Perl "Out of memory!" Error signal backtrace

[OnEhIppY]

Is there a way to get a Signal call for the Out of memory! call within a perl script?

Somewhere deep within the thread it gives me that message, yet there seems no way to catch that call, to do a traceback to the whole call structure, to see where things go wrong, like the warn and die have a signal, yet these are not triggered by this particular one.

There are some kill signals about memory, but these seem neither to be triggered.

Anyone an idea how to trace the code point?

[dave_the_m]

The problem is that once the perl interpreter has exhausted all memory, there's very little it can do. There is a reserved fixed buffer to allow it to display the the OOM error itself, but doing anything complex like displaying a Carp::confess()-like stack backtrace would require many memory allocations to generate the display, all of which would themselves fail.

Dave.

[eyepopslikeamosquito]

As LanX might have guessed, I maintain a list of nodes related to memory issues: Re: Perl Memory problem ... (Memory Tools)

From that node, I suggest you start by mulling over each item in the "General Areas to consider when faced with a Memory Leak" section ... while thinking like a detective. :)

The Profiling Memory Usage talk by Perl legend Tim Bunce (featuring his Devel::SizeMe module) is also excellent (it opens with a joke about your "Out of memory!" message I see).

[perlfan]

Just noticed this today, but Test::Valgrind is available as a package in Ubuntu (libtest-valgrind-perl). Valgrind is mentioned in the target page link by eyepopslikeamosquito, but I am pointing out it may offer an expedient way to research the issue on your system.

[LanX]

I'd say you need to track the growth of your memory consumption. Not sure what the codepoint can tell you.

You probably have a memory leak due to recursive data structures.

update

according to splain -v it's non-trappable

<Reveal this spoiler or all in this thread>

Well ...

... probably the trace options of the debugger can help?

If you log them to a file, you should see where they stop.

[perlfan]

> Somewhere deep within the thread

If you reproduce it, post the code here. I am guessing you're making some wrong assumption about threads and how they work in perl; without knowing what you're doing, I can blindly suggest: 1) get it working serially, and 2) if you need to workshare, use fork or better, Parallel::ForkManager. Also, if you're reading in a entire 1 TB file or grabbing a billion rows from mysqld, then there are better ways to do it.