Dump process core without killing the process

Question

Is there a way to get a core dump (or something similar) for a process without actually killing the processes? I have a multithreaded python process running on an embedded system. And I want to be able to get a snapshot of the process under normal conditions (ie with the other processes required to be running), but I don't have enough memory to connect gdb (or run it under gdb) without the python process being the only one running.

I hope this question makes sense.

Solution 1

The usual trick is to have something (possibly a signal like SIGUSR1) trigger the program to fork(), then the child calls abort() to make itself dump core.

from os import fork, abort
(...)
def onUSR1(sig, frame):
    if os.fork == 0:
        os.abort

and during initialization

from signal import signal, SIGUSR1
from wherever import onUSR1
(...)
signal.signal(signal.SIGUSR1, wherever.onUSR1)

Used this way, fork won't consume much extra memory because almost all of the address space will be shared (which is also why this works for generating the core dump).

Once upon a time this trick was used with a program called undump to generate an executable from a core dump to save an image after complex initialization; emacs used to do this to generate a preloaded image from temacs.

Solution 2

You could try using gcore. Is that an option for you?

Comments

[Gilles 'SO- stop being evil']

If this is only while you're debugging, have you considered something crazy like swap on an NFS file or a network block device?

[KatyB]

At some point gcore was a stand-alone program but I don't think it's part of the gdb package anymore - however you can run gdb --pid=<PID> and then use it's gcore command to dump a core file. gcore.c is a fairly simple program that is easily googlable if you want something lighter weight.

[PSkocik]

Doesn't quite work for multithreaded processes, unfortunately :/