Juan Fernandez
013718aacb
Fixes #47303 GetQueryResultsCounts and IncrQueryResultsCounts pipelined commands across multiple query_results_count:<id> keys on a single connection. These keys have no hash tag, so in a Redis Cluster they scatter across hash slots. A pipelined connection binds to the first key's slot, so every other key returned a MOVED redirect, producing recurring error log noise on host check-ins. IncrQueryResultsCounts additionally used ConfigureDoer, whose RetryConn does not support Send, so increments failed entirely in cluster mode. Group the keys by hash slot with redis.SplitKeysBySlot and run one pipeline per slot group, mirroring the existing QueriesForHost and CleanupInactiveQueries patterns in the same file. The write path uses a plain pooled connection (not ConfigureDoer) since all keys in a slot group share a slot and no redirect handling is needed.
2 lines
191 B
Plaintext
2 lines
191 B
Plaintext
- Fixed recurring Redis `MOVED` errors and silently-dropped report result-count increments on Redis Cluster deployments by grouping `query_results_count` keys by hash slot before pipelining.
|