A grab bag here. Some QA changes, but perhaps more importantly libpcp and pmie
bug fixes.
Changes committed to git://oss.sgi.com/kenj/pcp.git dev
qa/.gitignore | 2
qa/062 | 5
qa/062.out | 1156 ------------------
qa/083 | 17
qa/115 | 12
qa/514 | 10
qa/514.out | 8
qa/514.out.1 | 2430 +++++++++++++++++++++++++++++++++++++++
qa/514.out.2 | 2430 +++++++++++++++++++++++++++++++++++++++
qa/520 | 10
qa/520.out | 6
qa/520.out.1 | 594 +++++++++
qa/520.out.2 | 594 +++++++++
qa/524 | 102 +
qa/524.out | 64 +
qa/525 | 100 +
qa/525.out | 46
qa/check-group | 8
qa/common | 87 +
qa/common.check | 10
qa/group | 48
qa/new | 8
qa/src/chkacc1.c | 166 +-
qa/src/chkacc2.c | 181 +-
qa/src/chkacc3.c | 287 ++--
qa/src/fixhosts | 47
src/libpcp/src/err.c | 1
src/libpcp/src/lock.c | 2
src/libpcp/src/util.c | 23
src/libpcp_fault/src/GNUmakefile | 3
src/pmie/src/dstruct.c | 79 -
src/pmie/src/dstruct.h | 7
src/pmie/src/lexicon.c | 5
src/pmie/src/pragmatics.c | 43
src/pmlogger/pmlogger.c | 4
35 files changed, 7108 insertions(+), 1487 deletions(-)
commit 177f53196591d2a2fcda0f91d3cfdaf26acef6c8
Author: Ken McDonell <kenj@xxxxxxxxxxxxxxxx>
Date: Mon Apr 8 19:13:15 2013 +1000
qa/083 - fix systemic failure
This test has been failing ever since we switched from running
pmcd and pmlogger as "root".
But the root cause was obscured by the unrelated __pmOpenLog
bug that has only recently been fixed.
commit dcead9b696ba07148a15ee41d4b9f33c5094d38d
Author: Ken McDonell <kenj@xxxxxxxxxxxxxxxx>
Date: Mon Apr 8 18:05:28 2013 +1000
pmie - improved propagation of metadata units
We were not always propagating the units part of the metric
metadata up the expression tree. This is not an issue for pmie
per se, but becomes a problem when pmie is used in "secret agent"
mode underneath the summary PMDA.
Thanks to Jun Wang for raising this issue.
commit dfdc8c3f70e05e0d8d3bfd7977decb58d08d991e
Author: Ken McDonell <kenj@xxxxxxxxxxxxxxxx>
Date: Mon Apr 8 18:01:44 2013 +1000
qa/525 - new - exercise summary PMDA
Based on a report by Jun Wang, junwang123@xxxxxxxxx
commit 51c5dff227cdda71deb99847387d7f109f1764c1
Author: Ken McDonell <kenj@xxxxxxxxxxxxxxxx>
Date: Mon Apr 8 17:01:38 2013 +1000
qa/520 - new conditional output after pmie changes
pmie now does a better job of propagating metadata units up
the expression tree, so some of the diagnostic output is
slightly different.
commit 77be645c583acec4d5feb4ebf7c4cee323c5b134
Author: Ken McDonell <kenj@xxxxxxxxxxxxxxxx>
Date: Mon Apr 8 16:58:41 2013 +1000
qa/514 - new conditional output after pmie changes
pmie now does a better job of propagating metadata units up
the expression tree, so some of the diagnostic output is
slightly different.
commit 12ca09619cde7c6bfbf175acfa5880999501e98b
Author: Ken McDonell <kenj@xxxxxxxxxxxxxxxx>
Date: Mon Apr 8 16:54:27 2013 +1000
qa/115 - fix bizarre failure
If the remote hostname (e.g. bozo) is a proper prefix of the
local hostname (e.g. bozo-laptop) the test fails because the
filter used to rewrite the hostnames incorrectly!
commit a0f538fc84469a5e751a1c82da50ae923e7dc0fe
Author: Ken McDonell <kenj@xxxxxxxxxxxxxxxx>
Date: Sun Apr 7 13:17:52 2013 +1000
libpcp/err.c - fix small initialization logic error
In pmErrStr_r() we have a one-trip initialization to optionally
find the "Unknown error ..." message buffer from libc and strdup()
it. The code was not correctly dealing with a (possibly)
negative number after the "Unknown error " part of the message.
