Monitoring Redis Software with Prometheus and Datadog

Redis Software clusters expose a Prometheus-compatible metrics endpoint on each node, enabling robust observability for system performance, database health, and infrastructure metrics. This article explains how to monitor Redis Software using Prometheus for metric collection, Grafana for dashboarding, and Datadog for centralized alerting and analytics. It includes detailed Prerequisites, Step-by-Step Setup instructions for both Prometheus and Datadog, common Troubleshooting guidance, and relevant Additional Resources to support long-term monitoring.

Prerequisites

Redis Software Requirements

• Redis Software 8.0.2-17+ recommended: Use the /v2 metrics endpoint.

  • For 7.8-7.22: /v2 is available on supported patch levels; otherwise use legacy /metrics or /v1.

• metrics_exporter service running on all cluster nodes

• Metrics exposed on port 8070 over HTTPS

Monitoring Tools

• Prometheus instance (containerized or native)

• Optional: Grafana for custom dashboards

• Datadog Agent installed on a host that can reach the Redis cluster

Network Requirements

• Outbound access from Prometheus and Datadog Agent to all cluster nodes on port 8070. The cluster FQDN resolves to different node addresses over time, so every node must be reachable, but only the FQDN is configured as a scrape target. Do not add individual nodes as separate targets (see Step 2).

• Inbound access to port 9090 (Prometheus UI) and port 3000 (Grafana UI) for local access

Step-by-Step Setup for Prometheus and Grafana

1. Verify metrics_exporter is running and listening

• On each cluster node, run:

  supervisorctl status metrics_exporter

• Confirm the metrics_exporter service is in a RUNNING state.

• Optional:

• supervisorctl status | grep metrics_exporter

• If the command does not return the expected process status, you can also list all supervised processes and filter for metrics_exporter.

• Verify the endpoint is reachable:

• curl -k https://<node_ip>:8070/v2

• Any node returns the complete cluster-wide metric set, not just its own metrics, so this check confirms reachability from any single node. This is also why only one scrape target is configured in Step 2.

• To verify that the metrics endpoint is reachable, test access to the HTTPS metrics endpoint on port 8070 from the host that will run Prometheus or the Datadog Agent.

For Redis Software versions that do not support /v2, use the metrics endpoint path supported by your version, such as /metrics or /v1.

2. Create Prometheus configuration

global:
  scrape_interval: 15s
  evaluation_interval: 15s

scrape_configs:
  - job_name: redis-enterprise
    scrape_interval: 30s
    scrape_timeout: 30s
    metrics_path: /v2
    scheme: https
    tls_config:
      insecure_skip_verify: true
    static_configs:
      - targets: ["<cluster_fqdn>:8070"]

Replace <cluster_fqdn> with your cluster's FQDN.

Configure exactly one target. The v2 endpoint is cluster-wide: every node aggregates metrics from all other nodes and returns the same complete result. Adding a target per node duplicates every series and multiplies every sum()-based dashboard panel by the number of targets, with no error shown in Prometheus or Grafana. It also multiplies the internal scrape load on the cluster.

If you have no cluster FQDN available, point the single target at one node address. You will lose metrics if that specific node goes down, so an FQDN or load-balanced address is preferred.

3. Deploy Prometheus and Grafana using Docker

version: '3'
services:
  prometheus-server:
    image: prom/prometheus
    ports:
      - 9090:9090
    volumes:
      - ./prometheus/prometheus.yml:/etc/prometheus/prometheus.yml

  grafana-ui:
    image: grafana/grafana
    ports:
      - 3000:3000
    environment:
      - GF_SECURITY_ADMIN_PASSWORD=secret
    links:
      - prometheus-server:prometheus

• Start the stack:

  docker compose up -d

4. Verify data collection

• Visit http://localhost:9090 → Status > Targets → Confirm the redis-enterprise target is UP. Exactly one target should be listed for this job.

