Changes committed to git://git.pcp.io/kenj/pcp dev
Ken McDonell (6):
libpcp/AF.c: reduce chance of problems with signals
pmlogger: fix signal handler deadlock
qa/040: tweak timing for pmlogger
qa/1045: add more diagnostics to try and catch file clobbering
qa/301: fix small tail race
qa/752: expanded coverage, more robust filtering
Stan Cox (1):
Match current parse-datetime.y behavior for 'next *TODAY*'
man/man3/pmaf.3 | 26 +++++++
qa/040 | 13 ++-
qa/1045 | 3
qa/301 | 1
qa/752 | 92 +++++++++++++++++-----------
qa/752.out | 144 ++++++++++++++++++++++++--------------------
qa/src/rtimetest.c | 12 ++-
src/libpcp/src/AF.c | 9 ++
src/libpcp/src/getdate.y | 4 -
src/pmlogger/src/callback.c | 34 +++++++++-
src/pmlogger/src/logger.h | 3
src/pmlogger/src/pmlogger.c | 66 +++++++++++++++++++-
12 files changed, 290 insertions(+), 117 deletions(-)
Details ...
commit 343b79d183721db6b133170d798868bffd900a02
Author: Ken McDonell <kenj@xxxxxxxxxxxxxxxx>
Date: Mon Feb 16 07:08:39 2015 +1100
qa/752: expanded coverage, more robust filtering
commit 8053ade42eed280e803113305a03c398731df0aa
Author: Ken McDonell <kenj@xxxxxxxxxxxxxxxx>
Date: Mon Feb 16 07:05:31 2015 +1100
qa/301: fix small tail race
Need _wait_for_pmlogger after restarting pmcd at the end.
commit c081d8345903cad261b07362c4ec475385c1b0eb
Author: Ken McDonell <kenj@xxxxxxxxxxxxxxxx>
Date: Mon Feb 16 07:04:19 2015 +1100
qa/1045: add more diagnostics to try and catch file clobbering
Still trying to understand how pconf/cpu/load_average sometimes
ends up as an empty file.
commit 90bde5e658acee30830a35abd523907ad4203ee8
Author: Ken McDonell <kenj@xxxxxxxxxxxxxxxx>
Date: Mon Feb 16 07:01:06 2015 +1100
qa/040: tweak timing for pmlogger
With the changes to pmlogger to avoid the signal deadlock, it is
no longer possible to log reliably at a 5msec sample frequency.
(It never really was, but previously we never got out of the
interrupt handler!). Logging at 20msec seems to work fine.
Fine-tune the number of logging samples and the checking iterations
to match.
commit b365c37d9b321cb514fbdb39e4d1dba2028f3ab6
Author: Ken McDonell <kenj@xxxxxxxxxxxxxxxx>
Date: Mon Feb 16 06:54:55 2015 +1100
pmlogger: fix signal handler deadlock
All the AF* callbacks are now async-signal-safe.
Fixes http://oss.sgi.com/bugzilla/show_bug.cgi?id=1104
and addresses the pmlogger issues in
http://oss.sgi.com/bugzilla/show_bug.cgi?id=1069
commit 78cc8d020c59d15d3ced69a52b129abddfe6cc7b
Author: Ken McDonell <kenj@xxxxxxxxxxxxxxxx>
Date: Mon Feb 16 06:48:53 2015 +1100
libpcp/AF.c: reduce chance of problems with signals
As described in http://oss.sgi.com/bugzilla/show_bug.cgi?id=1069
we should restrict ourselves to async-signal-safe routines to be
called from onalarm().
This commit replaces the pmprintf()/pmflush() calls with printf()
(which is no better) but places them behind debug guards ... so
without setting -Daf there are no stdio calls in onalarm().
The free() remains, I need to figure out how to delay that one.
Also the pmaf(3) man page describes the restrictions that must
be applied to the func() passed to pmAFregister(), namely it must
also restrict itself to calling only async-signal-safe routines.
commit b47af16c219f78d96f3962e61d022ca9006ad3a8
Author: Stan Cox <scox@xxxxxxxxxx>
Date: Fri Feb 13 12:23:08 2015 -0500
Match current parse-datetime.y behavior for 'next *TODAY*'
* libpcp/src/getdate.y (__pmGlibGetDate): Consider 'next TODAY' to be next
week.
* qa/src/rtimetest.c: Test more relative DAYNAMEs.
* (qa/752, qa/752.out): Likewise
|