pcp
[Top] [All Lists]

pcp updates

To: pcp@xxxxxxxxxxx
Subject: pcp updates
From: Ken McDonell <kenj@xxxxxxxxxxxxxxxx>
Date: Thu, 14 Nov 2013 07:54:23 +1100
Delivered-to: pcp@xxxxxxxxxxx
User-agent: Mozilla/5.0 (X11; Linux x86_64; rv:24.0) Gecko/20100101 Thunderbird/24.1.0
Changes committed to git://oss.sgi.com/kenj/pcp.git dev

 qa/359                          |    2 
 qa/370                          |    8 ++-
 src/pmdas/linux_proc/proc_pid.c |  106 ++++++++++++++++++++++++++++------------
 3 files changed, 85 insertions(+), 31 deletions(-)

commit 32f7bc63b943e984cbb9817072b232e117aeb515
Author: Ken McDonell <kenj@xxxxxxxxxxxxxxxx>
Date:   Tue Nov 12 10:53:08 2013 +1100

    proc pmda - dodge wchan sometimes not available
    
    Recent thread semantics changes to the proc pmda have been resulting in
    qa/022 failing on some platforms.
    
    This was eventually tracked down to /proc/PID/task/PID/wchan being
    sometimes (unexpectedly) empty, and causing the other proc metrics to
    be unavailable.
    
    Fix this, and retain all the verbose diagnostics used to find this (all
    under -D libpmda AND -D desperate control).

commit 5bd200ba4db811e8e8bfcbffc9d63012c64e4b27
Author: Ken McDonell <kenj@xxxxxxxxxxxxxxxx>
Date:   Fri Nov 8 18:30:46 2013 +1100

    qa/359 - filtering change
    
    email summarizes all that matters ...
    
    Ken asks ...
    > Can anyone explain a recent change that produces this
    >
    > $ pminfo -v proc.psinfo
    > proc.psinfo.cgroups: Unknown or illegal instance identifier
    > proc.psinfo.labels: Unknown or illegal instance identifier
    >
    > on some systems and not others?
    
    And Nathan replies ...
    >> This is exporting /proc/PID/cgroup and /proc/PID/attr/current
    >> values.  It looks like systems where these files are empty or
    >> otherwise return an error on open, will cause the above inst
    >> error.  I guess thats because all the proc.psinfo metrics use
    >> the same ("all processes") indom.
    >>
    >> It can be safely filtered, thanks.

commit e18b37d50dfd50217b1479e8ffc0bef3df0721b7
Author: Ken McDonell <kenj@xxxxxxxxxxxxxxxx>
Date:   Fri Nov 8 09:01:14 2013 +1100

    qa/370 - fix regression
    
    Need proc.nprocs -> proc.nproc mapping in the pmdumplog filter after
    sar2pcp changes in commit 48e373f0609a0a0804bc052d5e9dec41a5a2928e
    so output matches the boatload of 370's "out" files.
    
    Should be proc.nprocs so the sar2pcp commit is correct, but all the
    old systems will match the old "out" files, so simplest to reverse
    the effects of the commit in the output filter.

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