feat: bucketize latency attribute

[deniseli]

This PR bucketizes the latency attribute (NOT the latency metric! That would be very bad!)

For better filtering capabilities in datadog, we need to bucket our high-cardinality numeric attributes. This uses log base8 because:

1. That seems to give the most useful results for current production numbers we're seeing in DD - good functional granularity for analysis purposes without being so granular that it's annoying to sort through the buckets. Current values range from 62 to 20279, concentrated in the 100-1000 range.
2. Base 8 bucket labels are all powers of 2, so it's relatively easy for our brains to skim

Sample output:

-> ftl.async_call.time_since_scheduled_at_ms.bucket: Str([8,64))

Issue: #2194

[alecthomas]

Can we perform this bucketing in DD instead?

[deniseli]

Can we perform this bucketing in DD instead?

I couldn't find a way to do it. Oddly, DD has decent support for advanced string filtering and bools, but not even basics for numeric-type attrs. At least, I couldn't find any docs on it, and nothing intuitive I tried in the UI worked either.

Worse, DD tries to show every single value of the attr in its filter dropdown, which fills up the dropdown to its max size and pushes all the other attrs out of view, so then you can't explore the values of any other attributes.

One option would be to reduce base to 4 or something, because we can add groups of allowed values for filtering... btw if anyone can find a good way to filter by < and > on int attrs in DD, I'd love to undo this