feat: formula queries in EndpointTraceItemTable

[kylemumma]

This PR implements support for formulas in the TraceItemTable endpoint. It is relevant to this ticket https://github.com/orgs/getsentry/projects/284/views/1?pane=issue&itemId=85243940&issue=getsentry%7Ceap-planning%7C27
It enables queries such as sum(my_attribute) / count(my_attribute) which werent possible before

Major changes

• the column to expression conversion logic used to take place here I extracted this logic into 2 separate new functions: _get_reliability_context_columns and _column_to_expression. (i.e. the logic stayed the same but its not inside these functions)
• I then extended _column_to_expression to support formulas https://github.com/getsentry/snuba/pull/6844/files#diff-e1e06d7f875a7c2870cc11bc4301dd6ab9fba73263c76260452c0b3176f66110R180-R191
• I also had to extend the existing result conversion and order by logic to support formulas https://github.com/getsentry/snuba/pull/6844/files#diff-e1e06d7f875a7c2870cc11bc4301dd6ab9fba73263c76260452c0b3176f66110R287-R288

Testing

I wrote 3 new tests for this feature:

• one that tests a simple formula on aggregates sum(my_attribute) / count(my_attribute)
• one that is the same as above uses extrapolation, to ensure formulas work with extrapolation.
• one that tests formulas on attributes without aggregation my_attribute + my_other_attribute

design decisions

• reliabilities dont work with formulas, if you do a formula on extrapolated aggregates you will not get any reliability information back. This decision was made because of the increased complexity it would add to support. If needed we can implement support for this in the future.
• I realized while implementing this that formulas using constants are not supported such as my_attribute * 10 if we need support for this it must be implemented as a follow up. and will require further modification of our protobuf grammar.
• it only supports spans not uptime or logs

[codecov]

❌ 1 Tests Failed:

Tests completed	Failed	Passed	Skipped
2847	1	2846	11

View the top 1 failed tests by shortest run time

tests.web.rpc.v1.test_endpoint_trace_item_table.test_endpoint_trace_item_table.TestTraceItemTable::test_non_agg_formula

Stack Traces | 0.342s run time

Traceback (most recent call last):
  File ".../v1/test_endpoint_trace_item_table/test_endpoint_trace_item_table.py", line 2450, in test_non_agg_formula
    assert response.column_values == [
AssertionError: assert [attribute_name: "kyles_measurement + my_other_attribute"\nresults {\n  val_double: 13\n}\nresults {\n  val_double: 5\n}\nresults {\n  val_double: 5\n}\nresults {\n  val_double: 0\n}\nresults {\n  val_double: 0\n}\nresults {\n  val_double: 0\n}\nresults {\n  val_double: 0\n}\n] == [attribute_name: "kyles_measurement + my_other_attribute"\nresults {\n  val_double: 0\n}\nresults {\n  val_double: 0\n}\nresults {\n  val_double: 0\n}\nresults {\n  val_double: 0\n}\nresults {\n  val_double: 5\n}\nresults {\n  val_double: 5\n}\nresults {\n  val_double: 13\n}\n]
  At index 0 diff: attribute_name: "kyles_measurement + my_other_attribute"\nresults {\n  val_double: 13\n}\nresults {\n  val_double: 5\n}\nresults {\n  val_double: 5\n}\nresults {\n  val_double: 0\n}\nresults {\n  val_double: 0\n}\nresults {\n  val_double: 0\n}\nresults {\n  val_double: 0\n}\n != attribute_name: "kyles_measurement + my_other_attribute"\nresults {\n  val_double: 0\n}\nresults {\n  val_double: 0\n}\nresults {\n  val_double: 0\n}\nresults {\n  val_double: 0\n}\nresults {\n  val_double: 5\n}\nresults {\n  val_double: 5\n}\nresults {\n  val_double: 13\n}\n
  Full diff:
    [
     attribute_name: "kyles_measurement + my_other_attribute"
  + results {
  +   val_double: 13
  + }
  + results {
  +   val_double: 5
  + }
  + results {
  +   val_double: 5
  + }
    results {
      val_double: 0
    }
    results {
      val_double: 0
    }
    results {
      val_double: 0
    }
    results {
      val_double: 0
    }
  - results {
  -   val_double: 5
  - }
  - results {
  -   val_double: 5
  - }
  - results {
  -   val_double: 13
  - }
    ,
    ]

To view more test analytics, go to the Test Analytics Dashboard
📢 Thoughts on this report? Let us know!

[kylemumma]

new protobuf definition with support for formulas, see getsentry/sentry-protos#105

[kylemumma]

if people try to use formulas in Timeseries Endpoint it returns a not-implemented-error. I will implement support for formulas in timeseries endpoint as my next PR.

[volokluev]

So this implements formulas only for spans. How much work would it be to support it for all trace item types

[kylemumma]

So this implements formulas only for spans. How much work would it be to support it for all trace item types

It would be non-trivial to extend this PR to support all data types, we will leave it as span only