Found while running the error injection QA tests and we were not
seeing the fault in the strdup() here because the PM_FAULT_POINT
was not being executed as a result of the logic error.
commit 5128d46420ee0ae43f7e5a8f8fe80745c5d2faf7
Author: Ken McDonell <kenj@xxxxxxxxxxxxxxxx>
Date: Sun Apr 7 08:59:35 2013 +1000
libpcp_fault - fix makefile bit rot
libpcp has moved on ... libpcp_fault needed to catch up.
commit ed1e37f55203ac14018b08eb29d4881023ae9e9d
Author: Ken McDonell <kenj@xxxxxxxxxxxxxxxx>
Date: Sun Apr 7 08:29:22 2013 +1000
libpcp/lock.c - fix compilation failure
Turned on PM_MULTI_THREAD_DEBUG and kaboom!
commit bd571fb3b331cc5083aa78d80f49c0e4f64fd46c
Author: Ken McDonell <kenj@xxxxxxxxxxxxxxxx>
Date: Sun Apr 7 08:22:33 2013 +1000
qa/group - cleanup
Retire the purify tests.
Add some missing groups.
Delete some groups that are never going to be useful here, e.g. the
PCP gui apps.
commit 54a0e7d7c92de1bbd3ca186c7e65376b584bb75e
Author: Ken McDonell <kenj@xxxxxxxxxxxxxxxx>
Date: Sun Apr 7 08:21:01 2013 +1000
qa - added "retired" tests support
Tagging a test with :retired immediately after the test sequence
number in the group file will mean the test is not run by default,
e.g. 584:retired libpcp pmie local #698774 oss purify
Add -r option to check et al to override the default behaviour and
allow retired tests to be run.
commit e4ae210c42fa7338d4c5c60fd68ec6d67dcdbad7
Author: Ken McDonell <kenj@xxxxxxxxxxxxxxxx>
Date: Sat Apr 6 06:42:11 2013 +1100
qa/common.check - cosmetic changes
commit d7b7fe742d897deb35f82eda2fc08528ad13ad74
Author: Ken McDonell <kenj@xxxxxxxxxxxxxxxx>
Date: Fri Apr 5 21:59:09 2013 +1100
qa/new - minor tweak for trap
commit 4db0f8b64657f4bf549de28bf8657b90c2b20ea8
Author: Ken McDonell <kenj@xxxxxxxxxxxxxxxx>
Date: Fri Apr 5 21:50:26 2013 +1100
qa/062 - remove ipv6 tests
This is the test that recently has been running for a _very_ long
time (like thousands of seconds) ... it now runs in under 2 secs.
For the src/chkacc?.c apps, add -4 (ipv4), -6 (ipv6) and -D cmd line
options ... default is -4.
Populate /etc/hosts with the IP addresses and bogus host names
needed to speed up the reverse DNS.
I'll leave it to someone who understands the ipv6 code and coverage
elsewhere in the QA suite to decide if and how it is appropriate
to reintroduce some variant of the ipv6 testing that used to be
in QA/062 in a (new) QA test.
commit 6ce0829cb900f67306ec0fc191da04020b88a534
Author: Ken McDonell <kenj@xxxxxxxxxxxxxxxx>
Date: Fri Apr 5 20:39:40 2013 +1100
libpcp - rework error path for __pmOpenLog
On the error path when the new log cannot be opened for writing,
the logic was not quite correct, and oldstream was not properly
restored ... this left applications like pmlogger writing stderr
into the void.
The problem is confined to the (rare, except for QA/083 and QA/524)
situation where the freopen() fails and we're able to recover
... but then did not leave the input oldstream attached to the
recovered i/o stream.
The fix is not without some issues ... refer to the comment in
the code.
Problem is actually in the internal routine logreopen() which is
called by both __pmOpenLog() and __pmRotateLog(), although the
likely use of the latter is unlikely to expose the problem.
commit a7dc53fa9d0dd827213257524df8cc4de0bc3ab7
Author: Ken McDonell <kenj@xxxxxxxxxxxxxxxx>
Date: Fri Apr 5 20:32:54 2013 +1100
pmlogger - extra diagnostic when log cannot be created
If pmlogger.log or the filename from -l cannot be created for
writing, it would be nice to warn about it. Except for a related
error in __pmOpenLog() (see next commit) output should go to
stderr rather than the log file that cannot be created.
commit 7c49a42e91d90497a83ea3f0e78ff22ed5bd7f19
Author: Ken McDonell <kenj@xxxxxxxxxxxxxxxx>
Date: Fri Apr 5 20:20:24 2013 +1100
qa/524 - new - odd permissions and pmlogger
Observed while trying to debug the recent regression in QA/083 ...
turn out that if pmlogger is started in a directory that is not
writeable, not only is the archive not created, but all stderr
is lost due to an obscure bug in __pmOpenLog() on the error
"recovery" path.
|