This is a very useful test. It does seem to test a lot of corner cases related to flush and "visible byte" reads.
I do see that this test has a lot of capability, so most of my comments are more related to utilizing that part. As for the code itself, I found the changes useful.
1. Since this test already has the capability to test position reads, can that also be enabled as a separate @test? Especially since that was something that brought out
HDFS-2021 and HDFS-1907.
2. Since it has the capability to do fc, could we also enable that as another @test?
3. Should the WR_NTIMES and WR_CHUNK_SIZE be increased to reflect the failed case of
HDFS-2021 and HDFS-1907? We can make this JIRA a dependent of those and thus possibly avoid having any test failures.