Jarkko Hietaniemi > BSD-Resource-1.2907 > BSD::Resource



Annotate this POD

Related Modules

By perlmonks.org


Open  0
View/Report Bugs
Module Version: 1.2907   Source  


BSD::Resource - BSD process resource limit and priority functions


        use BSD::Resource;

        # the process resource consumption so far

        ($usertime, $systemtime,
         $maxrss, $ixrss, $idrss, $isrss, $minflt, $majflt, $nswap,
         $inblock, $oublock, $msgsnd, $msgrcv,
         $nsignals, $nvcsw, $nivcsw) = getrusage($ru_who);

        $rusage = getrusage($ru_who);

        # the process resource limits

        ($nowsoft, $nowhard) = getrlimit($resource);

        $rlimit = getrlimit($resource);

        $success = setrlimit($resource, $newsoft, $newhard);

        # the process scheduling priority

        $nowpriority = getpriority($pr_which, $pr_who);

        $success = setpriority($pr_which, $pr_who, $priority);

        # The following is not a BSD function.
        # It is a Perlish utility for the users of BSD::Resource.

        $rlimits = get_rlimits();



        ($usertime, $systemtime,
         $maxrss, $ixrss, $idrss, $isrss, $minflt, $majflt, $nswap,
         $inblock, $oublock, $msgsnd, $msgrcv,
         $nsignals, $nvcsw, $nivcsw) = getrusage($ru_who);

        $rusage = getrusage($ru_who);

        # $ru_who argument is optional; it defaults to RUSAGE_SELF

        $rusage = getrusage();

The $ru_who argument is either RUSAGE_SELF (the current process) or RUSAGE_CHILDREN (all the child processes of the current process) or it maybe left away in which case RUSAGE_SELF is used.

The RUSAGE_CHILDREN is the total sum of all the so far terminated (either successfully or unsuccessfully) child processes: there is no way to find out information about child processes still running.

On some systems (those supporting both getrusage() with the POSIX threads) there can also be RUSAGE_THREAD. The BSD::Resource supports the RUSAGE_THREAD if it is present but understands nothing more about the POSIX threads themselves. Similarly for RUSAGE_BOTH: some systems support retrieving the sums of the self and child resource consumptions simultaneously.

In list context getrusage() returns the current resource usages as a list. On failure it returns an empty list.

The elements of the list are, in order: index name meaning usually (quite system dependent)

         0      utime           user time
         1      stime           system time
         2      maxrss          maximum shared memory or current resident set
         3      ixrss           integral shared memory
         4      idrss           integral or current unshared data
         5      isrss           integral or current unshared stack
         6      minflt          page reclaims
         7      majflt          page faults
         8      nswap           swaps
         9      inblock         block input operations
        10      oublock         block output operations
        11      msgsnd          messages sent
        12      msgrcv          messaged received
        13      nsignals        signals received
        14      nvcsw           voluntary context switches
        15      nivcsw          involuntary context switches

In scalar context getrusage() returns the current resource usages as a an object. The object can be queried via methods named exactly like the middle column, name, in the above table.

        $ru = getrusage();
        print $ru->stime, "\n";

        $total_context_switches = $ru->nvcsw + $ru->nivcsw;

For a detailed description about the values returned by getrusage() please consult your usual C programming documentation about getrusage() and also the header file <sys/resource.h>. (In Solaris, this might be <sys/rusage.h>).

See also "KNOWN ISSUES".


        ($nowsoft, $nowhard) = getrlimit($resource);

        $rlimit = getrlimit($resource);

The $resource argument can be one of

        $resource               usual meaning           usual unit

        RLIMIT_CPU              CPU time                seconds

        RLIMIT_FSIZE            file size               bytes

        RLIMIT_DATA             data size               bytes
        RLIMIT_STACK            stack size              bytes
        RLIMIT_CORE             coredump size           bytes
        RLIMIT_RSS              resident set size       bytes
        RLIMIT_MEMLOCK          memory locked data size bytes

