Open atolopko-czi opened 3 years ago
As of PR #2431.
Looks like the objective of the test, properly truncated values, is working properly, where it's going wrong is that the number of cells attached to the truncated value is varying from the expected screenshot. It might just be a matter of updating the snapshots.
Since smoke-test-annotations appears to pass reliably for the push-test
workflow, I tried running just the smoke-test-annotations
from within the compatibility-test
workflow. But alas it still fails for some steps. See https://github.com/chanzuckerberg/cellxgene/runs/3655021242?check_suite_focus=true.
I agree with Seve's assessment on this:
We might want to invest a good amount of time into this. The good thing is we realized this isn't necessarily a compatibility test issue, but an at large smoke test issue.
I don't think this is blocking 0.19 since this isn't an underlying issue of the FE, but of the tests themselves which aren't attached to a release version.
Recording Seve's input that the failing tests appear to be related lasso'ing, and the resultant selection of cells.
Retries reduce the number of failures. Adding retries allowed some smoke-test-annotations
steps to pass, after initial failure. Not a proper solution, but just confirming that the failures are in fact intermittent and non-deterministic.
Also fails on push-tests.
The cellxgene desktop GHA compatibility matrix tests have intermittently failing jobs, all failing at the
smoke-test-annotations
step. E.g. see https://github.com/chanzuckerberg/cellxgene/actions/runs/1243692946. May be timing-related.