Changes committed to git://git.pcp.io/kenj/pcp master
Ken McDonell (6):
qa/087: improve stability
qa/210: improve stability
qa/157: improve stability
qa/087: more tweaking
qa/valgrind-suppress-3.5.0: another false match
configure: restore default behaviour for compiler optimization
configure | 6 +++---
configure.ac | 6 +++---
qa/087 | 12 ++++++------
qa/087.out | 36 ++++++++++++++++++------------------
qa/157 | 4 ++++
qa/157.out | 1 -
qa/210 | 5 ++---
qa/valgrind-suppress-3.5.0 | 23 +++++++++++++++++++++++
8 files changed, 59 insertions(+), 34 deletions(-)
Details ...
commit e892dce201d6c95aa053a3af04feb5e157650f3a
Author: Ken McDonell <kenj@xxxxxxxxxxxxxxxx>
Date: Mon Oct 26 21:26:58 2015 +1100
configure: restore default behaviour for compiler optimization
Fix a couple of small logic errors ... oops.
commit d0cf5feaf9e44bc401c564d0c0d92389eb49488f
Author: Ken McDonell <kenj@xxxxxxxxxxxxxxxx>
Date: Mon Oct 26 20:02:22 2015 +1100
qa/valgrind-suppress-3.5.0: another false match
On
vm25 3.10.8 x86_64 CentOS 5.11 (Final)
qa/957 is failing with a false valgrind match ... suppress it.
commit aeb19b9aee5bacc269cfa90ef66ec428cfe3566b
Author: Ken McDonell <kenj@xxxxxxxxxxxxxxxx>
Date: Mon Oct 26 19:50:35 2015 +1100
qa/087: more tweaking
After another round of deep analysis (sigh) seems like 30-50 log reads
for all cases is needed to cover the statistical variability in the
(dynamically created) archives this test is using.
commit bff83547f6bfe354e9f9a1646817aa51469d4bc7
Author: Ken McDonell <kenj@xxxxxxxxxxxxxxxx>
Date: Mon Oct 26 17:13:40 2015 +1100
qa/157: improve stability
Wait until pmlogger has really started (and first record has been
written), then the number of size changes is deterministic.
commit 6c41f8bc5f11dfb72eaa73c0a4cc9dddaa1eb31c
Author: Ken McDonell <kenj@xxxxxxxxxxxxxxxx>
Date: Mon Oct 26 16:26:57 2015 +1100
qa/210: improve stability
After the "log early" changes for pmlogger, this QA test has become
unstable because 100msec logging and a duration of -T0.3 is all a
bit racy (sometimes get an extra record in the archive).
Change to 100msec logging and -s 3 ... and all's well.
commit 3886bcf12fb86ca971d3df5a7f579b66cd518438
Author: Ken McDonell <kenj@xxxxxxxxxxxxxxxx>
Date: Mon Oct 26 15:48:32 2015 +1100
qa/087: improve stability
A detailed analysis of failure of the QA test for the -t 10 -s 8 -a
src/interp case shows that sub-millisecond differences in the archive
timestamps can mean more reading to find the "next" record to bound
the values for interpolate mode ... 44 log reads is not incorrect
in these circumstances, so adjust the acceptable limits from (30-40)
reads to (30-45) reads.
|