        RLIMIT_NPROC            number of processes     1

        RLIMIT_NOFILE           number of open files    1
        RLIMIT_OFILE            number of open files    1
        RLIMIT_OPEN_MAX         number of open files    1

        RLIMIT_LOCKS            number of file locks    1

        RLIMIT_AS               (virtual) address space bytes
        RLIMIT_VMEM             virtual memory (space)  bytes

        RLIMIT_TCACHE           maximum number of       1
                                cached threads

        RLIMIT_AIO_MEM          maximum memory locked   bytes
                                for POSIX AIO
        RLIMIT_AIO_OPS          maximum number          1
                                for POSIX AIO ops

What limits are available depends on the operating system. See below for get_rlimits() on how to find out which limits are available, for the exact documentation consult the documentation of your operating system. The two groups (NOFILE, OFILE, OPEN_MAX) and (AS, VMEM) are aliases within themselves.

Two meta-resource-symbols might exist


RLIM_NLIMITS being the number of possible (but not necessarily fully supported) resource limits, see also the get_rlimits() call below. RLIM_INFINITY is useful in setrlimit(), the RLIM_INFINITY is often represented as minus one (-1).

In list context getrlimit() returns the current soft and hard resource limits as a list. On failure it returns an empty list.

Processes have soft and hard resource limits. On crossing the soft limit they receive a signal (for example the SIGXCPU or SIGXFSZ, corresponding to the RLIMIT_CPU and RLIMIT_FSIZE, respectively). The processes can trap and handle some of these signals, please see "Signals" in perlipc. After the hard limit the processes will be ruthlessly killed by the KILL signal which cannot be caught.

NOTE: the level of 'support' for a resource varies. Not all the systems

        a) even recognise all those limits
        b) really track the consumption of a resource
        c) care (send those signals) if a resource limit is exceeded

Again, please consult your usual C programming documentation.

One notable exception for the better: officially HP-UX does not support getrlimit() at all but for the time being, it does seem to.

In scalar context getrlimit() returns the current soft limit. On failure it returns undef.


        $nowpriority = getpriority($pr_which, $pr_who);

        # the default $pr_who is 0 (the current $pr_which)

        $nowpriority = getpriority($pr_which);

        # the default $pr_which is PRIO_PROCESS (the process priority)

        $nowpriority = getpriority();

getpriority() returns the current priority. NOTE: getpriority() can return zero or negative values completely legally. On failure getpriority() returns undef (and $! is set as usual).

The priorities returned by getpriority() are in the (inclusive) range PRIO_MIN...PRIO_MAX. The $pr_which argument can be any of PRIO_PROCESS (a process) PRIO_USER (a user), or PRIO_PGRP (a process group). The $pr_who argument tells which process/user/process group, 0 signifying the current one.

Usual values for PRIO_MIN, PRIO_MAX, are -20, 20. A negative value means better priority (more impolite process), a positive value means worse priority (more polite process).


        $success = setrlimit($resource, $newsoft, $newhard);

setrlimit() returns true on success and undef on failure.

NOTE: A normal user process can only lower its resource limits. Soft or hard limit RLIM_INFINITY means as much as possible, the real hard limits are normally buried inside the kernel and are very system-dependent.

NOTE: Even the soft limit that is actually set might be lower than what requested for various reasons. One possibility is that the actual limit on a resource might be controlled by some system variable (e.g. in BSD systems the RLIMIT_NPROC can be capped by the system variable maxprocperuid, try sysctl -a kern.maxprocperuid), or in many environments core dumping has been disabled from normal user processes. Another possibility is that a limit is rounded down to some alignment or granularity, for example the memory limits might be rounded down to the closest 4 kilobyte boundary. In other words, do not expect to be able to setrlimit() a limit to a value and then be able to read back the same value with getrlimit().


        $success = setpriority($pr_which, $pr_who, $priority);

        # NOTE! If there are two arguments the second one is
        # the new $priority (not $pr_who) and the $pr_who is
        # defaulted to 0 (the current $pr_which)

        $success = setpriority($pr_which, $priority);

        # The $pr_who defaults to 0 (the current $pr_which) and
        # the $priority defaults to half of the PRIO_MAX, usually
        # that amounts to 10 (being a nice $pr_which).

