I see. We are definitely not hitting the max row count:

[image: image.png]

As an aside from this, what is the right way to achieve perfect rollups?

On Fri, Aug 9, 2019 at 10:54 AM Gian Merlino <g...@apache.org> wrote:

> Fwiw, even if the rows are in order, they can end up not-rolled-up if they
> come into two separate Kafka tasks (or even the same task across a segment
> boundary based on max row count or intermediate handoff).
>
> On Fri, Aug 9, 2019 at 10:50 AM Prashant Deva <prashant.d...@gmail.com>
> wrote:
>
> > In this case there is only a single Kafka partition and rows are coming
> in
> > the order of timestamp and at the timestamp itself (no out of order rows
> > are being ingested).
> >
> > On Fri, Aug 9, 2019 at 10:21 AM Gian Merlino <g...@apache.org> wrote:
> >
> > > Hey Prashant,
> > >
> > > Thanks for the report, we'll look into this.
> > >
> > > Btw- I would generally not expect datasources that use Kafka ingestion
> to
> > > be _fully_ aggregated, since it doesn't guarantee perfect rollup (it'll
> > > roll up rows that come in to the same Kafka ingestion task, which is
> > based
> > > on when they come in and what partitions they come in from). But
> > > nevertheless we'll check out this issue to see if it looks like
> something
> > > is wrong.
> > >
> > > On Fri, Aug 9, 2019 at 9:57 AM Prashant Deva <prashant.d...@gmail.com>
> > > wrote:
> > >
> > > > by ' Data sources that dont aggregate' i mean data sources that have
> > same
> > > > query granularity as the data that is generated.
> > > >
> > > > Prashant
> > > >
> > > >
> > > > On Fri, Aug 9, 2019 at 9:55 AM Prashant Deva <
> prashant.d...@gmail.com>
> > > > wrote:
> > > >
> > > > > Only posting this because I think this is a major issue.
> > > > > See bug #8276 <
> https://github.com/apache/incubator-druid/issues/8276
> > >
> > > > for
> > > > > details.
> > > > > Seems to be happening on every data source that is aggregating
> during
> > > > > ingestion.
> > > > >
> > > > > Data sources that dont aggregate dont have this issue.
> > > > >
> > > > > Prashant
> > > > >
> > > >
> > >
> > --
> > Prashant
> >
>

Reply via email to