High CPU usage when reading parquet with wide schema about trino HOT 13 OPEN

Sure I will check it

from trino.

cc @raunaqmorarka

from trino.

@davseitsev could you try benchmarking your case with #22451 ?
There could be further improvements made, but I wanted to see if those changes are sufficient for your scenario.

from trino.

Now it's a little bit better.

In comparison to previous results I would say it's about 1.35x times better.

The flame graph looks similar:

from trino.

That still looks strange because I removed the usage of ColumnPath#get from getColumnChunkMetaData, but it's still somehow showing in your flamegraph.

from trino.

I will check it

from trino.

Ok, it's my bad, I didn't deploy the change properly.

from trino.

Okay, so we have improved significantly with the new changes but this needs further improvement still. Would you be able to share a sample file, table definition and test query with me so that I can assess potential optimizations more easily ?

from trino.