On 12/8/11 3:53 PM, Mikolaj Golub wrote:
On Thu, 08 Dec 2011 13:30:36 -0500 John Baldwin wrote:
JB> On 12/3/11 3:58 PM, Mikolaj Golub wrote:
>>
>> On Mon, 28 Nov 2011 13:30:11 -0500 John Baldwin wrote:
>>
>> JB> On Thursday, November 24, 2011 3:54:06 pm Mikolaj Golub wrote:
>> >> Author: trociny
>> >> Date: Thu Nov 24 20:54:06 2011
>> >> New Revision: 227956
>> >> URL: http://svn.freebsd.org/changeset/base/227956
>> >>
>> >> Log:
>> >> usr.bin/procstat
>> >>
>> >> Add -l flag to display resource limits.
>> >>
>> >> PR: bin/161257
>> >> Reviewed by: kib
>> >> MFC after: 2 weeks
>>
>> JB> Thanks for doing this! Did you consider making the procstat -l
output use
>> JB> "pretty" output similar to the output of /usr/bin/limits? For
example,
>> JB> using "infinity" instead of -1 and using humanize_number() for
finite limits
>> JB> that are in units of bytes?
>>
>> I tried several variants, from one where for rlimit names rlimit_ident
>> constants from sys/resource.h are used and units are printed as suffixes:
>>
>> PID COMM RLIMIT SOFT HARD
>> 46150 zsh cpu 100000S infinity
>> 46150 zsh fsize infinity infinity
>> 46150 zsh data 524288kB 524288kB
>> 46150 zsh stack 65536kB 65536kB
>> 46150 zsh core 9765625kB 9765625kB
>> 46150 zsh rss infinity infinity
>> 46150 zsh memlock infinity infinity
>> 46150 zsh nproc 5547 5547
>> 46150 zsh nofile 11095 11095
>> 46150 zsh sbsize infinity infinity
>> 46150 zsh vmem infinity infinity
>> 46150 zsh npts infinity infinity
>> 46150 zsh swap infinity infinity
>>
>> to one where rlimit names are the same as in limits(1) and units are
printed
>> in separate column:
>>
>> PID COMM RLIMIT SOFT HARD UNIT
>> 48885 zsh cputime 100000 infinity secs
>> 48885 zsh filesize infinity infinity bytes
>> 48885 zsh datasize 524288k 524288k bytes
>> 48885 zsh stacksize 65536k 65536k bytes
>> 48885 zsh coredumpsize 95367M 95367M bytes
>> 48885 zsh memoryuse infinity infinity bytes
>> 48885 zsh memorylocked infinity infinity bytes
>> 48885 zsh maxprocesses 5547 5547
>> 48885 zsh openfiles 11095 11095
>> 48885 zsh sbsize infinity infinity bytes
>> 48885 zsh vmemoryuse infinity infinity bytes
>> 48885 zsh pseudo-terminals infinity infinity
>> 48885 zsh swapuse infinity infinity bytes
>>
>> Personally I like the first variant as the most compact and the easiest to
>> maintain but would be glad to learn what other think about this or may be
have
>> other suggestions.
>>
>> A couple other variations:
>>
>> PID COMM RLIMIT SOFT HARD UNIT
>> 47062 zsh cpu 100000 infinity secs
>> 47062 zsh fsize infinity infinity bytes
>> 47062 zsh data 524288k 524288k bytes
>> 47062 zsh stack 67108864 67108864 bytes
>> 47062 zsh core 9765625k 9765625k bytes
>> 47062 zsh rss infinity infinity bytes
>> 47062 zsh memlock infinity infinity bytes
>> 47062 zsh nproc 5547 5547
>> 47062 zsh nofile 11095 11095
>> 47062 zsh sbsize infinity infinity bytes
>> 47062 zsh vmem infinity infinity bytes
>> 47062 zsh npts infinity infinity
>> 47062 zsh swap infinity infinity bytes
>>
>> PID COMM RLIMIT SOFT HARD UNIT
>> 48798 zsh cputime 100000 infinity secs
>> 48798 zsh filesize infinity infinity bytes
>> 48798 zsh datasize 524288k 524288k bytes
>> 48798 zsh stacksize 65536k 65536k bytes
>> 48798 zsh coredumpsize 95367M 95367M bytes
>> 48798 zsh memoryuse infinity infinity bytes
>> 48798 zsh memorylocked infinity infinity bytes
>> 48798 zsh maxprocesses 5547 5547
>> 48798 zsh openfiles 11095 11095
>> 48798 zsh sbsize infinity infinity bytes
>> 48798 zsh vmemoryuse infinity infinity bytes
>> 48798 zsh pseudo-terminals infinity infinity
>> 48798 zsh swapuse infinity infinity bytes
JB> Hmm, I would stick as close to limits output as possible. I would
JB> consider duplicating the unit field in each of soft and hard, so you
JB> end up with something like this:
JB> PID COMM RLIMIT SOFT HARD
JB> 48798 zsh cputime 100000 secs infinity secs
JB> 48798 zsh filesize infinity kb infinity kb
JB> 48798 zsh datasize 524288 kb 524288 kb
JB> etc.
Ok.
JB> (Things like 'openfiles' is simply more intuitive than 'nofile' (no
JB> file?, huh? oh, num open files.. (except not all users will make the
JB> last step there).
Then why do we have so non-intuitive rlimit_ident names?
It looks like they are used only in procfs_rlimit.c. Do procfs(5) users always
make that last step with 'nofile'? :-)
Well, I suspect it's best not to change the names in procfs in case
there are existing binaries that parse the output of that file
(unfortunately).
Is it possible to change rlimit_ident names? Just to ones that are used by
limit(1) or (if they look too long) to something like below:
Hmm, I have no idea what other things might use rlimit_ident. Probably
not many. (Also, for fun, note that the 'ulimit' and 'limit' built-in
commands in sh and csh also have their own sets of names, fun!) I would
maybe add a rlimit_names[] (just leave rlimit_ident alone), and give
that the names from limits(1), and change both procstat and sh's
'ulimit' command to use those.
--
John Baldwin
_______________________________________________
svn-src-all@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/svn-src-all
To unsubscribe, send any mail to "svn-src-all-unsubscr...@freebsd.org"
