Sizing and capacity
Minimum support envelopes for the single-node, split and HA-ready self-hosted profiles.
RC
Exact release scope. This page documents the reviewed v0.0.1 source at fcc5871. Check release status before enabling a gated capability.
Profile baselines
| Profile | API / control plane | Agents | PostgreSQL | Artifacts |
|---|---|---|---|---|
| profile-single-node | 1 API: 4 vCPU / 16 GiB | 1 agent: 2 vCPU / 4 GiB | PostgreSQL: 2 vCPU / 8 GiB / ≥200 GiB / ≥1,000 IOPS | Artifacts: 50 GiB baseline |
| profile-split | 1–4 APIs: 4 vCPU / 16 GiB each | ≥2 agents: 4 vCPU / 8 GiB each | PostgreSQL: 4 vCPU / 16 GiB / ≥500 GiB / ≥3,000 IOPS | Shared artifacts: ≥500 GiB |
| profile-ha-ready | 2–12 APIs: 8 vCPU / 24 GiB each | ≥4 agents: 8 vCPU / 16 GiB each | HA PostgreSQL primary: ≥8 vCPU / 32 GiB / ≥2 TiB / ≥6,000 IOPS | Shared artifacts: ≥2 TiB |
Capacity signals
| Signal | Warning interpretation | First response |
|---|---|---|
| clusterpilot_storage_usage_bytes | ≥80% for 15m warns; ≥90% is critical; ≥95% blocks promotion | Review retention, exports, holds and storage headroom |
| clusterpilot_run_queue_depth | Queue freshness is outside the profile SLO | Check agent availability and bounded concurrency before adding API replicas |
| clusterpilot_db_pool_utilization_ratio | Database is the limiting dependency | Inspect query latency, IOPS and pool configuration |
| SSE active streams | Long-lived stream pressure approaches the profile limit | Review clients, proxies and stream quotas |
| Retry / dead-letter backlog | Dependencies or idempotency are not converging | Classify failures; do not increase retries first |
Storage planning
Size PostgreSQL for run history, projections and audit. Size artifact storage for logs, command evidence, conformance archives, exports, holds and retention.
- ✓Document retention by environment and evidence class
- ✓Budget export and temporary-object headroom
- ✓Keep backup storage independent from the active artifact store
- ✓Use shared filesystem or S3-compatible storage for multiple replicas
- ✓Benchmark backup and restore time against the business RTO
Scale safely
Scale API, workers, agents, PostgreSQL and artifacts as one queueing system. Adding API replicas without leader election, shared artifacts and agent headroom is unsupported. Re-run soak, failure and restore acceptance after any topology change.