Lukasz Stec comments

Results 28 comments of


                                            Lukasz Stec

Use "union" in partial aggregation output

> What about `dec serde opt` + `push_partial_aggregation_through_join`? see row 1 in the summary

Use "union" in partial aggregation output

I ran another orc part sf1k for this rebased on a master with https://github.com/trinodb/trino/pull/13573. There are noticeable tpch gains (6% CPU). One thing that is surprising is that the baseline...

Row wise group by on fixed width types

> Your benchmark PDF is missing results for partitioned/unpartitioned data (for such a big change) and peak memory metrics partitioned is there. I will run not partitioned

Row wise group by on fixed width types

I updated partitioned benchmark with peek memory statistics

Row wise group by on fixed width types

unpartitioned benchmark results added

Row wise group by on fixed width types

> @lukasz-stec how could that peak memory didn't go up (or even dropped). Are we correctly accounting mem in this PR? @sopel39 I suspect that peak memory is not "caused"...

Row wise group by on fixed width types

I ran a benchmark for this (**tpch/tpcds orc part sf1K**) on top of the latest master that includes adaptive partial aggregation. There is a 5% gain for tpch overall so...

Row wise group by on fixed width types

> Rather than generating source code, we can use MethodHandle composition, see https://github.com/trinodb/trino/pull/14178 There are pros and cons to both approaches. it's way easier to read, understand and profile generated...

Row wise group by on fixed width types

Let's close it. Project [hummingbird](https://github.com/trinodb/trino/issues/14237) is going in this direction anyway.

Flaky Parquet tests by ParquetMemoryManagerRuntimeException in Hive connector

https://github.com/trinodb/trino/actions/runs/3228435010/jobs/5285315134