qa/787 will fail for everyone else for the time being ... I have libpcp changes
that reduce the number of log I/Os to 1/3 or 1/4 of their previous values in
the bad cases. The qa test shows these numbers. But unfortunately the libpcp
code is not yet ready for checkin.
This commit is to (a) reserve qa/787, and (b) show Frank that I have not
forgotten about this problem ... 8^)>
Expect the real libpcp fix in late June when I'll be back.
Changes committed to git://oss.sgi.com/kenj/pcp.git dev
qa/787 | 82 +++
qa/787.out | 1283 ++++++++++++++++++++++++++++++++++++++++++++++++++++
qa/group | 1
qa/src/GNUlocaldefs | 3
qa/src/interp1.c | 37 -
qa/src/slow.0 |binary
qa/src/slow.index |binary
qa/src/slow.meta |binary
8 files changed, 1389 insertions(+), 17 deletions(-)
commit 068ab84d26645a4895220edb70b8cdca0320b734
Author: Ken McDonell <kenj@xxxxxxxxxxxxxxxx>
Date: Sun May 11 19:58:31 2014 +1000
qa/787 (new) - exercise slow archive reading
Uses the reduced version of Frank's archive that used to show
really excessive log I/O for some metrics in the presence of
<mark> records.
commit 76d90572c199f472318bd238e5d60711fd45e62d
Author: Ken McDonell <kenj@xxxxxxxxxxxxxxxx>
Date: Sun May 11 06:55:05 2014 +1000
qa/slow archive
This is a reduced version of Frank's archive
(super-SLOW-merged-archive-20140124.225919) that demonstrates
really excessive log I/O for some metrics in the presence of
<mark> records.
See http://oss.sgi.com/bugzilla/show_bug.cgi?id=1044 for details.
|