Aled Sage created BROOKLYN-484:
----------------------------------
Summary: JBoss7 entity restart fails (launch ssh session returns
before process running?)
Key: BROOKLYN-484
URL: https://issues.apache.org/jira/browse/BROOKLYN-484
Project: Brooklyn
Issue Type: Bug
Reporter: Aled Sage
Priority: Minor
With version 0.11.0-rc1...
We've seen a failure of the {{restart}} effector for {{JBoss7Server}}. The
post-launch step failed (waiting for the url to be reachable/responsive).
Unfortunately there's no additional debugging information available - the VMs
are gone, and the debug log is not available.
However, I've identified a reason why this might happen.
On {{start}}, the {{JBoss7SshDriver.launch}} script will redirect stdout/stderr
to a file named {{console}}, and will then wait for that file to say 'starting'.
Importantly, there is an old comment saying:
{noformat}
// We wait for evidence of JBoss running because, using SshCliTool,
// we saw the ssh session return before the JBoss process was fully
running
// so the process failed to start.
{noformat}
On {{restart}}, it stops the process, and then calls {{JBoss7SshDriver.launch}}
again. However, it appends to the file {{console}}. Therefore when it checks if
the file says 'starting' it will return immediately. This means the ssh session
could return before the JBoss process was fully running.
A solution would be to change launch, to first move the previous {{console}}
file. This would mean the subsequent calls to the {{launch}} script would wait
for the process to be running.
This same problem would also apply to other entities, such as
{{TomcatSshDriver.launch}}.
A way to reproduce this would probably be to repeatedly call the {{restart}}
effector (waiting for serviceUp to be true again between each). It almost
always works - I've personally only seen this failure once.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
