Skip to main content

Production Playbook

This page is the operator-oriented map for adopting Spring Prism in real environments.

Decision Matrix​

NeedRecommended Path
Single application instancevault.type=in-memory
Multiple nodes or Kubernetesvault.type=redis
Deterministic detector set onlystarter without NLP extension
Person-name redaction with conservative rolloutNLP extension with backend=heuristic
Person-name redaction with stronger recallNLP extension with backend=hybrid and explicit model

Minimal production baseline​

Every serious deployment should have:

  • a non-default spring.prism.app-secret
  • starter autoconfiguration enabled
  • metrics endpoint available through Actuator or /prism/metrics
  • integration tests covering the runtime path you actually use
  • docs and rollout notes synchronized with the chosen deployment mode

spring.prism.failure-mode=FAIL_SAFE remains the default posture and preserves the established fail-open behavior with metrics. For stricter regulated environments, prefer spring.prism.failure-mode=FAIL_CLOSED once you have validated the rollout and operational dependencies. See Configuration: Failure Mode.

Single-node rollout​

Use this when request tokenization and response restoration happen inside the same process or node.

Recommended settings:

spring:
  prism:
    app-secret: ${PRISM_APP_SECRET}
    vault:
      type: in-memory
    ttl: 30m

Use this path when:

  • the app is local or developer-facing
  • the workload is simple and non-distributed
  • you do not need cross-node restore

Multi-node rollout​

Use this when restore can happen on a different pod or node.

Recommended settings:

spring:
  data:
    redis:
      host: redis.internal
      port: 6379
  prism:
    app-secret: ${PRISM_APP_SECRET}
    vault:
      type: redis
    ttl: 30m

This path is only production-ready when:

  • every node uses the same shared Redis deployment
  • every node uses the same non-default spring.prism.app-secret
  • TTL covers the realistic request-to-restore latency window

NLP rollout​

Phase 1: heuristic​

Start here when you want person-name redaction with low operational friction.

spring:
  prism:
    extensions:
      nlp:
        enabled: true
        backend: heuristic
        confidence-threshold: 4

Phase 2: hybrid​

Promote to this path only after validating the model and blocked phrases on realistic text.

spring:
  prism:
    extensions:
      nlp:
        enabled: true
        backend: hybrid
        model-resource: file:${PRISM_NLP_MODEL}
        confidence-threshold: 4

Production rules for hybrid:

  • the model artifact must be versioned and deployed consistently on every node
  • startup should fail if the model is missing or unreadable
  • blocked phrases should include domain-specific technical vocabulary

Definition of done for feature work​

Any new runtime feature or behavior change should include:

  • unit tests in the owning module
  • at least one end-to-end scenario in prism-integration-tests when behavior crosses module or runtime boundaries
  • Docusaurus docs updates before merge
  • release_notes.v1.1.0.md updates while the release line is active

Release-train checklist​

Before merging into v1.1.0, confirm:

  • the chosen runtime mode is documented
  • example config exists or is updated
  • integration tests cover the user-facing behavior
  • metrics or readiness output is still truthful after the change
  • README still reflects the recommended onboarding path

Where to go next​