This is kind of a followup of
As the name imply, the out-of-order scrub test (ScrubTest.testScrubOutOfOrder) test scrubs detects a sstable with out-of-order partitions, which imply we have a generated sstable with out of order partitions. That latter part being not trivial since we have a sanity check in BigTableWriter that we never write out-of-order partitions. And hence, generating such a sstable used to be manual: you had to comment the check in BigTableWriter and uncomment some code in ScrubTest to generate the sstable. That did imply you had to manually regenerate the sstable each time the sstable format was changed.
I had do to such manipulation in
CASSANDRA-9705 (as there was some change to the sstable format) and unfortunately I screwed up and included the uncommenting of the sanity check in BigTableWriter in the commit.
In the meantime, came
CASSANDRA-9880, which aimed to make the manual process of generating out-of-order sstable automatic. Unfortunately, that ticket was done with the BigTableWriter sanity check commented, which did make things a lot easier for the test, but was kind of wrong.
So anyway, commenting the check in BigTableWriter was an oversight and I uncommented it on the cassandra-3.0/trunk branches. The problem is that ScrubTest.testScrubOutOfOrder now fails due to that check. We need to find a way to make that test work without disabling our sanity check on writes. Not sure what's the best solution though to be honest. Maybe we'll need a flag in BigTableWriter to disable that check just for the test?