Second round of changes related to timestamp arithmetic and OpenIndiana
QA failures.
On OpenIndiana QA is now down to 9 failures ... these are non-trivial,
but smell like QA failures (i.e. problems in the QA tests) rather that
failures in the PCP code.
Changes committed to git://oss.sgi.com/kenj/pcp.git dev
src/libpcp_fault/src/GNUmakefile | 7 -
src/pmdas/shping/Install | 46 +++++++-----
src/pmdas/shping/sample.conf | 7 +
src/pmdas/shping/shping.c | 5 +
src/pmdas/solaris/data.c | 9 +-
src/pmdas/solaris/pmns.hinv | 1
src/pmdas/solaris/solaris.c | 8 +-
src/pmdas/solaris/vnops.c | 11 ++
src/pmdumplog/pmdumplog.c | 30 ++++++--
src/pmlogextract/gram.y | 31 +-------
src/pmlogextract/logger.h | 36 ---------
src/pmlogextract/metriclist.c | 6 -
src/pmlogextract/pmlogextract.c | 146 ++++++++++++++++++++++-----------------
src/pmlogreduce/dometric.c | 2
14 files changed, 180 insertions(+), 165 deletions(-)
commit fe1c497d2151da1063e4fa48e656110bbcc6b463
Author: Ken McDonell <kenj@xxxxxxxxxxxxxxxx>
Date: Tue May 29 06:08:42 2012 +1000
Solaris PMDA - add hinv.nfilesys fix other hinv.* metadata
hinv.nfilesys was missing and QA would lie to use this.
hinv.ncpu and hinv.ndisk should not have "count" semantics.
commit 247a34205bea0937950b9c856c232ef7c52f70b6
Author: Ken McDonell <kenj@xxxxxxxxxxxxxxxx>
Date: Tue May 29 06:07:06 2012 +1000
shping PMDA - try to stop SIGALRM delivery to main thread
The PMDA is unreliable on OpenIndiana because the timeouts intended
to wake the shell command running thread are interrupting the thread
that is communicating with PMCD.
commit 4e80bd0498e9db8083bb7546dec10426fcfc8e27
Author: Ken McDonell <kenj@xxxxxxxxxxxxxxxx>
Date: Mon May 28 15:47:07 2012 +1000
pmdumplog - fix timestamp arithmetic
Similar to pmlogextract, pmdumplog has been flawed from the outset
in the choice of doubles to hold some timestamps. This involves
integer to double conversions from struct timeval and __pmTimeval
timestamps which cannot be accurate.
Fixed by removing all the double timestamp arithmetic and operating
exclusively on tv_sec and tv_usec.
commit c5eb29d621c44d8959f167de611c891feac16f1b
Author: Ken McDonell <kenj@xxxxxxxxxxxxxxxx>
Date: Thu May 24 06:45:38 2012 +1000
pmlogreduce - fix missing %p prefix
commit ded6cef86a4a85db1a6b2e664a33d7c25bd30e9e
Author: Ken McDonell <kenj@xxxxxxxxxxxxxxxx>
Date: Wed May 23 10:12:46 2012 +1000
shpinf PMDA - fix botched format string
commit 891a28a0e7e622e2ddd6477b00f4051798d68733
Author: Ken McDonell <kenj@xxxxxxxxxxxxxxxx>
Date: Wed May 23 09:42:04 2012 +1000
shping PMDA - fine tuning for nslookup command
For some platforms (like OpenIndiana) nslookup is not on pmcd's
$PATH.
Extend the existing auto-configuration mechanism for sample.conf.
commit 882f75ca948def0e342210017f1cbdaa5fb32768
Author: Ken McDonell <kenj@xxxxxxxxxxxxxxxx>
Date: Tue May 22 06:35:43 2012 +1000
pmlogextract - dead code removal
After the timestamp changes, a chunk of dead code and unused data
structure elements were culled.
commit 5cfdf702196ea476a1545b0c38939526a451b3ce
Author: Ken McDonell <kenj@xxxxxxxxxxxxxxxx>
Date: Tue May 22 06:16:32 2012 +1000
pmlogextract - fix timestamp arithmetic
pmlogextract has been flawed from the outset in the choice of
doubles to hold all timestamps. This involves integer to double
conversions from struct timeval and __pmTimeval timestamps which
cannot be accurate.
Eventually it blew up on OpenIndiana where timestamps that should
have been equal were slightly different and the code took a segv ...
this was killing QA as I could not even make the archives needed
for some of the QA tests.
Fixed by removing all the double timestamp arithmetic and operating
exclusively on tv_sec and tv_usec.
commit 1b1f2d3d720f04605af269459093f18d8d563d18
Author: Ken McDonell <kenj@xxxxxxxxxxxxxxxx>
Date: Tue May 22 06:15:52 2012 +1000
libpcp_fault - fix .h dependencies in GNUmakefile
|