From: Chuck Lever <chuck.lever@oracle.com>
To: Lans Carstensen <Lans.Carstensen@dreamworks.com>
Cc: NFS list <linux-nfs@vger.kernel.org>
Subject: Re: [PATCH 4/4] nfs-utils: nfs-iostat.py autofs cleanup and option to sort by ops/s
Date: Wed, 26 Aug 2009 11:47:18 -0400 [thread overview]
Message-ID: <08E0F2AC-2747-4F98-9686-CF665490B0F1@oracle.com> (raw)
In-Reply-To: <4A954C4D.3060701@dreamworks.com>
On Aug 26, 2009, at 10:53 AM, Lans Carstensen wrote:
> Greetings! Comments inline. Chuck Lever wrote:
>> Hi Lans-
>> I was thinking about this feature a little more. Would it be
>> interesting to allow sorting by other metrics besides ops/s ?
>> That's another feature that the "top" command has that this new
>> option doesn't.
>> I also don't like the bare integer options. What if, at some
>> point, we want "-4" and "-6" for IPv6 support (not that we do, but
>> what if)? Plus, having the bare integers doesn't give any clues
>> about what the number means.
>> In which case I propose:
>> --sort=ops --list=<n>
>> which provides more flexibility and helps make the options more
>> self-documenting.
>
> Again, thanks for looking this over.
>
> I had thought about the sorting a little bit, but decided that every
> case we cared about devolved to sorting by ops/s. We'd use a tool
> like this in an operational group where they've lost control of a
> particular rogue workload in the mix of thousands of other
> workloads, and the operational group is trying to figure out what
> workload is generating a disproportionate amount of traffic (ops/
> s). If someone wanted to extend the implementation further where
> leaving "--sort" unadorned sorts by ops and using "--
> sort=readdirplus" or somesuch, I'd be all for it - but it didn't
> seem worth the work for our use. So I'd encourage that to be
> adopted as-is and if someone wants to put the extra effort into it
> I'm all for it.
Well, the problem is once we choose a command line option, it's hard
to change it later. I think having two choices, like --sort=ops and --
sort=def[ault] would be fine for now, and give us room to grow later.
> For the bare integer options I was thinking in terms of the existing
> "tail" and "head" implementations - that's what made the most sense
> to me and the couple of others looking over this here. Changing it
> to "--list=<n>" is fine if that suites you and others more. I can't
> quite fathom going after IPv4 or IPv6 stats with this, as /proc/self/
> mountstats a) doesn't expose any relevant data that would be IPv4/
> IPv6 oriented, and b) really never would as that's at an
> implementation layer below NFS ops. So the existing patch makes
> most sense to me, but if you and others want it changed I'd be fine
> w/ doing it.
The modern command line option on head/tail is --lines=<n>, which is
also an OK name for this option.
IPv4/IPv6 was just an example... some applications, like ssh, use "-4"
and "-6" already, which makes the use of -<n> ambiguous, I think.
>> On Aug 26, 2009, at 12:59 AM, Lans Carstensen wrote:
>>> commit 062577877bd3f1d8da1edeb889a09d380da83364
>>> Author: Lans Carstensen <Lans.Carstensen@dreamworks.com>
>>> Date: Tue Aug 25 21:53:54 2009 -0700
>>>
>>> Adds --sort option to display mount point stats sorted by ops/s
>>> Adds -<n> option to only display stats for first <n> mount points
>>> E.g. the use of "--sort -1" should be useful in seeing stats for
>>> only the mountpoint with the highest ops/s.
>>>
>>> diff --git a/tools/nfs-iostat/nfs-iostat.py b/tools/nfs-iostat/nfs-
>>> iostat.py
>>> index 9f3b3eb..5df9bfb 100644
>>> --- a/tools/nfs-iostat/nfs-iostat.py
>>> +++ b/tools/nfs-iostat/nfs-iostat.py
>>> @@ -20,7 +20,7 @@ along with this program; if not, write to the
>>> Free Software
>>> Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
>>> 02111-1307 USA
>>> """
>>>
>>> -import sys, os, time
>>> +import sys, os, time, re
>>>
>>> Iostats_version = '0.3'
>>>
>>> @@ -353,6 +353,12 @@ class DeviceData:
>>> print '\t%7.3f' % rtt_per_op,
>>> print '\t%7.3f' % exe_per_op
>>>
>>> + def ops(self, sample_time):
>>> + sends = float(self.__rpc_data['rpcsends'])
>>> + if sample_time == 0:
>>> + sample_time = float(self.__nfs_data['age'])
>>> + return (sends / sample_time)
>>> +
>>> def display_iostats(self, sample_time, which):
>>> """Display NFS and RPC stats in an iostat-like way
>>> """
>>> @@ -421,6 +427,11 @@ def print_iostat_help(name):
>>> print ' If one or more <mount point> names are specified,
>>> statistics for only these'
>>> print ' mount points will be displayed. Otherwise, all NFS
>>> mount points on the'
>>> print ' client are listed.'
>>> + print
>>> + print ' You can also specify "--sort" to sort the NFS mount
>>> points by ops/second,'
>>> + print ' and specify a number of mount points to return with -
>>> <num>, e.g. -1.'
>>> + print ' For example, use of "--sort -1" will iterate only
>>> showing you the stats'
>>> + print ' for the mount point with the highest ops/second.'
