- changed status to open
Poor failure behavior for tests with RANKS=1
If one runs make dev-check-debug RANKS=1 NETWORKS=smp
, the results include the following three failures and their -par-
equivalents:
Running test-nebr_exchange-seq-debug-smp
*** FATAL ERROR (proc 0):
rerun the program with GASNET_FREEZE_ON_ERROR=1 in the environment.
*** Caught a fatal signal (proc 0): SIGABRT(6)
FAILED (exitcode=134)
Running test-put_flood-seq-debug-smp
*** FATAL ERROR (proc 0):
rerun the program with GASNET_FREEZE_ON_ERROR=1 in the environment.
*** Caught a fatal signal (proc 0): SIGABRT(6)
FAILED (exitcode=134)
Running test-copy-par-debug-smp
*** FATAL ERROR (proc 0):
rerun the program with GASNET_FREEZE_ON_ERROR=1 in the environment.
*** Caught a fatal signal (proc 0): SIGABRT(6)
FAILED (exitcode=134)
I have confirmed that in all three cases failures are due to checks such as the following from test/copy.cpp
:
UPCXX_ASSERT_ALWAYS(upcxx::rank_n() >= 2, "Set ranks>=2 please.");`
The "polite" way for such tests to fail is illustrated in test/rput-cover.cpp
:
if (ranks > 1 && ranks % 2) {
print_test_skipped("test requires unary or even rank count");
upcxx::finalize();
return 0;
}
There may be similar "ugly" failure modes for codes in the extras (or guide?). While those do run in nightly regression testing, I consider them out-of-scope for this issue in which I want to focus just on the tests and example to appear in the release.
Comments (5)
-
reporter -
- changed milestone to 2022.3.0 release
Mass roll-over of open issues to next release milestone
-
- changed milestone to 2022.9.0 release
Mass roll-over of open issues to next release milestone
-
- changed status to resolved
Fix issue
#501: Poor failure behavior for tests with RANKS=1→ <<cset 4614fe7126f8>>
-
Merge pull request #425 into develop
- issue501-single-rank-tests:
Update ChangeLog
ChangeLog: Create new pending section
test/view: Fix a subtle memory leak seen in single-rank runs
test/view: Add a sched_yield
Fix issue
#501: Poor failure behavior for tests with RANKS=1 Fix issue#539: PG: dmap-quiescence-test hangs with one process
→ <<cset 3378a7878e78>>
- issue501-single-rank-tests:
Update ChangeLog
ChangeLog: Create new pending section
test/view: Fix a subtle memory leak seen in single-rank runs
test/view: Add a sched_yield
Fix issue
- Log in to comment
I anticipate addressing this in time for the "code freeze" (Sep 22) and not for the "feature freeze" (Sep 15). However, I am entering this issue now for tracking/planning purposes.