Very Computer

Hello,

I'd like to know the reason defunct processes are caused.

We have defunct processes being caused because of the following logic

Use vfork to create child process
The child process uses execlp to launch a seperate binary.
The parent process waits in a loop and calls waitpid() to determine if
the child process has terminated.
If the child process has not terminated in a predefined time interval
the parent process issues a kill -15 signal to kill the child process.

We observed that the child process thus killed leaves a trail of a
defunct process.

The parent process did not have a signal handler to trap any signals
from this errant child process. Could this be a problem.

Could some one explain why this could be happening and how can I solve
this issue.

Thank you in advance
Maya

Maya,

The parent process is waiting to see if the child process
exits...it doesn't so the parent kills the child.  The defuncts
you are seeing are processes that were waiting for a condition
to be satisfied by your child process (i.e. the child's children).
Since the child process is being forced to terminate with a "kill"
it is leaving that condition unsatisfied and thus the "defunct"
process will not go away because the condition they are waiting
for is never met.  You need to see which process are becoming
defunct and use truss and snoop and the like to see what
condition they are waiting for.

  --Vince

   -------------------------------------------------------------------
   | Vince Arcuri   http://www.obs.net   Open Business Systems, Inc. |
   |  Disclaimer:   My thoughts & opinions are exclusively my own.   |
   -------------------------------------------------------------------

Originally, we had code in the listener to catch the SIGCHLD
signal such as: 'signal(SIGCHLD, (sigHandler)reapchild);', where
reapchild called waitpid(). For some reason this wouldn't work
under Solaris (although it worked fine on other Unix platforms).

We found the solution to the problem is to replace the signal(...)
function call with a call to 'sigignore(SIGCHLD);'. Ignoring the
signal allows children to terminate without becoming defunct.
(This behaviouris actually mentioned in the man pages.)

Hope this helps.

Mike.

--
Mike Biggin, Database Services
DENR, South Australian Government

Phone: +61 8 8204 9217; Fax: +61 8 8204 9017

Unless explicitly attributed, the opinions expressed are personal
and not those of DENR or the South Australian Government.