Operate self-managed Intent Router

Check health, rotate credentials, watch memory, update configuration, back up, upgrade, and uninstall Redis Intent Router.

This page describes how to operate self-managed Intent Router after you deploy it and give Routers to developers. It covers routine checks, such as service health and data-plane Redis memory, and how the deployment responds when Redis is unavailable. It also covers the maintenance tasks you run with helm upgrade: rotating credentials, updating Secrets, and upgrading the chart. It ends with backing up and uninstalling the deployment.

Check service health

Every service answers /health/liveness and /health/readiness without a credential:

curl -sS http://localhost:9400/health/readiness
curl -sS http://localhost:9401/health/readiness
curl -sS http://localhost:9200/health/readiness

Check one Router with its health route. It returns ok and a checks list that covers the Router's state, the embedding provider, the intent store, and the search index. A failing check carries a message that says why:

curl -sS "$IRIS_INTENT_ROUTER_URL/iris/data/v1/intent/routers/$IRIS_INTENT_ROUTER_ID/health" \
  -H "Authorization: Bearer $IRIS_INTENT_ROUTER_API_KEY" | jq .

With tests.enabled=true, helm test intent -n <namespace-name> probes every service and runs the shared security-profile checks.

Rotate credentials

Rotate the control plane and Identity Service tokens as described in Rotate credentials. To rotate an embedding provider credential, update embedders_connection_details in the inference Secret, update inference.existingSecretChecksum, and run helm upgrade, so the data plane rolls.

Watch the data-plane Redis

The data-plane Redis holds every Router's examples and vector index. Watch its memory against the estimate in Size memory, and add memory before a large authoring run. Intent limits stop a Router growing past dataplane.limits.max_intents_per_router, but not past the memory of the database. A Router whose database runs out of memory answers 424 with /errors/database-out-of-memory.

The data plane rebuilds a Router's index if it is lost, and reports the Router ready only after the index is built. While it rebuilds, match requests return 503 with Retry-After.

Handle Redis outages

A Router whose database is unreachable answers 503 with Retry-After until the database returns. See Customer databases for what happens to a new Router.

Update configuration and license Secrets

When you change a Secret, update its existingSecretChecksum in the values and run helm upgrade, so the pods roll and read the new content. Changing a customer database URL needs this restart.

Back up the deployment

Back up the metadata Redis, which holds the Router records, and the data-plane Redis, which holds intents and examples. Back up the Identity Service metadata too, so agent keys survive a restore. Keep the Secrets and the values file in your secret store and source control.

A restored data-plane Redis rebuilds its Router indexes. Embeddings are stored with each example, so a restore doesn't call the embedding provider.

Upgrade the chart 

helm repo update redis-ai
helm upgrade intent redis-ai/redis-intent-router \
  --version <chart-version> \
  --namespace <namespace-name> \
  -f intent-values.yaml \
  --wait --timeout 5m

Set the image tags to the new chart version. Check readiness on every service afterward.

Uninstall the chart 

helm uninstall intent --namespace <namespace-name>

Uninstalling doesn't delete the Redis databases or the Secrets you created. The chart keeps the auto-generated token Secrets (helm.sh/resource-policy: keep), so delete them yourself if you don't plan to reinstall.

RATE THIS PAGE
Back to top ↑