Gian Merlino

@ColeAtCharter I was working on something recently that reminded me of this report. I realized it's possible to see the "unexpected result example 1" if the filter is filtering on...

Oops, I missed the latest message on this issue. Certainly, if the filter value is not ambiguously in two possible segments, then only one (per time chunk) should be queried....

The deepest cause of the "Index 180 out of bounds for length 129" error is the stack trace below, which I think can only happen if the `unionMap` is corrupted...

The first one, `sum(pe1_0.payment_amount)`, is now `NULL` due to SQL compliance (this is how SQL expects `SUM` to behave). The behavior is no longer configurable, since Druid is now always...

@carltal It depends on how the SQL query is translated. The planner makes some effort to translate `coalesce` in a low-impact way. Ideally for a situation like this, it will...

I'm not familiar with how the autocomplete is coded, in particular I'm not sure how much control we have over it (vs. it coming in as part of a component...

The error messages seem to be talking about the host `druid-druid-coordinators-7889b9b98d-jcrxs` not being resolvable. I'm not a kubernetes expert so I am not sure what exactly could be causing that....

Is there something we can or should change in the bundled script to improve this case? Otherwise I suppose we should close the issue, since the cause has been figured...

It seems that the intent here is to prevent sampling from Kafka getting stuck when fewer than `numRows` exist in the stream. Instead of adding a second shorter timeout, could...

Btw, on general design. Sampling is intended to be used in a very specific situation— it's there so the web console (& any similar third-party UIs) can show people a...