Date: Fri, 10 Mar 2000 03:09:55 GMT From: lists@efinley.com (Elliot Finley) To: Bhishan Hemrajani <bhishan@cytosine.dhs.org> Cc: freebsd-questions@FreeBSD.ORG Subject: Re: Code that crashes 3.4-Stable Message-ID: <38c96656.29361402@mail.afnetinc.com> In-Reply-To: <200003100251.SAA05266@cytosine.dhs.org> References: <200003100251.SAA05266@cytosine.dhs.org>
next in thread | previous in thread | raw e-mail | index | archive | help
True, I could impose those limits, but the point is that the system
crashes when it starts to swap... Should this be the case? How
SHOULD FreeBSD handle this?
In reality, I wouldn't WANT the system to swap, and if I was running
low on RAM, I would add more... But if I didn't catch it in time, I
would rather have the system take a BIG performance hit rather than
crash...
On Thu, 9 Mar 2000 18:51:38 -0800 (PST), Bhishan Hemrajani wrote:
>It does work when there are no limits on the users.
>
>I recommend that everyone audits their users using /etc/login.conf.
>
>I have my users limited to 32 processes, and 16m per process.
>
>It didn't crash my system when I was logged in as a limited user.
>
>--bhishan
>
>P.S, after editing /etc/login.conf you have to 'cap_mkdb =
/etc/login.conf'
>
>
>> I'm in the process of stress testing a system for a program that I'm
>> writing. This program will be starting lots of processes, so I wrote
>> a short bit a code to test this out.
>>=20
>> The machine is 3.4-stable as of a couple of days ago, PII-350 384M
>> RAM, and 500M swap. The kernel has MAXUSERS 512, which should give us
>> about 8200 process slots. After forking approx. 5100 children
>> processes, the machine runs out of RAM, and starts using swap, after
>> about 64K of swap is used (as shown by top), the machine freezes
>> solid. No core, no logged messages, no socket activity, no messages
>> on console, nothing.
>>=20
>> If someone can see a problem with the code, I would appreciate some
>> pointers. But even if there IS a problem... You shouldn't be able to
>> crash the system with a user program running with no privileges.
>>=20
>> If I take out the 'sleep(30)' from the child process, then this
>> program runs fine, even with the loop counter set to 100,000 (one
>> hundred thousand), because the child processes go away fast enough
>> that I can only get about 400 of them running at any one time. The
>> problem seems to be when the machine runs out of memory, and starts to
>> swap.
>>=20
>> Here is the code, it was compiled with 'g++ -o test test.c'
>> -----------------------------------------------------------
>>=20
>> #include <iostream.h>
>> #include <sys/types.h>
>> #include <sys/wait.h>
>> #include <signal.h>
>> #include <unistd.h>
>> //#include <stdlib.h>
>>=20
>> void sig_child(int i)
>> {
>> int pid;
>> int status;
>>=20
>> while ( (pid =3D wait3(&status, WNOHANG, (struct rusage *) 0)) > 0) =
;
>> }
>>=20
>> main()
>> {
>> cout << "Hello World!" << endl;
>>=20
>> void sig_child(int i);
>>=20
>> signal(SIGCHLD, sig_child);
>>=20
>> int iIsParent;
>>=20
>> for(int i =3D 0; i < 8000; i++) {
>> iIsParent =3D fork();
>> if(iIsParent < 0) { cout << "Error Forking!" << endl; }
>> if(!iIsParent) {
>> cout << "I'm child #" << i << endl;
>> sleep(30);
>> return 0;
>> }
>> }
>>=20
>> sleep(10);
>> cout << "Parent done." << endl;
>> return 0;
>> }
>>=20
>> --=20
>> Elliot (efinley@efinley.com) Weird Science!
>>=20
>>=20
>> To Unsubscribe: send mail to majordomo@FreeBSD.org
>> with "unsubscribe freebsd-questions" in the body of the message
>>=20
>
--=20
Elliot (efinley@efinley.com) Weird Science!
To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-questions" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?38c96656.29361402>