bench: aggregations over decimals

[sploiselle]

A very simple test of an aggregation over a single decimal column; this will help us understand how much the rust-dec-backed decimal implementation degrades performance and increases memory usage.

This supersedes #6658 and #6643

[cirego]

Looks great to me!

We need to figure out why the smoke test failed in CI.

[cirego]

What's creating the data file and/or directory?

[cirego]

You might consider using INSERT instead of UPSERT, as the upsert operator is quite a bit more expensive. Using insert should make the performance delta between implementations more apparent.

[cirego]

It looks like this failed with the following error message:

CREATE SOURCE
psql:/data/aggregation_views.sql:21: ERROR:  column name "OuterRecord" is ambiguous
psql:/data/aggregation_views.sql:23: ERROR:  unknown catalog item 'aggregationtest'
$ docker wait 15c81efce39dd2dcf9575085d9e51cb288a90ebdba2f625bad69ffba79c26b47

[cirego]

I'm able to reproduce this locally:

[chris-work:~/materialize/materialize:(dec-agg-test[|])] docker logs aggregations_create-views_run_f4dc208bfa72 
CREATE SOURCE
psql:/data/aggregation_views.sql:21: ERROR:  column name "OuterRecord" is ambiguous
psql:/data/aggregation_views.sql:23: ERROR:  unknown catalog item 'aggregationtest'

[cirego]

I also see kgen exiting with the following errors:

Spawning 4 generator processes, writing 25000 messages each
thread 'main' panicked at 'max value of 1000000000 exceeds value expressable with precision 15', src/kafka-util/src/bin/kgen.rs:294:13
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace
thread 'main' panicked at 'max value of 1000000000 exceeds value expressable with precision 15', src/kafka-util/src/bin/kgen.rs:294:13
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace
thread 'main' panicked at 'max value of 1000000000 exceeds value expressable with precision 15', src/kafka-util/src/bin/kgen.rs:294:13
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace
thread 'main' panicked at 'max value of 1000000000 exceeds value expressable with precision 15', src/kafka-util/src/bin/kgen.rs:294:13
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace
0/4 processes finished: pid=7 returncode=101
1/4 processes finished: pid=8 returncode=101
2/4 processes finished: pid=9 returncode=101
3/4 processes finished: pid=10 returncode=101

[cirego]

I think the "OuterRecord" error may be a red-herring, when we really should have exited with error when kgen failed.

Edit: forget what I said about the red herring. I was contemplating that perhaps the schema didn't exist in the schema registry yet, but I can see that it was created:

[chris-work:~/materialize/materialize/test/bench/aggregations:(dec-agg-test[|])] curl -s http://localhost:8081/subjects/aggregationtest-value/versions/latest | jq .
{
  "subject": "aggregationtest-value",
  "version": 1,
  "id": 13,
  "schema": "{\"type\":\"record\",\"name\":\"aggregationtest\",\"namespace\":\"com.acme.avro\",\"fields\":[{\"name\":\"Key1Unused\",\"type\":\"long\"},{\"name\":\"OuterRecord\",\"type\":{\"type\":\"record\",\"name\":\"OuterRecord\",\"fields\":[{\"name\":\"DecValue\",\"type\":{\"type\":\"bytes\",\"logicalType\":\"decimal\",\"precision\":15,\"scale\":2}}]}}]}"
}

Once #6675 lands, we should pull those improvements into this benchmark too!

[sploiselle]

@cirego tysm for checking into this; the root of the problem was just an inverted inequality that caused everything else to topped down around it. re: UPSERT vs. append-only envelope, I want the ability to control the total size of the view to examine memory consumption of the refactor. I guess it's possible to also throttle this with the number of messages we send, so idk––going to go with UPSERT because it also feels like a slightly more realistic.