mirror of
https://github.com/scylladb/scylladb.git
synced 2026-06-06 06:53:12 +00:00
a9dbd89478
We don't expect the cluster to be functioning at all after a failed test. The whole cluster might have crashed, for example. In these situations the framework would report multiple errors (one for the actual failure, another for a failed post-condition check because the cluster was down) which would only obscure the report and make debugging harder. It's also not safe in general to reuse the cluster in another test - if the test previous failed, we should not assume that it's in a valid state. Therefore, mark the cluster as dirty after a failed test. This will let us recycle the cluster based on the dirty flag and it will disable post-condition check after a failed test (which is only done on non-dirty clusters). To implement this in topology tests, we use the `pytest_runtest_makereport` hook which executes after a test finishes but before fixtures finish. There we store a test-failed flag in a stash provided by pytest, then access the flag in the `manager` fixture.