2.6.38-8.42-server 2.6.38.2
Uname: Linux 2.6.38-8-server x86_64
Architecture: amd64
Date: Thu Jul 7 12:23:38 2011
PackageArchitecture: all
ProcEnviron:
 LANGUAGE=en_US:
 LANG=en_US
 SHELL=/bin/bash
SourcePackage: nova
UpgradeStatus: No upgrade log present (probably fresh install)

[Tom Ellis (tellis)]

After overcoming some network configuration issues, I can now launch instances and ssh to them.

Virsh console still shows this PTY error message though. I noticed the same thing on the lxc console.

[Thierry Carrez (ttx)]

I wonder if it's an issue with Nova (for example, does eucalyptus succeed in the same situation), with libvirt, or with the Ubuntu cloud images...

[Chuck Short (zulcss)]

Tom,

Is this still a problem for you?

Regards
chuck

[Matt Thompson (mattt416)]

Hi All,

I was able to replicate this running --libvirt_type=qemu, and see similar "virsh dumpxml" output as reported in https://www.redhat.com/archives/libvir-list/2009-July/msg00460.html. If I move the <serial type='pty'> tag above <serial type="file"> in libvirt.xml (despite libvirt.xml stating that the order should be "file" first), I can connect to console but cannot see console output using using "euca-get-console-output". Previously, I couldn't connect to console, but could see console output.

Interestingly, but as an aside, when using --libvirt_type=uml, I by default can connect to the console but cannot see console output using "euca-get-console-output".

Anyway, after reading the libvirt-list post pasted above, I'm inclined to think only one <serial> tag should be defined in libvirt.xml, but perhaps someone more familiar w/ libvirt can clarify?

-Matt

[Tom Ellis (tellis)]

@Chuck
Yes, tried todays daily oneiric image on diablo-3 and still see the same issue:

ubuntu@sugarloaf:~/images$ euca-describe-instances
RESERVATION r-vchanq4i proj default
INSTANCE i-00000023 ami-00000008 192.168.3.3 192.168.3.3 running psekey (proj, perce) 0 m1.small 2011-08-02T10:00:06Z nova
ubuntu@sugarloaf:~/images$ ping 192.168.3.3 -c 1
PING 192.168.3.3 (192.168.3.3) 56(84) bytes of data.
64 bytes from 192.168.3.3: icmp_req=1 ttl=64 time=0.155 ms

--- 192.168.3.3 ping statistics ---
1 packets transmitted, 1 received, 0% packet loss, time 0ms
rtt min/avg/max/mdev = 0.155/0.155/0.155/0.000 ms
ubuntu@sugarloaf:~/images$ ssh ubuntu@192.168.3.3 -i ~/creds/psekey.priv
PTY allocation request failed on channel 0

FYI, I'm using VLAN network mode.

euca-get-console-output shows: UnknownError: An unknown error has occurred. Please try your request again.

A much better error was produced in nova-api.log though, attached.

[Tom Ellis (tellis)]

@Matt Interesting, I'll have a play around with it.

[Dave Walker (davewalker)]

@Tom, can you provide an update if that resolved the issue?

Thanks.

[Tom Ellis (tellis)]

So, I've tried the latest oneiric cloud images from today and two days ago and I still get the "PTY allocation request failed on channel 0" connecting via ssh.

With euca-get-console-output <instance id> I still get (from nova-api.log):
(nova.api): TRACE: RemoteError: Error local variable 'fpath' referenced before assignment
(nova.api): TRACE: [u'Traceback (most recent call last):\n', u' File "/usr/lib/pymodules/python2.7/nova/rpc.py", line 232, in _process_data\n rval = node_func(context=ctxt, **node_args)\n', u' File "/usr/lib/pymodules/python2.7/nova/exception.py", line 97, in wrapped\n return f(*args, **kw)\n', u' File "/usr/lib/pymodules/python2.7/nova/compute/manager.py", line 1023, in get_console_output\n output = self.driver.get_console_output(instance_ref)\n', u' File "/usr/lib/pymodules/python2.7/nova/exception.py", line 124, in wrapped\n raise Error(str(e))\n', u"Error: local variable 'fpath' referenced before assignment\n"]

However, on the compute node I can see the correctly booted system with 'virsh -c lxc:/// console <instance-id>'

[Tom Ellis (tellis)]

Also, it seems moving the serial tag makes no difference for what I can see

[Chuck Short (zulcss)]

This should be fixed now, please re-open if you are still having problems.

Regards
chuck

[Tom Haddon (mthaddon)]

I'm seeing this on trusty/icehouse:

root@furud:~# virsh list --all
 Id Name State
----------------------------------------------------
 32 instance-00002bb1 running
 37 instance-00002bb2 running
 4237 instance-00006b86 running
 4238 instance-00006b87 running

root@furud:~# virsh console instance-00006b87
Connected to domain instance-00006b87
Escape character is ^]
error: internal error: character device <null> is not using a PTY

[James Page (james-page)]

Its quite likely that this is intended function - the console output of the instance is mapped to a log file so that 'nova console-log' can be used to retrieve log data via the public API of nova.

[James Page (james-page)]

You can use the following:

virsh console --devname serial1 instance-XXXX

[James Page (james-page)]

However that may not be usable - anyway I think this is by design - end users won't normally have direct access to compute hosts!

[Varhegyi Balazs (varh1i-balazs)]

Hi, I wanted to use "virtsh console instance-xxxx" as described here: http://www.jaredlog.com/?p=1484 for debugging this issue: https://ask.openstack.org/en/question/89583/cant-ping-machine-after-unrescue/ but I encountered this problem. My goal was to connect to the instance and see the console log and get a console to the machine because ssh doesn't work (details in the referenced link)