[apaddobs]

 use sigtrap qw (handler signal_handler normal-signals);

If I run a process that has a sleep (1000), and kill the program using "kill pid", the signal is trapped by virtue of the line above. However, if I kill it using "kill -9 pid", the signal is not trapped.

How do i trap that?

[jon.kiparsky]

I believe that what -9 means is precisely that it can't be trapped. That's always how I've understood the situation, at least.
It's a last-resort signal: it'll always work, but it's always messy.

[sepp2k]

jon.kiparsky, on 15 January 2013 - 05:26 PM, said:

It's a last-resort signal: it'll always work, but it's always messy.

Not always. There are some cases where a process can't be killed (if, say, a syscall is hanging because of a buggy driver or faulty hardware).

That's only tangentially relevant though - you're right that SIGKILL can't be trapped.

[apaddobs]

So if I have a plugin runnin a perl program and the java code fails with FATAL:null, with a SIGKILL , then I cannot trap it. Is that right ?

[jon.kiparsky]

Correction noted. Would it be better to say, "if anything short of powercycle will work, kill -9 will work"?

[dsherohman]

apaddobs, on 15 January 2013 - 05:43 PM, said:

So if I have a plugin runnin a perl program and the java code fails with FATAL:null, with a SIGKILL , then I cannot trap it. Is that right ?

Yes. Your process never actually receives the SIGKILL, so it can't handle or respond to it in any way. While most other signals are requests for the target process to do something, SIGKILL is handled directly by the kernel and tells it to stop giving processing time to the target process and to de-allocate all of the target's resources as soon as it is safe to do so.

jon.kiparsky, on 15 January 2013 - 05:44 PM, said:

Correction noted. Would it be better to say, "if anything short of powercycle will work, kill -9 will work"?

Yes and no...

While a process is waiting on an I/O operation (a process status of 'D' in top or ps), the kernel will basically just ignore it, just in case aborting the I/O would cause problems. If the I/O call never completes, then kill -9 doesn't have any effect on it and nothing short of a power cycle will work. However, it sometimes does happen that a process will recover from an I/O hang, such as if it's trying to read from an NFS-mounted volume when the network is down, then the network comes back. In this case, the kill -9 will have its normal effect once the I/O call is able to complete.

So, yes in that, for as long as kill -9 doesn't work, nothing else will; no in that something else may happen which allows the process to continue, at which point kill -9 will work normally.