Red Hat Bugzilla – Bug 443691
Valgrind errors in store tests.
Last modified: 2012-12-07 12:46:01 EST
Description of problem:
After enabling valgrind for all the store C++ test executables, there are
valgrind errors that need to be resolved.
All tests pass without valgrind (make check VALGRIND=)
Steps to Reproduce:
On a machine with valgrind installed:
> make check
The jrnl/ tests are extremely slow (40 minutes) under valgrind. They are
probably re-testing the same code paths reundantly in loops, and can probably be
re-factored to run in a reasonable amount of time under valgrind without losing
any code coverage.
In particular _ut_enq_map.stress seems like an inappropriate build test, perhaps
should be separated into a stress or performance test suite.
Created attachment 303384 [details]
Results of make check with valgrind. jrnl tests prevented by jtt failure
Created attachment 303388 [details]
make check for jrnl subdir with valgrind errors.
I reduce the loop counter in the stress tests to 10 iterations so the tests run
in a reasonable time now. Looking at the logs there are actually not many
valgrind errors, it should be possible to clean them up in less than a day.
Most errors were in the tests themselves (being over-simple in their approach)
but one corner case in the journal itself was uncovered. All fixed.
qpidc-0.2.656926-1.el5, qpidd-0.2.656926-1.el5, and rhm-0.2.2058-1.el5 have been pushed to the staging repo for testing