fix(slo): Correctly report TOO_MANY_BYTES from allocation policy as rate-limited

[volokluev]

Human summary

Our SnQL API supposedly has a very high error rate:

https://app.datadoghq.com/s/FH6-Y3/7s2-qbw-ap9

but these errors are not being captured in sentry. I dug into the querylog and noticed that almost all of them scanned too many bytes. This is a decision the BytesScannedRejectingPolicy can make to allow small queries through. It's surfaced as a rate limit to the caller but the metrics call it an error. This PR fixes it

Robot Summary

Fixes SLO reporting for TOO_MANY_BYTES errors that occur when queries exceed the max_bytes_to_read limit set by the BytesScannedRejectingPolicy allocation policy.

Problem

When the BytesScannedRejectingPolicy determines a customer is over their scan limit, it can set max_bytes_to_read on the ClickHouse query settings to allow smaller queries through instead of rejecting them outright. When a query exceeds this limit, ClickHouse raises a TOO_MANY_BYTES error (code 307).

Previously, this was being reported as a generic ERROR in SLO metrics, which counts against the SLO. However, this is actually rate limiting behavior and should be reported as RATE_LIMITED, which does not count against the SLO.

Changes

1. Track allocation policy context (snuba/web/db_query.py:860)

   • Added max_bytes_to_read_set_by_policy flag to stats when the allocation policy sets max_bytes_to_read

2. Update error handling (snuba/web/db_query.py:479-483)

   • Modified TOO_MANY_BYTES handling to only convert to RateLimitExceeded when the flag is set
   • Preserves generic error classification for TOO_MANY_BYTES from other sources

3. Context-aware status determination (snuba/querylog/query_metadata.py:128-149)

   • Updated get_request_status() to accept optional context parameter
   • Checks for max_bytes_to_read_set_by_policy flag when handling TOO_MANY_BYTES errors
   • Returns RequestStatus.RATE_LIMITED only when the flag is present

4. Comprehensive test coverage

   • Added unit tests for get_request_status() context handling
   • Added integration tests for both allocation policy and non-policy scenarios
   • Verifies correct SLO classification in both cases

Result

• ✅ TOO_MANY_BYTES from allocation policy → RATE_LIMITED (SLO "for")
• ✅ TOO_MANY_BYTES from other sources → ERROR (SLO "against")
• ✅ Allocation policy rate limiting no longer incorrectly impacts SLO metrics

Test Plan

# Run unit tests
pytest tests/querylog/test_query_metadata.py -v

# Run API status tests
pytest tests/test_api_status.py::TestApiCodes::test_too_many_bytes_from_allocation_policy -v
pytest tests/test_api_status.py::TestApiCodes::test_too_many_bytes_without_allocation_policy -v

All tests pass ✅