Mailing-List: contact issues-help@cloudstack.apache.org; run by ezmlm
Precedence: bulk
Reply-To: dev@cloudstack.apache.org
Date: Sat, 10 May 2014 22:16:23 +0000 (UTC)
From: "David Scott (JIRA)" <jira@apache.org>
To: cloudstack-issues@incubator.apache.org
Message-ID: <JIRA.12713281.1399629169512.324326.1399760183689@arcas>
In-Reply-To: <JIRA.12713281.1399629169512@arcas>
References: <JIRA.12713281.1399629169512@arcas>
Subject: [jira] [Created] (CLOUDSTACK-6621) Intermittent failure when
 management server connects to hypervisor via ssh
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: 7bit

David Scott created CLOUDSTACK-6621:
---------------------------------------

             Summary: Intermittent failure when management server connects to hypervisor via ssh
                 Key: CLOUDSTACK-6621
                 URL: https://issues.apache.org/jira/browse/CLOUDSTACK-6621
             Project: CloudStack
          Issue Type: Bug
      Security Level: Public (Anyone can view this level - this is the default.)
          Components: Management Server
    Affects Versions: 4.5.0
         Environment: I'm running a management server locally (from master c/s 6511b96088af75b7e37a5f8b0cce609b006021fb) and attempting to add a CentOS 6.4 host via the libvirt/KVM plugin
            Reporter: David Scott


The management server attempts to verify the presence of kvm by using ssh to talk to the host via sshExecuteCmd:

https://github.com/apache/cloudstack/blob/master/utils/src/com/cloud/utils/ssh/SSHCmdHelper.java#L63

The work is done by sshExecuteCmdOneShotWithExitCode (called in a loop)

https://github.com/apache/cloudstack/blob/master/utils/src/com/cloud/utils/ssh/SSHCmdHelper.java#L94

This function waits until either EXIT_STATUS or EOF is set, and then calls sshSession.getExitStatus. For me this fails with a NullPointerException
{noformat}
ERROR [c.c.u.s.SSHCmdHelper] (581293855@qtp-1130716142-0:ctx-57482224 ctx-b2286596 ctx-e73d2678) Ssh executed failed
java.lang.NullPointerException
{noformat}

I added some extra logging and I believe that EOF can be set *before* EXIT_STATUS i.e. before the exit status is ready. I think if we want there to be a readable exit code, we must wait for EXIT_STATUS.

Perhaps my system has unusual timing, but this hits me every time. Note the ssh command is repeated multiple times (e.g. 3) which could hide the bug for many people.

I've prepared a simple patch which fixes the issue and makes ssh reliable for me. I'll upload it to review board shortly.


--
This message was sent by Atlassian JIRA
(v6.2#6252)