PSARC-- I made a grave error in filing this case, which is that I forgot
to update the IAM file to change the state of the case to Waiting Fast
Track after I mailed the proposal out to the case. As a result this
case was not considered yesterday at ARC business. This was a careless
oversight on my part and should in no way reflects on the author of this
case, Chad Mynhier.
I have apologized to Chad, and I would appreciate it very much if the
ARC would allow me to set the timer to tomorrow, Friday, Feb 27 2009, at
close-of-business pacific time. In this way we can avoid further delay.
This case has thus far been non-controversial.
I have altered the IAM file to set the timer to tomorrow; please
let us know if this is not acceptable.
Thanks. Again, please accept my apologies.
-dp
On Sat 14 Feb 2009 at 08:35PM, Dan Price wrote:
>
> I'm pleased to sponsor the following fast-track for Chad Mynhier.
> Chad is enhancing some of the various "stat" commands to support
> a time stamp option in keeping with a similar option provided by
> iostat(1m). This case also defines a uniform specification for
> this option which future cases can reference in order to add
> timestamp options to other "stat" commands.
>
> For those who use the *stat commands to do monitoring over long
> periods of time, this will be a welcome addition.
>
> Please ensure that you CC Chad on all correspondance.
>
> -dp
>
> ---------------------------------------------------------------------------
> Time stamp option for xxstat commands
> Chad Mynhier <cmynhier at gmail.com>
>
> SUMMARY
> This fast-track enhances the mpstat(1M), vmstat(1M), and
> prstat(1M) commands to address an existing RFE[1] requesting an
> option for each of these commands to emit a time stamp. In
> addition, this fast-track enhances the existing time stamp
> functionality in iostat(1M) to print a locale-specific time
> stamp, if the standard date format is requested.
>
> This fast-track also sets a precedent that all "*stat" commands
> that print statistics at a regular interval should implement a
> timestamp feature similar to the one described here.
>
> DETAILS
>
> Overview
>
> Currently, the iostat(1M) and fsstat(1M) commands provide an
> option to print a time stamp in either standard date format or
> traditional Unix time for each set of data. For example:
>
> # iostat -T d 1
> Sat Jan 24 12:34:34 2009
> tty sd0 sd1 sd2 cpu
> tin tout kps tps serv kps tps serv kps tps serv us sy wt id
> 0 8 39 1 24 39 1 22 0 0 0 1 1 0 98
> Sat Jan 24 12:34:35 2009
> 0 220 0 0 0 0 0 0 0 0 0 0 1 0 99
> Sat Jan 24 12:34:36 2009
> 0 92 0 0 0 0 0 0 0 0 0 0 1 0 99
> ^C#
> #
> # fsstat -T u tmpfs 1
> new name name attr attr lookup rddir read read write write
> file remov chng get set ops ops ops bytes ops bytes
> 2.57M 1.63M 25.9K 36.3M 70.8K 26.1M 4.67K 118M 464G 65.8M 253G tmpfs
> 1232818548
> 0 0 0 0 0 0 0 0 0 0 0 tmpfs
> 1232818549
> 0 0 0 0 0 0 0 0 0 0 0 tmpfs
> 1232818550
> 0 0 0 0 0 0 0 0 0 0 0 tmpfs
> ^C#
>
> The mpstat(1M), vmstat(1M), and prstat(1M) commands do not
> provide such an option. This option is useful when gathering
> data to correlate with a transient system or application
> problem whose time is known after the fact. While this problem
> can be worked around by wrapping the command in a script that
> adds a time stamp to the output, it would be better to provide
> consistency with iostat(1M) and fsstat(1M).
>
> Additionally, the time stamp currently emitted by iostat(1M) is
> printed unconditionally in the format specified by the C
> locale. For example:
>
> # locale
> LANG=
> LC_CTYPE="C"
> LC_NUMERIC="C"
> LC_TIME=fr
> LC_COLLATE="C"
> LC_MONETARY="C"
> LC_MESSAGES="C"
> LC_ALL=
> # iostat -T d 3
> Mon Jan 26 12:03:42 2009
> tty sd0 sd1 sd2 cpu
> tin tout kps tps serv kps tps serv kps tps serv us sy wt id
> 0 8 43 1 24 43 1 22 0 0 0 1 1 0 98
> Mon Jan 26 12:03:45 2009
> 0 73 0 0 0 0 0 0 0 0 0 0 0 0 99
> ^C#
>
> We plan to address these issues as follows:
>
> 1. We will add the "[-T d | u]" option to vmstat(1M) and mpstat(1M)
> to mirror that option in iostat(1M) and fsstat(1M).
