Botond Dénes

v6 - complete re-write, see cover-letter. @bhalevy @raphaelsc @avikivity please re-review.

> > Indeed very good point, the coordinator can crash, releasing the erm and the replica side can keep executing. The last thing we want in this situation is replicas...

BTW I just realized that we have a window between `cleanup_tablet()` and `update_effective_replication_map()` where inactive reads can be produced and these will survive and can cause data resurrection.

> @denesb what do you think? is it safe to start using Python 3.9 ? I have no idea :laughing:. I may have written this script, but I just used...

I cannot find any logs. I wonder if it is python in fact segfaulting here, and this is a duplicate of https://github.com/scylladb/scylladb/issues/17564. As of https://github.com/scylladb/scylladb/commit/5c2e5890a610fc963b30aae8546e8af21c73dd6f C++ unit tests are also...

Yes, coredump uploaded to gs://upload.scylladb.com/24551/encryption_at_r.114046.1750160309.gz

Apparently we didn't save the executable for this `encryption_at_rest_test`, so the coredump is useless. We only save `combined_tests` binary, but this test is not part of it.

New in v2: * Added `--cache-hit-rate-read-balancing=0`, I think this was the reason for the occasional flakiness to begin with. * Also use `LOCAL_ONE` instead of `ONE` (shouldn't make a difference).

> > currently the tests use scylla-sstable write + upload but this won't work with tablets since upload with tablets implies --load-and-stream which means data is streamed to all replicas...

@scylladb/scylla-maint ping this PR improves a flaky test and makes it work with tablets.