Changes committed to git://oss.sgi.com/nathans/pcp.git
src/pmdas/linux/cgroups.c | 185 +++++++++++++++++++++++++++++++++++++++++-
src/pmdas/linux/cgroups.h | 7 -
src/pmdas/linux/clusters.h | 7 +
src/pmdas/linux/dynamic.c | 25 +++--
src/pmdas/linux/dynamic.h | 8 -
src/pmdas/linux/filesys.c | 35 +++++++
src/pmdas/linux/filesys.h | 4
src/pmdas/linux/help | 8 +
src/pmdas/linux/indom.h | 47 ++++++----
src/pmdas/linux/pmda.c | 198 +++++++++++++++++++++++++++++++++++++++------
src/pmdas/linux/proc_pid.c | 98 ++++++++++------------
src/pmdas/linux/proc_pid.h | 9 ++
src/pmdas/linux/root_linux | 18 ++++
13 files changed, 529 insertions(+), 120 deletions(-)
commit 91bfbf2332996877472e2560c0100640228772ee
Author: Nathan Scott <nathans@xxxxxxxxxx>
Date: Wed Jun 2 11:22:33 2010 +1000
In-progress commit of cgroup metrics for Linux kernel PMDA.
This makes the following changes to pmdalinux, as we move toward
properly supporting cgroups:
- keep track of mount options provided to filesystems (needed to
tell which cgroup subsystems in use for a cgroup mount point).
- add a filesys.readonly metric now that this is trivial to do.
- refactor the proc metric code so that we can use subsets of all
PIDs instead of all PIDs (to be used later by cgroup task metrics)
- added initial metrics for cgroups - subsystems and mounts.
- switch indom definitions from cpp macros to enum like clusters,
add in comments explaining what each one is.
commit deaba427710144f687529bc1eb0d72a11c4d6487
Author: Nathan Scott <nathans@xxxxxxxxxx>
Date: Wed Jun 2 08:48:05 2010 +1000
Remove unnecessary Linux PMDA wide cluster refresh in init code.
This is undesirable behaviour when running in local context mode,
and from my testing there's no reason to be doing this - it just
slows pmdalinux startup and burns CPU - we end up (re)doing only
the bits we need during the following fetch/indom lookups anyway.
commit 3d1720cb93876bc28a831d2a310cc01750fdba82
Author: Nathan Scott <nathans@xxxxxxxxxx>
Date: Wed Jun 2 08:28:47 2010 +1000
Ensure proc_cpuinfo node_indom element is initialised, initially.
Otherwise there's an earth-shattering kaboom when we fire up pminfo
in local context mode...
Starting program: pminfo -L -K clear -K add,60,./pmda_linux.so,linux_init
-f kernel.all.load
Program received signal SIGSEGV, Segmentation fault.
0xb7dfb9c9 in refresh_proc_stat (proc_cpuinfo=0xb7e11290,
proc_stat=0xb7e10640) at proc_stat.c:117
117 n = proc_cpuinfo->node_indom->it_numinst * sizeof(unsigned
long long);
(gdb) p proc_cpuinfo->node_indom
$1 = (pmdaIndom *) 0x0
|