rhodo opened a new pull request, #17457:
URL: https://github.com/apache/pinot/pull/17457
## Summary
Introduce two global broker meters to provide a 1:1, SLA-friendly view of
query failures:
- querySystemError
- queryUserError
These are emitted once per query response that contains any exception,
alongside the existing per-error-code meters.
## Motivation
Per-error-code meters can double-count when a single query surfaces multiple
error codes. For SLA/reporting, a single per-query failure signal is desirable.
This change adds a clean per-query classification while preserving existing
detailed error metrics.
## Changes
- Add QUERY_SYSTEM_ERROR and QUERY_USER_ERROR in `BrokerMeter`
- Update `BrokerResponse.emitBrokerResponseMetrics` to:
- Continue emitting existing per-error-code meters for each exception.
- Emit per query querySystemError or queryUserError
- Centralize system error classification as a static `SYSTEM_ERROR_CODES`
set on `BrokerResponse`.
## System error classification
A query is counted as system error if any exception has one of:
- SQL_RUNTIME
- INTERNAL
- QUERY_SCHEDULING_TIMEOUT
- EXECUTION_TIMEOUT
- BROKER_TIMEOUT
- SERVER_SEGMENT_MISSING
- BROKER_SEGMENT_UNAVAILABLE
- SERVER_NOT_RESPONDING
- BROKER_REQUEST_SEND
- MERGE_RESPONSE
- QUERY_CANCELLATION
- SERVER_SHUTTING_DOWN
- QUERY_PLANNING
All other error codes are counted as user errors.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