>
> 2. We will add the the "[-d d | u]" option to prstat(1M) to print a
> time stamp.
>
> 3. We will modify iostat(1M) to print the time stamp in a
> locale-specific format (i.e., printing local time using
> strftime(3C) with the correct format string for the locale.)
>
> There are two possible options for the placement of the
> time stamp in the output. The first option is to emit the
> time stamp on a separate line prior to each set of data. The
> second option is to print the time stamp as an additional column
> in the output. We choose the first option for two reasons.
> The first of these is that there is a precedent in how
> iostat(1M) and fsstat(1M) output the time stamp, as shown
> above. The second is that the time stamp would consume a large
> number of columns, and these utilities are designed (for the
> most part) to fit in 80 columns.
>
> Because the output of mpstat(1M) and prstat(1M) typically
> includes multiple rows, we also have the option of printing one
> time stamp per row. We choose instead to print one time stamp
> per sampling period. This follows the precedent set by
> iostat(1M), and it allows the most data to be printed on a
> single screen. Note that this also applies to those modes of
> prstat(1M) that involve printing two separate sets of data,
> such as per-process and per-zone statistics. Only one
> time stamp is shown in these cases.
>
> It could be argued that the time stamp option only makes sense
> in conjunction with the "-c" option of prstat(1M), given that
> otherwise the time stamp is overwritten with each sample. For
> the sake of completeness, however, we propose implementing the
> time stamp option as an independent option that can be used in
> either mode.
>
> EXAMPLE
>
> This example shows mpstat(1M) output using a standard date
> format time stamp:
>
> # ./mpstat -T d 1
> Sat Jan 24 13:14:59 2009
> CPU minf mjf xcal intr ithr csw icsw migr smtx srw syscl usr sys wt idl
> 0 201 0 77 2219 127 87 3 2 14 1 241 1 1 0 98
> 1 209 0 73 173 26 92 3 2 15 1 246 1 1 0 98
> 2 203 0 105 165 24 94 2 2 14 1 240 1 1 0 98
> 3 207 0 73 177 25 97 3 2 14 1 243 1 1 0 98
> Sat Jan 24 13:15:00 2009
> CPU minf mjf xcal intr ithr csw icsw migr smtx srw syscl usr sys wt idl
> 0 18387 0 3246 10899 111 455 222 82 518 79 17262 71 29 0 0
> 1 24517 0 3635 8790 9 482 210 66 624 86 22727 63 37 0 0
> 2 20614 0 8964 7016 75 460 189 68 573 83 17889 61 39 0 0
> 3 23650 0 13319 5535 21 442 186 69 546 75 17211 64 36 0 0
> Sat Jan 24 13:15:01 2009
> CPU minf mjf xcal intr ithr csw icsw migr smtx srw syscl usr sys wt idl
> 0 21763 0 4950 13096 130 491 187 89 521 46 15574 62 38 0 0
> 1 19702 0 12835 8601 48 478 219 86 482 76 16697 64 36 0 0
> 2 16842 0 11385 9045 62 396 195 69 418 62 14106 74 26 0 0
> 3 29149 0 8459 10058 18 538 216 85 665 103 24843 55 45 0 0
> Sat Jan 24 13:15:02 2009
> CPU minf mjf xcal intr ithr csw icsw migr smtx srw syscl usr sys wt idl
> 0 27958 0 10245 21765 125 657 212 127 817 144 21165 46 52 0 3
> 1 26394 0 8131 20657 28 703 196 118 847 142 19734 45 52 0 3
> 2 26998 0 44518 8242 114 478 174 106 778 100 17544 38 61 0 1
> 3 16175 0 4904 20502 33 553 140 115 506 55 14710 54 38 0 8
> ^C#
>
>
> This example shows vmstat(1M) output using a traditional Unix
> time stamp:
>
> # ./vmstat -T u 1
> kthr memory page disk faults cpu
> r b w swap free re mf pi po fr de sr f0 s0 s1 s2 in sy cs us sy id
> 1232820946
