Production readiness
This page explains what it takes to run DIAL reliably in production and links to the task-by-task guides that get you there. It is for DevOps and platform engineers who have a working deployment and now need it to survive load, failure, upgrades, and audit. Read it top to bottom once, then use the linked how-tos as a checklist. No setup steps are here.
The model that drives every decision
Almost every production-readiness choice in DIAL follows from one fact about how the platform stores state:
- DIAL Core keeps durable resources outside its service instances. Optional services can have their own databases or state stores.
- Blob storage persists resources — conversations, files, published resources, and application state.
- Valkey holds cached resources, pending writes, and coordination state. Protect its availability and account for unsynchronized changes in recovery. See State and data lifecycle.
Scale the service tier separately from its stores. Protect blob data, pending writes, optional component databases, and configuration. Preserve the keys required to address existing resources and decrypt stored credentials.
Readiness checklist
Each item below is a single how-to. Work through them in order for a new environment, or jump to the one you need.
| Area | Goal | Guide |
|---|---|---|
| High availability | Reduce disruption from node or zone failure | Run DIAL with high availability |
| Scaling | Scale for the measured workload and upstream capacity | Scale DIAL for load |
| Secrets management | Keep keys and credentials out of values files | Manage secrets |
| Backup and restore | Recover permanent data after a failure | Back up and restore DIAL |
| Upgrade procedure | Move to a new chart version safely | Upgrade DIAL |
| Cost control | Cap and monitor model spend | Control cost |
| Security hardening | Reduce the attack surface and control data retention | Harden DIAL for production |
How this section relates to the rest of Operating DIAL
Production readiness is platform-neutral. It states the principle and the Helm or DIAL Core setting, then points to the deployment guide for your cloud, which carries the provider-specific detail.
- Cloud deployment — the per-cloud guides hold the concrete service tiers, networking, and disaster-recovery notes for AWS, Azure, GCP, and generic Kubernetes.
- Observability — you cannot operate what you cannot see; metrics, alerting, and analytics underpin every page here.
- Configuration reference — the authoritative description of every setting these guides change.