Site search

Find architecture, research, and terms

Start typing to search the editorial index.

Production practiceAdvanced

MIR Deployment and Operations

Operational deployment guidance for logical boundaries, environments, secrets, atomic release manifests, rollback, runbooks, SLOs, and evidence compatibility.

Logical versus physical architecture

Logical boundaries are mandatory; physical distribution is optional. A modular monolith can be correct, and a microservice deployment can still be poorly bounded. Preserve lifecycle, authority, adapters, context, tools, recovery, and evidence even when they share a process.

Environment model

Local development

Deterministic fakes, disposable workspace, and no production credentials.

Integration and staging

Representative policies, schemas, queues, evidence, and failure injection.

Production

Version-pinned routes, least privilege, monitored SLOs, and tested rollback.

Isolated or edge runtime

Separate high-risk or disconnected execution with explicit synchronization and evidence export.

Configuration and secrets

  • Version non-secret configuration; store secrets in a vault or secret manager.
  • Never place secrets in prompts or evidence.
  • Pin environment-specific model routes, policy versions, and schema versions.
  • Bind credentials to tenant, action, destination, and expiration.

Atomic release manifest

Version together application/runtime code, runtime profile, model route and model version, prompt contract, tool schemas, policy version, evidence schema, database/event migrations, and container/image digests.

Rollback

  • Retain previous adapters and preserve schema compatibility.
  • Drain or explicitly migrate active runs.
  • Do not silently change tool semantics.
  • Record control-plane deployment events.
  • Test current evidence readers against historical events.

Operational runbook

Capacity and dependency incidents

Queue saturation, model provider outage, policy-service outage, evidence-store degradation, and approval backlog.

Control incidents

Stuck recovery loop, credential incident, context contamination, deployment rollback, and emergency termination.

SLOs and metrics

Outcome

Task success and first-attempt success.

Recovery

Recovery rate and operator intervention.

Governance

Policy-compliant completion and approval latency.

Operations

Tool reliability and cost per completed workflow.

Evidence

Completeness, integrity, and read compatibility.

Source record

References

Suggest a correction
  1. Microsoft. Azure Architecture Center. Published Current documentation; last reviewed 2026-06-23 UTC. Official architecture guidance.

  2. National Institute of Standards and Technology. NIST. Published 2023-01-26; last reviewed 2026-06-20 UTC. Government framework.

  3. OpenTelemetry project. Cloud Native Computing Foundation. Published Current specification repository; last reviewed 2026-06-24 UTC. Official specification.