Hi All,
As per this issue <https://issues.apache.org/jira/browse/QUARKS-219>, the
keyedTimeBatchWindowTest has been failing intermittently.
In general, the goal of a time-based window test is to assert one of two
things:
A) Tuples are aggregated on time
B) Tuples are routed to the correct window partition
Tests for B) can be written deterministically, but for A) it's harder. What
does "on time" mean? Currently a tolerance is set, and the intermittent
failures are due to the test exceeding the tolerance, but even if the
tolerance is increased the tests is still not robust in a strict sense.
Given this, I'm wondering if anyone has an idea for how to make the
time-based windowing tests more reliable.
-Will
