query-log-hardening.md

Query-log hardening and sampling

This document explains how to control persistent query-log growth under high traffic and how logging.query_log_sampling behaves.

Quick start

• Full example: example_configs/logging/query_log_hardening.yaml
• Related README section: README.md → 2.5 Stats and query log.

Sampling configuration

Use logging.query_log_sampling to reduce persistent query-log volume before rows are written to the configured stats backend.

logging:
  query_log_sampling:
    enabled: true
    sample_rate: 0.25

Fields:

• enabled (bool, default true)
  • When false, suppresses persistent query-log writes from the stats collector.
• sample_rate (number, 0..1)
  • 1.0: write all rows.
  • 0.25: keep about 25% of rows.
  • 0.0: write none.
• rate (number, 0..1)
  • Compatibility alias for sample_rate.

Legacy alias:

• logging.query_log_sample_rate is supported as a compatibility alias.

Runtime behavior notes

• Sampling applies to persistent query-log rows. It does not disable the in-memory statistics counters used for runtime snapshots/log summaries.
• Sampling is evaluated before query-log dedupe.
• Dedupe can further reduce rows via logging.query_log_dedupe.
• For sampling to have persistent effect, you must have stats enabled and a configured persistence backend selected by stats.source_backend.

Recommended companion controls

Sampling works best when combined with:

• logging.query_log_retention:
  • max_records
  • days
  • max_bytes
  • prune_interval_seconds
  • prune_every_n_inserts
• logging.max_logging_queue to bound async queue memory.
• logging.query_log_dedupe.window_seconds to suppress short-window repeats.