vol-async icon indicating copy to clipboard operation
vol-async copied to clipboard

Failing tests with HDF5 API tests for VOLS.

Open brtnfld opened this issue 1 year ago • 5 comments

For the serial tests (test/API in HDF5), only h5_api_test_attribute fails with:


1: Testing shared datatype for attributes                                *FAILED*
1:     reference count of the named datatype is wrong: 1

For the parallel tests (testpar/API), only h5_api_test_parallel_async fails with:

9: **********************************************
9: *                                            *
9: *      API Parallel Async Tests              *
9: *                                            *
9: **********************************************
9: 
9: Testing single dataset I/O                         
9:   Testing test setup                                                  HDF5-DIAG: Error detected in HDF5 (1.15.0) MPI-process 0:
9:   #000: ../../src/H5VLcallback.c line 6321 in H5VLintrospect_get_conn_cls(): NULL obj pointer
9:     major: Invalid arguments to routine
9:     minor: Bad value
9: HDF5-DIAG: Error detected in HDF5 (1.15.0) MPI-process 0:
9:   #000: ../../src/H5VL.c line 658 in H5VLobject_is_native(): can't determine if object is a native connector object
9:     major: Virtual Object Layer
9:     minor: Can't get value
9:   #001: ../../src/H5VLint.c line 1077 in H5VL_object_is_native(): can't get VOL connector class
9:     major: Virtual Object Layer
9:     minor: Can't get value
9:   #002: ../../src/H5VLcallback.c line 6289 in H5VL_introspect_get_conn_cls(): can't query connector class
9:     major: Virtual Object Layer
9:     minor: Can't get value
9:   #003: ../../src/H5VLcallback.c line 6256 in H5VL__introspect_get_conn_cls(): can't query connector class
9:     major: Virtual Object Layer
9:     minor: Can't get value
9:   #004: ../../src/H5VLcallback.c line 6321 in H5VLintrospect_get_conn_cls(): NULL obj pointer
9:     major: Invalid arguments to routine
9:     minor: Bad value
9: *FAILED*

brtnfld avatar May 31 '24 19:05 brtnfld

@brtnfld, for the failed serial test, I cannot find the root cause of how the reference count is wrong, can we skip this test when testing async vol for now? I'm looking into the parallel test issue.

houjun avatar Jun 07 '24 21:06 houjun

We should be able to skip the serial test for now. Were you able to reproduce the failures? Thanks for looking into this.

brtnfld avatar Jun 10 '24 14:06 brtnfld

Yes, I can reproduce the error with the parallel test, the error is due to the file object not being created (because of async) when H5VL_object_is_native is called, thus leading to the NULL obj pointer error. I will work on a fix.

houjun avatar Jun 11 '24 18:06 houjun

@houjun, any progress on this? Strangely, the serial test fails as it never calls any _async APIs.

brtnfld avatar Dec 03 '24 02:12 brtnfld

@brtnfld I just created a pull request to disable the "shared datatype for attributes" test for vol-async, so the serial test will all pass. I'll get back to work on the parallel tests with the latest HDF5 this week.

houjun avatar Dec 03 '24 21:12 houjun