(2.2.15) Resource limit problems.

From: Scott A Crosby (crosby@qwes.math.cmu.edu)
Date: Fri May 26 2000 - 18:11:10 EDT

  • Next message: Francois romieu: "[uPatch] Compile fixe - msr.c"

    Hello, I tend to run software systems like SML-NJ, or CMUCL. (These are
    software-systems which use heap-style allocation, similar to JAVA.) They
    can suddenly consume a large amount of memory without warning.

    As linux doesn't handle this well, I need to have resource limits designed
    to kill off these processes before they impact system stability.

    I cannot do this..

    For some reason, the DATA or RSS limits (ulimit -Sd or -Sm) will not catch
    the runaway process.

    Only the VM (ulimit -Sv) will catch a runaway sml-nj (Standard ML of New
    Jersey) process: Here's an example strace:

    old_mmap(NULL, 48431104, PROT_READ|PROT_WRITE|PROT_EXEC, MAP_PRIVATE, 3,
                                 0) = -1 ENOMEM (Cannot allocate memory)
    /usr/local/sml: Error -- unable to map 48365568 bytes, errno = 12

    SML-NJ allocates memory as it needs it, occasionally doubling its memory
    consumption when it's doing a garbage collection.

    The other program, CMUCL (Carnegie Mellon University Common Lisp) MMAP's
    all of its heap before it uses it, when I run it under 'ulimit -Sv 40000':

    old_mmap(0x10000000, 268431360, PROT_READ|PROT_WRITE|PROT_EXEC,
        MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS|MAP_NORESERVE, -1, 0) = -1 ENOMEM
             (Cannot allocate memory)

    It's normal heap-allocation sequence is this:

    old_mmap(0x10000000, 268431360, PROT_READ|PROT_WRITE|PROT_EXEC,
            MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS|MAP_NORESERVE, -1, 0) = 0x10000000
    old_mmap(0x28000000, 268431360, PROT_READ|PROT_WRITE|PROT_EXEC,
            MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS|MAP_NORESERVE, -1, 0) = 0x28000000
    old_mmap(0x48000000, 1879048192, PROT_READ|PROT_WRITE|PROT_EXEC,
            MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS|MAP_NORESERVE, -1, 0) = 0x48000000
    old_mmap(0x38000000, 134213632, PROT_READ|PROT_WRITE|PROT_EXEC,
            MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS|MAP_NORESERVE, -1, 0) = 0x38000000
    old_mmap(0x20000000, 134213632, PROT_READ|PROT_WRITE|PROT_EXEC,
            MAP_PRIVATE|MAP_FIXED|MAP_ANONYMOUS|MAP_NORESERVE, -1, 0) = 0x20000000

    As such, it obviously cannot be run under 'ulimit -Sv 40000', or any other
    reasonable 'ulimit -Sv' limit.

    What I wish is to be able to limit the number of pages that a program can
    use, not allocate, so that I can catch runaway SML-NJ or CMUCL processes,
    without impacting their normal operation.

    Doing some grep's through my 2.2.15 code tree, RLIMIT_RSS seems to be used
    in exactly one place, printing out ``/proc/this/stat''. RLIMIT_DATA seems
    to only be checked with brk().

    Any idea how to do get resource limits that will work? (Please CC any
    replies.)

    Scott 

    --
Table of testing results. Here, I test each system in a
runaway-situation under various limits:

                    SML             CMUC
-Sv 50000       kills           won't start
-Sd 40000       won't kill      won't kill
-Sm 40000       won't kill      won't kill

    -
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/

    This archive was generated by hypermail 2b29 : Fri May 26 2000 - 18:16:46 EDT