feat(api): return TOO_MANY_BYTES as clickhouse error instead of 429#7782
feat(api): return TOO_MANY_BYTES as clickhouse error instead of 429#7782
Conversation
cvxluo
force-pushed
the
cvxluo/too-many-bytes-error-changes
branch
from
e0c6985 to
6a3004f
Compare
cvxluo
changed the title
…limit Stop converting ClickHouse TOO_MANY_BYTES (code 307) errors into RateLimitExceeded. This lets consumers (e.g. sentry) distinguish "query scanned too many bytes" from snuba-level rate limiting. - Remove RateLimitExceeded conversion for TOO_MANY_BYTES in db_query.py - Add TOO_MANY_BYTES to ACCEPTABLE_CLICKHOUSE_ERROR_CODES (returns 400) - Update test to assert clickhouse error type with code 307 Co-Authored-By: Claude <noreply@anthropic.com>
cvxluo
force-pushed
the
cvxluo/too-many-bytes-error-changes
branch
from
6a3004f to
c66b632
Compare
![cursor[bot]](https://avatars.githubusercontent.com/in/1210556?s=60&v=4){kind=small}
There was a problem hiding this comment.
Cursor Bugbot has reviewed your changes and found 2 potential issues.
Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.
| "Query scanned more than the allocated amount of bytes", | ||
| quota_allowance=stats["quota_allowance"], | ||
| ) | ||
|
|
There was a problem hiding this comment.
calculated_cause variable is now redundant dead code
Low Severity
The calculated_cause variable is assigned to cause on line 472 and is never reassigned in any branch. It was previously needed because the TOO_MANY_BYTES branch would reassign it to a RateLimitExceeded instance. With that branch removed, calculated_cause is always identical to cause, making it dead code. The from calculated_cause on line 509 can simply use from cause directly.
| # queries that can never return the amount of data requested by the user are not an internal error | ||
| ErrorCodes.MEMORY_LIMIT_EXCEEDED, | ||
| # query exceeded the max_bytes_to_read limit set by an allocation policy | ||
| ErrorCodes.TOO_MANY_BYTES, |
There was a problem hiding this comment.
TOO_MANY_BYTES missing from non-retryable error codes set
Medium Severity
TOO_MANY_BYTES is added to ACCEPTABLE_CLICKHOUSE_ERROR_CODES but not to NON_RETRYABLE_CLICKHOUSE_ERROR_CODES. Previously this error was converted to RateLimitExceeded, so it never hit the non-retryable check. Now it flows through as a ClickhouseError, so in the subscription executor consumer it triggers a SubscriptionQueryException (retry), even though the error is deterministic and retrying won't help since the max_bytes_to_read limit is fixed by allocation policy.
1 participant
Stop converting ClickHouse TOO_MANY_BYTES (code 307) errors into RateLimitExceeded. This lets consumers (e.g. sentry) distinguish "query scanned too many bytes" from snuba-level rate limiting.