• Open http://localhost:3000 to access Grafana (default login: admin / secret)

• Import Redis dashboards from the [Redis observability GitHub repo] for cluster, node, and shard views

Step‑by‑Step Setup for Datadog

1. Install the Redis Enterprise Integration

• In Datadog, navigate to Integrations > Redis Enterprise by Redis, Inc.

• Click Install Integration

2. Configure the Datadog Agent

• Create or edit the following file: /etc/datadog-agent/conf.d/redis_enterprise.d/conf.yaml

instances:
  - openmetrics_endpoint: https://<cluster_fqdn>:8070/v2
    namespace: rdse
    ssl_verify: false

• Replace <cluster_fqdn> with your cluster's FQDN. Configure a single instance: the v2 endpoint already returns the whole cluster's metrics, so one instance per node would duplicate every metric.

• Use secure TLS configuration for production (set ssl_verify: true with valid certs).

• Restart the Datadog Agent to apply changes:

sudo systemctl restart datadog-agent

3. Validate Integration and Explore Dashboards

• In Datadog, go to Infrastructure > Host Map or Metrics > Explorer

• Search for metrics in the rdse.* namespace

• Open and use the built‑in Redis Enterprise dashboards to monitor cluster health

Migrating from v1 to v2

• /v2 exports raw (not pre-aggregated) time-series metrics, enabling greater accuracy and scalability.

• Use Redis’ v1→v2 PromQL mapping or prebuilt dashboards to migrate existing charts and panels.

• For details, see the Redis 8.0.2-17 Release Notes.

Version Comparison: v1 vs v2 Metrics Exporter

Use this table to understand key differences between Redis Software versions prior to and after 8.0.2-17:

Troubleshooting

Dashboard values look too high, or series appear duplicated

• Almost always caused by scraping more than one node. Each node returns the whole cluster's metrics, so N targets means N copies of every series and sum() panels inflated N times.

• Confirm your target count in Prometheus under Status > Targets: the redis-enterprise job should show exactly one target.

• To check from the query side, run count(up{job="redis-enterprise"}) and count(node_available_memory_bytes{node="1"}). Both should return 1. Any higher value is the number of duplicate scrape targets.

• Fix by reducing static_configs to a single cluster-FQDN target.

Prometheus can’t connect to targets

• Check that metrics_exporter is running and port 8070 is open on each node

• Verify you're using the correct /v2 endpoint for Redis Enterprise 7.8.2+

Grafana dashboards are empty

• Check that the Prometheus data source is correctly configured

• Confirm that scrape intervals are appropriate for the metric frequency

Datadog metrics not appearing

• Confirm conf.yaml contains correct node IPs and valid endpoint paths

• Review agent logs using:

  journalctl -u datadog-agent

High shard or proxy CPU

• Before acting on CPU figures, confirm you are scraping a single target (see "Dashboard values look too high" above). Duplicate scraping inflates CPU panels by the number of targets and has led customers to reshard databases unnecessarily.

• Note that per-shard and per-node CPU panels sum across threads, where 100% equals one core, so values above 100% are expected for multi-threaded processes and are not by themselves a saturation signal. A Redis shard executes commands on a single main thread, so main-thread CPU approaching 100% of one core is the meaningful saturation indicator.

• Use rdse.shard_cpu_pct to identify load hotspots

• Investigate whether a hot key or slot is concentrating load on one shard before resharding, since resharding does not help when a single key is the bottleneck

• Consider resharding the database or scaling cluster resources

• Cannot scrape /v2: The cluster may still run a pre-8.0.2 version—use /v1 instead.

• 401/403 errors: Check metrics_auth and verify Prometheus/Datadog credentials.

• Empty dashboards after upgrade: Import the v2 preconfigured dashboards or update queries using the mapping guide.

• Metrics mismatch between nodes: First confirm you are scraping a single target (see "Dashboard values look too high" above), since duplicate scraping is the most common cause. If the target count is correct, verify all nodes are on the same exporter version and that scrape intervals are consistent.

Additional Resources

Prometheus and Grafana integration with Redis Enterprise

Datadog integration with Redis Enterprise