The other big missing bit is that the working variables can't be complex
data.

That means that I can't write:

- my own form of count unique

- approximate aggregates like hyper-log-log, tdigest

- anything that constructs complex output like list_aggregate

This is just as bad as the lack of two-phased aggregate.




On Wed, Sep 23, 2015 at 12:00 PM, Jacques Nadeau <[email protected]> wrote:

> Merging is the main missing thing. Drill supports building custom aggregate
> functions. However, those are currently run in a single thread per
> grouping. Generally, it is much better to do a two-phased aggregate for
> custom functions, however the interface doesn't yet support that
> functionality.
>
> --
> Jacques Nadeau
> CTO and Co-Founder, Dremio
>
> On Wed, Sep 23, 2015 at 10:36 AM, Neeraja Rentachintala <
> [email protected]> wrote:
>
> > https://drill.apache.org/docs/developing-an-aggregate-function/
> > See the customer aggregate functions are marked as alpha and experimental
> > usage only.
> > What features or aspects are missing to make this a 'ready to deploy in
> > production' capability.
> > Appreciate response.
> >
> > thanks
> > -Neeraja
> >
>

Reply via email to