Eric Fu

After some extra thinking, I would like to propose to use a single error_table rather than one error_table per operator. As mentioned in the RFC, our targets include > 1....

> > Instead, in the error_table-per-operator fashion, we might need to provide an extra 'view' for that purpose. Note that this is not an actual 'view' because when a new...

> How do we synchronize access to the state table? Or we use uuid here to synchronize, instead of row_id? If yes that makes sense 👍 Yes, uuid should be...

**Meeting Notes** - We will do it and start by supporting memory estimation for batch executors, as stated in the "Implementation" section. - We will work on the "Without Overselling"...

> Can we close the RFC? I will rewrite the "Measure: How to measure memory usage?" part

As pointed out by @st1page, the plan node proto (actually fragment node) need to be kept compatible, otherwise the existing streaming jobs will fail.

After the dicussion, I'm a little wavering now. 1. It introduces some extra complexity, not only in programing but also in analyzing a bug afterwards. 2. Sometimes it's hard to...