> 0 0 0 3666444 2642924 59 824 0 0 0 0 4 0 1 1 -1 2735 974 370 1 1 98
> 1232820947
> 5 0 0 2521192 1540320 5071 92284 0 0 0 0 0 0 22 20 0 54794 83465 1661 63 37 > 0
> 1232820948
> 4 0 0 2495436 1511176 5642 87339 0 0 0 0 0 0 28 28 0 39021 59198 1570 67 33 > 0
> 1232820949
> 3 0 0 2525936 1545120 5767 81687 0 0 0 0 0 0 20 20 0 31799 60757 1697 69 31 > 0
> ^C#
>
>
> This example shows prstat(1M) output using a standard date
> format time stamp (with the -c option):
>
> # amd64/prstat -c -d d -n 5 1
> Sat Jan 24 13:17:42 2009
> PID USERNAME SIZE RSS STATE PRI NICE TIME CPU PROCESS/NLWP
> 6920 mynhier 5760K 4664K sleep 0 4 0:00:00 0.5% dmake/1
> 13928 mynhier 25M 24M sleep 0 4 0:00:00 0.3% dmake/1
> 9229 mynhier 21M 17M cpu3 0 4 0:00:00 0.3% dmake/1
> 1504 noaccess 101M 76M sleep 59 0 0:43:09 0.0% java/18
> 9315 root 3772K 2568K cpu1 39 0 0:00:00 0.0% prstat/1
> Total: 95 processes, 256 lwps, load averages: 6.93, 4.99, 2.46
> Sat Jan 24 13:17:43 2009
> PID USERNAME SIZE RSS STATE PRI NICE TIME CPU PROCESS/NLWP
> 6920 mynhier 5840K 4748K sleep 0 4 0:00:00 0.5% dmake/1
> 9229 mynhier 22M 21M sleep 38 4 0:00:00 0.4% dmake/1
> 9528 mynhier 15M 12M run 0 4 0:00:00 0.4% acomp/1
> 9430 mynhier 10M 9628K sleep 0 4 0:00:00 0.4% dmake/1
> 13928 mynhier 25M 24M sleep 38 4 0:00:00 0.3% dmake/1
> Total: 114 processes, 274 lwps, load averages: 6.91, 4.99, 2.47
> #
>
> This example shows a screenshot of the above command without
> the -c option. (There's little difference to be seen between
> the previous example and this one, but this one is included in
> order to explicitly show the placement of the time stamp in
> this mode.):
>
> Sat Jan 24 13:20:25 2009
> PID USERNAME SIZE RSS STATE PRI NICE TIME CPU PROCESS/NLWP
> 4992 mynhier 29M 26M cpu2 0 4 0:00:00 1.1% ube/1
> 537 mynhier 13M 12M sleep 39 4 0:00:01 1.1% dmake/1
> 5045 mynhier 45M 41M run 0 4 0:00:00 0.4% ube/1
> 5145 mynhier 21M 12M run 0 4 0:00:00 0.1% iropt/1
> 386 mynhier 30M 29M sleep 47 4 0:00:00 0.1% dmake/1
> Total: 144 processes, 302 lwps, load averages: 6.76, 5.72, 3.17
>
> This example shows iostat(1M) output using the French locale:
>
> # ./iostat -T d 3
> lundi 26 janvier 2009 12 h 08 EST
> tty sd0 sd1 sd2 cpu
> tin tout kps tps serv kps tps serv kps tps serv us sy wt id
> 0 8 43 1 24 43 1 22 0 0 0 1 1 0 98
> lundi 26 janvier 2009 12 h 08 EST
> 0 76 0 0 0 0 0 0 0 0 0 0 0 0 100
> ^C#
>
> DOCUMENTATION
>
> This fast-track proposes adding the following text to the man
> pages for mpstat(1M), vmstat(1M), and prstat(1M), with the
> exception that the prstat(1M) shall be listed as "-d":
>
> -T u | d
> Emit a time stamp.
>
> Specify u for a printed representation of the internal
> representation of time. See time(2). Specify d for
> standard date format. See date(1).
>
> Additionally, the option should be added to the SYNOPSIS
> section of the man pages for each of these.
>
> (The iostat(1M) man page should be changed to read, "See
> date(1)." rather than "See ctime(3C)." to reflect the
> localization change.)
>
> REFERENCES
>
> [1] would like a timestamp option like iostat has added to mpstat, vmstat
> and prstat
> (http://bugs.opensolaris.org/view_bug.do?bug_id=4775687)
> ---------------------------------------------------------------------------
>
> --
> Daniel Price, Solaris Kernel Engineering http://blogs.sun.com/dp
--
Daniel Price, Solaris Kernel Engineering http://blogs.sun.com/dp