        $success = setpriority($pr_which);

        # The $pr_which defaults to PRIO_PROCESS.

        $success = setpriority();

setpriority() is used to change the scheduling priority. A positive priority means a more polite process/process group/user; a negative priority means a more impolite process/process group/user. The priorities handled by setpriority() are [PRIO_MIN,PRIO_MAX]. A normal user process can only lower its priority (make it more positive).

NOTE: A successful call returns 1, a failed one 0.

See also "KNOWN ISSUES".


        use BSD::Resource qw(times);

        ($user, $system, $child_user, $child_system) = times();

The BSD::Resource module offers a times() implementation that has usually slightly better time granularity than the times() by Perl core. The time granularity of the latter is usually 1/60 seconds while the former may achieve submilliseconds.

NOTE: The current implementation uses two getrusage() system calls: one with RUSAGE_SELF and one with RUSAGE_CHILDREN. Therefore the operation is not `atomic': the times for the children are recorded a little bit later.

NOTE: times() is not imported by default by BSD::Resource. You need to tell that you want to use it.

NOTE: times() is not a "real BSD" function. It is older UNIX.


        use BSD::Resource qw{get_rlimits};
        my $limits = get_rlimits();

NOTE: This is not a real BSD function. It is a convenience function introduced by BSD::Resource.

get_rlimits() returns a reference to hash which has the names of the available resource limits as keys and their indices (those which are needed as the first argument to getrlimit() and setrlimit()) as values. For example:

        use BSD::Resource qw{get_rlimits};
        my $limits = get_rlimits();
        for my $name (keys %$limits) {
          my ($soft, $hard) = BSD::Resource::getrlimit($limits->{$name});
          print "$name soft $soft hard $hard\n";

Note that a limit of -1 means unlimited.



        # the user and system times so far by the process itself

        ($usertime, $systemtime) = getrusage();

        # ditto in OO way

        $ru = getrusage();

        $usertime   = $ru->utime;
        $systemtime = $ru->stime;

        # get the current priority level of this process

        $currprio = getpriority();


In AIX (at least version 3, maybe later also releases) if the BSD compatibility library is not installed or not found by the BSD::Resource installation procedure and when using the getpriority() or setpriority(), the PRIO_MIN is 0 (corresponding to -20) and PRIO_MAX is 39 (corresponding to 19, the BSD priority 20 is unreachable).

In HP-UX the getrusage() is not Officially Supported at all but for the time being, it does seem to be.

In Mac OS X a normal user cannot raise the RLIM_NPROC over the maxprocperuid limit (the default value is 266, try the command sysctl -a kern.maxprocperuid).

In NetBSD RLIMIT_STACK setrlimit() calls fail.

In Cygwin RLIMIT_STACK setrlimit calls fail. Also, setrlimit() RLIMIT_NOFILE/RLIMIT_OFILE/RLIMIT_OFILE calls return success, but then the subsequent getrlimit calls show that the limits didn't really change.

Because not all UNIX kernels are BSD and also because of the sloppy support of getrusage() by many vendors many of the getrusage() values may not be correctly updated. For example Solaris 1 claims in <sys/rusage.h> that the ixrss and the isrss fields are always zero. In SunOS 5.5 and 5.6 the getrusage() leaves most of the fields zero and therefore getrusage() is not even used, instead of that the /proc interface is used. The mapping is not perfect: the maxrss field is really the current resident size instead of the maximum, the idrss is really the current heap size instead of the integral data, and the isrss is really the current stack size instead of the integral stack. The ixrss has no sensible counterpart at all so it stays zero.


Copyright 1995-2013 Jarkko Hietaniemi All Rights Reserved

This module free software; you can redistribute it and/or modify it under the terms of the Artistic License 2.0 or GNU Lesser General Public License 2.0. For more details, see the full text of the licenses at <http://www.perlfoundation.org/artistic_license_2_0>, and <http://www.gnu.org/licenses/gpl-2.0.html>.


Jarkko Hietaniemi, jhi@iki.fi

syntax highlighting: