Re: [Wrapper-user] wrapper not restarting java process

[Leif M. <le...@ta...>]

Patrick,
    I have a customer running about 10 copies of the Wrapper (3.0.1) 
under Solaris.
It has been working there without any problems for close to a year.  I 
am only aware
of one JVM crash but the Wrapper restarted it correctly.

    It sounds like you have already done what I would have asked.   
Turning on debug
output.  That will tell me more information about exactly what is 
happening and why
the Wrapper is not noticing that the JVM process has disappeared.

    This is surprising because the Wrapper actually uses two methods 
that would both
have to be failing before what you describe could be happening.
    1) The Wrapper launched the JVM by forking its process and then 
calling the JVM.
The parent process then is able to notice almost immediately if the 
forked process dies.
    2) The Wrapper pings the JVM every 5 seconds so it can detect when 
the JVM is
frozen.  1) will only detect a crash.

    Either one of the above should be able to detect that the JVM has 
died.   That makes
me wonder if the Wrapper process itself has somehow become frozen.   
This is something
that I have never even once seen however.
    If it were to become frozen then the JVM is designed to quit and 
allow the Wrapper
to relaunch it in the event that it ever stops receiving ping requests 
from the Wrapper.
This would end up looking exactly like what you are seeing, so it is an 
obvious
possibility.    Enabling the DEBUG output will confirm this, and 
hopefully give some
clue as to why it happened.

Cheers,
Leif

Patrick Woodworth wrote:

>Using wrapper 3.0.4 on Solaris I've seen on two
>separate boxes the 'java' process disappear, but the
>'wrapper' process is still running (according to ps). 
>The wrapper log unfortunately had *nothing* unusual,
>not even a shutdown message.  On one box this occurred
>after 3 hours of running, on another box it took 15
>days.  I've since changed the conf file to spit out
>DEBUG logging, but have yet to reproduce the problem. 
>Has anyone else seen anything like this?
>  
>