Changes committed to git://oss.sgi.com/pcp/pcp.git dev
man/man1/pmstore.1 | 2
qa/733 | 50 ++++++++
qa/733.out | 239 +++++++++++++++++++++++++++++++++++++++
qa/group | 1
qa/src/20130706.0.bz2 |binary
qa/src/20130706.index |binary
qa/src/20130706.meta |binary
qa/src/GNUlocaldefs | 3
src/pmcollectl/pmcollectl.py | 19 ++-
src/pmdas/linux_proc/GNUmakefile | 7 -
src/pmdas/linux_proc/Install | 29 ++++
src/pmdas/linux_proc/Remove | 23 +++
src/pmdas/linux_proc/cgroups.c | 10 -
src/pmdas/linux_proc/root | 6
src/pmdas/linux_xfs/GNUmakefile | 7 -
src/pmdas/linux_xfs/Install | 29 ++++
src/pmdas/linux_xfs/Remove | 23 +++
src/pmdas/linux_xfs/root | 6
src/pmie/src/binary.sk | 8 -
19 files changed, 440 insertions(+), 22 deletions(-)
commit c5ff24cbfab2ae393599391f99ea05c404bc4159
Author: Nathan Scott <nathans@xxxxxxxxxx>
Date: Tue Sep 24 15:50:32 2013 +1000
Fix pmstore man page typo re pmStore vs pmstore restrictions
commit 32feb0fa4a9f55fe4a7f597fe0f2cce8e23c49df
Author: Nathan Scott <nathans@xxxxxxxxxx>
Date: Tue Sep 24 15:48:18 2013 +1000
Allow opt-out with the default-enabled xfs and proc PMDAs
Nowadays we run pmdaproc and pmdaxfs separately to the main
kernel PMDA on Linux, and under the root user account. As
some people may not wish to run additional daemons as root,
provide a Remove script for each. Of course, someone will
want to undo such an action, at some point, so we'll need to
have Install scripts too. These need "root" PMNS files to
ensure a successful install with all the bits in the right
places, with the right domain numbers and so on.
commit f5e9b634fddf6cd5a3aae2eab6479dd463ca33c4
Author: Joseph White <jpwhite4@xxxxxxxxxxx>
Date: Tue Sep 24 12:53:22 2013 +1000
Fix the recursive expansion of child cgroups
I was investigating how to get the linux_proc pmda to generate cgroup
metrics for child cgroups (in my case to view the cgroups created by
slurm). The source code in cgroup.c in the linux_proc pmda appeared
to support creating metrics for child cgroups, but the pmda was not
producing any of these metrics on my system.
The problem was caused because the full directory path (including the
mount point) is passed to cgroup_scan() when it is called recursively.
The recursively called cgroup_scan() prepends the mount point again to
the path and then fails trying to open the non-existant directory. There
was also a follow-on bug where the cgroupname was not set correctly when
processing a subdirectory.
This change enables the pmda to parse cgroup subdirectories and create
metrics for all the child cgroups.
commit fa78803e15e7e919afcf4fef0376191383ac226c
Author: Nathan Scott <nathans@xxxxxxxxxx>
Date: Tue Sep 24 12:37:01 2013 +1000
Resolve a pmcollectl divide-by-zero picked up by qa/709
Now that pmcollectl is defaulting to local: connections, it
seems we are much more likely to observe a zero delta in the
network in/out packet values. This is causing a div-by-zero
python stack trace in pmcollectl.
Resolve this by taking care to ensure the divisor is never
zero. Also fix a typo in the suppression of loopback for the
network.interface.out metric sum and add a comment suggesting
pmDelProfile would be a better option for tackling this class
of problem (currently works only because lo is presented as
the first entry in /proc/net/dev by the kernel - but actually
multipe loopback devices can be configured, and there may not
be a guarantee that lo0 is always first anyway).
commit c2ce6ba257e4d73fd312d3b33d3d3990952e7214
Merge: e00b973 1481b71
Author: Nathan Scott <nathans@xxxxxxxxxx>
Date: Tue Sep 24 12:29:30 2013 +1000
Merge branch 'dev' of git://oss.sgi.com/kenj/pcp into dev
commit 1481b719c210e588cd8e99116a3d951f66c9a8b2
Author: Ken McDonell <kenj@xxxxxxxxxxxxxxxx>
Date: Sat Sep 21 07:43:05 2013 +1000
pmie bug - archive, interp mode, mark record, ...
Found investigating a pmie core dump on a specific archive.
Problem triggered by an arithmetic expression with two metrics over
the same instance domain, one with INSTANT semantics, the other with
DISCRETE semantics, and an archive with records and in one of
the periods between records, the DISCRETE metric had defined
values, but the INSTANT one did not (although it did have values
earlier and later in the archive).
An loop in the binary.sk template was protected by a guard that was
not strong enough in this case ... fixed and now the expression is
undefined in these circumstances as it should be.
Very specific QA case added.
|