Support grouping multiple rows of first match in an interval

[metalmatze]

Right now Parca queries FrostDB for all the timestamps and on the server side chooses to ignore metric samples when a sample has been found in the step's bucket.

https://github.com/parca-dev/parca/blob/ed90dbeb684186e9cdb295bc0f62c723ed3c5a9f/pkg/parcacol/querier.go#L571-L579

This isn't ideal since the data is still queried...

FrostDB should add support to find all rows for a bucket of interval and only then start aggregating.

The row data would look like this:

timestamp	stacktrace	value
1	stack1	3
1	stack2	5
1	stack3	8
2	stack1	2
2	stack2	3

What we want. We want to only return the sum(value) for the first timestamp we find in the bucket that ranges from 0-9.
So what we want as a result is:

timestamp	sum(value)
1	16

What we currently get is however the very first row that falls within that bucket (1, stack1, 3) and then the sum of it's value, which is basically a noop:

timestamp	sum(value)
1	3

[metalmatze]

This issue is especially bad for Parca itself. On our cloud with the aggregate_view table, it shouldn't be such a bad problem for now.

[albertlockett]

@metalmatze As far as I an tell the Duration aggregation does what we want. I added a test to demonstrate.
https://github.com/polarsignals/frostdb/pull/503/files

That said, I think this is different than what we do here in parca, where we take the first result in the interval.
parca-dev/parca#2598

I'm wondering if maybe what we want is a First aggregation instead like this (or something like it)?
#504

[github-actions]

This issue is stale because it has been open 30 days with no activity. Remove stale label or comment or this will be closed in 5 days.

[github-actions]

This issue is stale because it has been open 30 days with no activity. Remove stale label or comment or this will be closed in 5 days.