>>>
>>> def parse_stats_file(filename):
>>> """pop the contents of a mountstats file into a dictionary,
>>> @@ -446,7 +457,10 @@ def parse_stats_file(filename):
>>>
>>> return ms_dict
>>>
>>> -def print_iostat_summary(old, new, devices, time, ac):
>>> +def print_iostat_summary(old, new, devices, time, ac, sortbyops,
>>> entrycount):
>>> + diff_stats = {}
>>> + count = 1
>>> +
>>> if old:
>>> # Trim device list to only include intersection of old and
>>> new data,
>>> # this addresses umounts due to autofs mountpoints
>>> @@ -455,15 +469,34 @@ def print_iostat_summary(old, new, devices,
>>> time, ac):
>>> devicelist = devices
>>>
>>> for device in devicelist:
>>> + count += 1
>>> stats = DeviceData()
>>> stats.parse_stats(new[device])
>>> if not old:
>>> stats.display_iostats(time, ac)
>>> + if (count>entrycount):
>>> + return
>>> else:
>>> old_stats = DeviceData()
>>> old_stats.parse_stats(old[device])
>>> - diff_stats = stats.compare_iostats(old_stats)
>>> - diff_stats.display_iostats(time, ac)
>>> + diff_stats[device] = stats.compare_iostats(old_stats)
>>> + if not sortbyops:
>>> + diff_stats[device].display_iostats(time, ac)
>>> + if (count>entrycount):
>>> + return
>>> +
>>> + if old and sortbyops:
>>> + # We now have compared data and can print a comparison
>>> + # ordered by mountpoint ops per second
>>> + count = 1
>>> +
>>> + devicelist.sort(key=lambda x: diff_stats[x].ops(time),
>>> reverse=True)
>>> +
>>> + for device in devicelist:
>>> + count += 1
>>> + diff_stats[device].display_iostats(time, ac)
>>> + if (count>entrycount):
>>> + return
>>>
>>> def list_nfs_mounts(givenlist, mountstats):
>>> """return a list of NFS mounts given a list to validate or
>>> @@ -497,6 +530,8 @@ def iostat_command(name):
>>> which = 0
>>> interval_seen = False
>>> count_seen = False
>>> + sortbyops = False
>>> + entrycount = sys.maxint
>>>
>>> for arg in sys.argv:
>>> if arg in ['-h', '--help', 'help', 'usage']:
>>> @@ -507,6 +542,19 @@ def iostat_command(name):
>>> print '%s version %s' % (name, Iostats_version)
>>> return
>>>
>>> + if arg in ['-s', '--sort', 'sort']:
>>> + sortbyops = True
>>> + # sorted display infers a loop, default to 1 second
>>> + if not interval_seen:
>>> + interval = 1
>>> + interval_seen = True
>>> + continue
>>> +
>>> + stop_re = re.compile('-[0-9]+')
>>> + if stop_re.match(arg):
>>> + entrycount = int(arg.lstrip('-'))
>>> + continue
>>> +
>>> if arg in ['-a', '--attr']:
>>> which = 1
>>> continue
>>> @@ -546,12 +594,12 @@ def iostat_command(name):
>>> sample_time = 0.0
>>>
>>> if not interval_seen:
>>> - print_iostat_summary(old_mountstats, mountstats, devices,
>>> sample_time, which)
>>> + print_iostat_summary(old_mountstats, mountstats, devices,
>>> sample_time, which, sortbyops, entrycount)
>>> return
>>>
>>> if count_seen:
>>> while count != 0:
>>> - print_iostat_summary(old_mountstats, mountstats,
>>> devices, sample_time, which)
>>> + print_iostat_summary(old_mountstats, mountstats,
>>> devices, sample_time, which, sortbyops, entrycount)
>>> old_mountstats = mountstats
>>> time.sleep(interval)
>>> sample_time = interval
>>> @@ -562,7 +610,7 @@ def iostat_command(name):
>>> count -= 1
>>> else:
>>> while True:
>>> - print_iostat_summary(old_mountstats, mountstats,
>>> devices, sample_time, which)
>>> + print_iostat_summary(old_mountstats, mountstats,
>>> devices, sample_time, which, sortbyops, entrycount)
>>> old_mountstats = mountstats
>>> time.sleep(interval)
>>> sample_time = interval
>>>
>>>
>>>
>>> --
>>> To unsubscribe from this list: send the line "unsubscribe linux-
>>> nfs" in
>>> the body of a message to majordomo@vger.kernel.org
>>> More majordomo info at http://vger.kernel.org/majordomo-info.html
>> --
>> Chuck Lever
>> chuck[dot]lever[at]oracle[dot]com
>
--
Chuck Lever
chuck[dot]lever[at]oracle[dot]com
next prev parent reply other threads:[~2009-08-26 15:47 UTC|newest]
Thread overview: 7+ messages / expand[flat|nested] mbox.gz Atom feed top
2009-08-26 4:59 [PATCH 4/4] nfs-utils: nfs-iostat.py autofs cleanup and option to sort by ops/s Lans Carstensen
2009-08-26 14:37 ` Chuck Lever
2009-08-26 14:53 ` Lans Carstensen
2009-08-26 15:47 ` Chuck Lever [this message]
2009-08-26 16:57 ` Trond Myklebust
2009-08-26 17:24 ` Chuck Lever
-- strict thread matches above, loose matches on Subject: below --
2009-09-15 4:57 Lans Carstensen
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=08E0F2AC-2747-4F98-9686-CF665490B0F1@oracle.com \
--to=chuck.lever@oracle.com \
--cc=Lans.Carstensen@dreamworks.com \
--cc=linux-nfs@vger.kernel.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox