fix orc test

[pxLi]

Signed-off-by: Peixin Li peixinl@nvidia.com

I cannot reproduce on ubuntu env. But I can easily reproduce this on integration tests w/ our integration centos image:

1. test_read_round_trip-fail
2. test_simple_partitioned_read-fail

If I increase the time start a bit larger than the epoch (1970.1.1, not the same as since that would bring the weighted special cases) , then the test works well. Not sure what's the root cause here.

This PR does not need to be merged if anyone else has better context about this, related to #140

[pxLi]

build

[jlowe]

@nvdbaranec has been trying to reproduce this in an interactive environment without success. @pxLi if you can reproduce this easily, it would be great if you could work with Dave to help setup a repro case for him to debug. I don't want to sweep this bug under the rug, which is what this change wants to do.

[pxLi]

@nvdbaranec has been trying to reproduce this in an interactive environment without success. @pxLi if you can reproduce this easily, it would be great if you could work with Dave to help setup a repro case for him to debug. I don't want to sweep this bug under the rug, which is what this change wants to do.

Yeah, I provided him the env locally to reproduce yesterday. this PR is just a workaround which was intended to be taken as a real bug fix.

[jlowe]

Based on @nvdbaranec's investigations, this appears to be an issue with the GMT timezone in the CentOS environment. As such I think the approach in #1337 is more appropriate since we don't need to unnecessarily restrict the range of timestamps being tested.

[pxLi]

sure, let me close this one