Adding query analytics to PMM’s Valkey monitoring in 10 minutes

Percona Monitoring and Management (PMM) already watches your Valkey instances. This page covers where BetterDB Monitor fits alongside it, and how to wire the two together without replacing anything PMM already does.

What PMM already gives you

PMM’s Valkey/Redis monitoring covers host-level and server vitals: CPU, memory, disk and network for the node, plus the server-level INFO metrics — uptime, connected clients, keyspace hits/misses, replication status — on its stock Valkey dashboards.

What BetterDB adds

BetterDB Monitor adds the metric families PMM doesn’t collect today:

  • Slowlog pattern analysis — betterdb_slowlog_pattern_count, betterdb_slowlog_pattern_avg_duration_us, betterdb_slowlog_pattern_percentage, betterdb_slowlog_length: which query patterns are slow, not just which individual commands were.
  • Commandlog pattern analysis — betterdb_commandlog_large_request_by_pattern, betterdb_commandlog_large_reply_by_pattern: oversized requests and replies, grouped by pattern.
  • Cluster slot distribution — betterdb_cluster_slot_keys, betterdb_cluster_slot_reads_total, betterdb_cluster_slot_writes_total, betterdb_cluster_slot_expires: per-slot key counts and read/write volume, for spotting hot or unbalanced slots.
  • Anomaly and forecast series — betterdb_anomaly_events_total, betterdb_anomaly_by_severity, betterdb_correlated_groups_total, betterdb_metric_forecast_time_to_limit_seconds: detected anomalies grouped by severity and correlated pattern, and a forecast of when a trending metric will hit its limit.

Wiring it up

Pick whichever fits your Prometheus setup better:

  1. Run the monitor next to your Valkey, pointed at it with DB_HOST / DB_PORT (and credentials, if set), then add a scrape job to the Prometheus instance PMM reads from:

    scrape_configs:
  - job_name: 'betterdb'
    metrics_path: '/api/prometheus/metrics'
    static_configs:
      - targets: ['<monitor-host>:3001']

  2. Or, if you’d rather not add a scrape target, point the monitor’s OTLP mirror at a Collector you already run: OTEL_EXPORTER_OTLP_ENDPOINT=http://<collector>:4318, then scrape the Collector’s Prometheus exporter instead. See OpenTelemetry Collector for the config.

Either path lands the same betterdb_* series in the Prometheus PMM already queries.

Importing the dashboards

Import BetterDB · Query Patterns and BetterDB · Anomalies from deploy/observability/dashboards/ into the same Grafana that renders your PMM dashboards — see the observability pack README for the import steps. They sit fine next to PMM’s stock Valkey dashboards; both read from the same Prometheus, just different metric families.

What’s not covered yet

The docker-compose demo in deploy/observability/demo/ ships Grafana only — it doesn’t include a PMM container. If you want to try the wiring above end-to-end, bring your own PMM install and point it at the same Prometheus the demo (or your own monitor) is feeding.