pcp
[Top] [All Lists]

pmdagfs2: updates

To: Nathan Scott <nathans@xxxxxxxxxx>
Subject: pmdagfs2: updates
From: Paul Evans <pevans@xxxxxxxxxx>
Date: Mon, 07 Jul 2014 17:49:10 +0100
Cc: PCP Mailing List <pcp@xxxxxxxxxxx>
Delivered-to: pcp@xxxxxxxxxxx
User-agent: Mozilla/5.0 (X11; Linux x86_64; rv:17.0) Gecko/20130625 Thunderbird/17.0.7
Changes committed to git://github.com/pauljevans/pcp.git dev

 qa/654                       |   2 +-
 qa/654.out                   |   2 +-
 qa/655                       |   5 +-
qa/655.out | 504 +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++-
 src/pmdas/gfs2/control.c     |  19 ++-
 src/pmdas/gfs2/control.h     |   3 +-
 src/pmdas/gfs2/ftrace.c      | 161 ++++++++++-----------
 src/pmdas/gfs2/ftrace.h      |   3 +-
 src/pmdas/gfs2/help          | 233 +++++++++++++-----------------
 src/pmdas/gfs2/pmda.c        |  48 ++++++-
 src/pmdas/gfs2/pmns          |  24 +---
 src/pmdas/gfs2/worst_glock.c | 310 +++++++++++++++++++++++++++------------
 src/pmdas/gfs2/worst_glock.h |  50 +++++--
 13 files changed, 1004 insertions(+), 360 deletions(-)

commit ee4f2f5382a1894ba9c6a2be95fa411f049d7902
Author: Paul Evans <pevans@xxxxxxxxxx>
Date:   Mon Jul 7 16:55:04 2014 +0100

    pmdagfs2: Set default trace_pipe buffer size and allow control

Set a default value for the size of the trace_pipe buffer (per cpu) upon the init of the PMDA. Have also created a control metric which allows the user to
    set a custom size for buffer_size_kb up to 128MB (per cpu).

When there is high load on the GFS2 filesystems mounted there will be many events captured in the trace_pipe, if the buffer is not sufficiently large enough then entries may be overwritten before being read by the PMDA. By
    increasing the size of the buffer we are minimizing the risk of this
    occurring and ensuring we catch as many results as possible.

commit 1a516b3d56cb64fe1998cd6655dfd9d87e8572fb
Author: Paul Evans <pevans@xxxxxxxxxx>
Date:   Mon Jul 7 15:58:18 2014 +0100

    pmdagfs2: Update full help text in order to pass qa 469

Update the formatting of the full help text for the pmdagfs2 metrics in order for each line of the full helptext to fit nicely in normal sized consoles.

    We now pass qa 469 without any warnings :)

    Cheers to Nathan for pointing this one out!

commit 36ca15436ae121ee8aa856b8b9fb614d971c6d67
Author: Paul Evans <pevans@xxxxxxxxxx>
Date:   Mon Jul 7 15:26:43 2014 +0100

    pmdagfs2: Add top10 worst_glocks and stop storing intermediate values

Update worst_glock metric to now list the "top 10" computed worst glock for each GFS2 filesystem mounted on the system. Have at the same time moved to using dynamic pmns generation for these metrics (as is being used with sbstats
    metrics).

Likewise with the ftrace metrics have removed the use of arrays to store the intermediate metric values before they are processed resulting in both reduced
    memory usage and speed increases.

Have updated the QA tests to reflect the changes from the increased number of metrics (now 293) and add an additional test on the worst_glock metrics.

commit 43b8277540e7cd30e117c26c8bbd10f56c64d7a3
Author: Paul Evans <pevans@xxxxxxxxxx>
Date:   Fri Jun 20 15:55:06 2014 +0100

pmdagfs2: ftrace.c Stop storing intermediate variables in array for metrics

Have removed the use of arrays to store the intermediate ftrace data for the
    tracepoint metrics before they are processed.

    This has results in a reduced amount of memory required to process the
tracepoint metrics and at the same time reduced the amount of needless data
    copying.

commit 1bd79a5019df21600c4a3c716fc4409102b88166
Author: Paul Evans <pevans@xxxxxxxxxx>
Date:   Tue May 13 17:45:08 2014 +0100

    pmdagfs2: Enable all GFS2 tracepoints on PMDA load

    Added gfs2_tracepoints_init() function to enable all GFS2 tracepoints
    when pmdagfs2 is enabled in PCP.

commit 9de7a7986517f9af99ab7b18fbaaf1e46c993f9a
Author: Paul Evans <pevans@xxxxxxxxxx>
Date:   Tue May 13 17:43:53 2014 +0100

    pmdagfs2: Fix issue with counting total for gfs2_glock_queue metric

    Fixed an issue where we were not counting a metric value for the total
    number of gfs2_glock_queue tracepoint calls seen.

commit 5127e7f86d8124b10ba0044c0d8ba4f3ddd8b4f2
Author: Paul Evans <pevans@xxxxxxxxxx>
Date:   Fri May 9 19:33:56 2014 +0100

    pmdagfs2: Correct enum naming scheme worst_glock.h

    Correct the naming scheme for the members of an enum in worst_glock.h
    the names hadn't be updated since re-factoring a couple of version ago
    and this brings them inline with the rest of the enum names in the
    header file.

commit bfc7a6e197b2e6ae513caf94480a47a985522bf0
Author: Paul Evans <pevans@xxxxxxxxxx>
Date:   Fri May 9 19:26:24 2014 +0100

    pmdagfs2: Fix variable type in ftrace.c

    Fixed an issue where "int" was used instead of "unsigned int" as the
    the type for major and minor in ftrace.c these should be unsigned int
    to match the value converted in their corresponding sscanf statements.

commit 9c75a60d5f79c8c5144c8944cd0e38068036a67d
Author: Paul Evans <pevans@xxxxxxxxxx>
Date:   Fri May 9 19:11:30 2014 +0100

    pmdagfs2: Fix reading of gfs2_glock_queue tracepoint

    Fix issue where incorrect formatting of the argument for sscanf when
    reading gfs2_glock_queue tracepoint entries would mean that queuing
    events would not be recorded properly.

    Also altered changed the corresponding followed strncmp argument to
    reflect this proper reading of the gfs2_glock_queue tracepoint.

Collection of fixes and updates that I've been working on over the last couple of weeks, there is a few more metric additions I would like to get in for the next release (hopefully committed by the end of the week) but need a bit further
testing first.

Also includes the fixes to pass qa test 469, all of the patches have been
tested and have had covscan run on them. As always please let me know if there
are any issues and feedback is welcome :)

Cheers,

Paul

<Prev in Thread] Current Thread [Next in Thread>