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

The CWRUSAGE_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() and the POSIX threads) there is also CWRUSAGE_THREAD. The BSD::Resource supports the CWRUSAGE_THREAD if it is present but understands nothing more about the POSIX threads themselves. Similarly for CWRUSAGE_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 CW<sys/resource.h>. (In Solaris, this might be CW<sys/rusage.h>).

See also KNOWN ISSUES.

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

        $rlimit = getrlimit($resource);

The CW$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 CWget_rlimits() on how to find out which limits are available, for the exact documentation consult the documentation of your operatgiing system. The two groups (CWNOFILE, CCWOFILE, <OPEN_MAX>) and (CWAS, CWVMEM) are aliases within themselves.

Two meta-resource-symbols might exist

        RLIM_NLIMITS
        RLIM_INFINITY

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

In list context CWgetrlimit() 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 CWSIGXCPU or CWSIGXFSZ, corresponding to the CWRLIMIT_CPU and CWRLIMIT_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 CWKILL 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 and hard resource limits as an object. The object can be queried via methods CWcur and CWmax, the current and maximum resource limits for the CW$resource, respectively.

        $nowpriority = getpriority($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 CWundef (and CW$! is set as usual).

The priorities returned by getpriority() are in the (inclusive) range CWPRIO_MIN...CWPRIO_MAX. The CW$pr_which argument can be any of PRIO_PROCESS (a process) CWPRIO_USER (a user), or CWPRIO_PGRP (a process group). The CW$pr_who argument tells which process/user/process group, 0 signifying the current one.

Usual values for CWPRIO_MIN, CWPRIO_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 CWundef on failure.

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

        $success = setpriority($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 impoite process/process group/user. The priorities handled by setpriority() are [CWPRIO_MIN,CWPRIO_MAX]. A normal user process can only lower its priority (make it more positive).

NOTE: A successful call returns CW1, a failed one CW0.

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: BItimes() is not a real BSD function. It is older UNIX.

        $rlimits = get_rlimits();

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

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:

        $r = get_rlimits();
        print "ok.\n" if ($r->{'RLIM_STACK'} == RLIM_STACK);

•

        Your vendor has not defined BSD::Resource macro ...

The code tried to call getrlimit/setrlimit for a resource limit that your operating system vendor/supplier does not support. Portable code should use get_rlimits() to check which resource limits are defined.

•

        use PRIO..., not "PRIO_..."

getpriority() and setpriority() use symbolic names, not strings, for the constants.

•

        use RLIMIT..., not "RLIMIT_..."

getrlimit() and setrlimit() use symbolic names, not strings, for the constants.

        # 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 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 CWPRIO_MIN is 0 (corresponding to -20) and CWPRIO_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 10.3.2 (and probably earlier) the t/setrlimit.t subtest #8 will fail because of bug(s?) in the setrlimit/getrlimit functionality, setting/getting resource limits on the maximum number of processes (RLIM_NPROC) behaves nonsensically. The bug has been reported to Apple.

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 CW<sys/rusage.h> that the CWixrss and the CWisrss fields are always zero. In SunOS 5.5 and 5.6 the getrusage() leaves most of the fiels 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 1995-2004 Jarkko Hietaniemi All Rights Reserved

This library is free software; you may redistribute it and/or modify it under the same terms as Perl itself.

Jarkko Hietaniemi, CWjhi@iki.fi