Description
From slack:
dralves @hahao we discussed offline. for the record we should start with making sure we have a test that is testing RYR and RWY, the register test might or might not test both. specifically RYR requires that a clients reads don’t go back in time, which might or might not be covered depending on what gets written to the register. then we’d need a way to split ledgers between clients, i.e. we don’t want to verify a global history as in the linearizable case, but instead we want to have analyze the history per-client (even though the checking is largely the same). finally we likely want to make sure that the test is good (that is tests something and fails when constraints are not met) and maybe make it a bit more adversarial to make sure the circumstances under which the constraints are broken occur often
Attachments
Issue Links
- relates to
-
KUDU-1704 Add a new read mode to perform bounded staleness snapshot reads
- Resolved