Open Admin → SEMP traffic with admin:manage permission. This page controls background broker polling and shared metadata caching, and shows request usage. Changing settings requires recent MFA and records an audit event.
Cadence and caching
| Setting | Default | Allowed range |
|---|---|---|
| Environment heartbeat | Enabled, every 30 seconds | 5–86,400 seconds |
| Scheduled queue-inventory refresh | Enabled, every 1,440 minutes | 5–43,200 minutes |
| VPN-list cache lifetime | 300 seconds | 1–3,600 seconds |
| Queue-detail cache lifetime | 60 seconds | 1–3,600 seconds |
| Queue-statistics cache lifetime | 10 seconds | 1–60 seconds |
| Additional stale-on-error window | 60 seconds | 0–60 seconds |
Set global defaults and optional per-environment heartbeat/inventory overrides. A removed override restores inheritance. Precedence is environment override, saved platform settings, deployment fallback, then built-in default. Once saved, the complete platform settings become authoritative. Disable a task using its enabled switch, not a zero interval. Disabled environments are excluded from background polling.
API and Worker reload settings within their polling loops, normally at most five seconds apart. Manual inventory refresh remains available independently of scheduled cadence. These settings do not impose a request budget or delay interactive operations.
Metadata caches are shared across replicas and callers; permission and license checks still apply per request. Explicit refresh obtains new data or shares an overlapping refresh. A bounded stale value may be returned after a transient broker failure, with freshness information; authentication and not-found errors do not become stale successes. Environment connection edits invalidate metadata.
Message payloads, message browse results and queue message metadata are not cached by this feature. Test Connection and heartbeat perform live reads. If a single queue’s statistics fail, Queue explorer can show — while retaining healthy queues; refresh or investigate the affected broker path.
Interpret usage
Usage counts real SEMP HTTP attempts, including retries and failures. Cache hits, misses and refreshes are recorded separately. Counts aggregate into five-minute buckets and retain 30 days of history; they do not require an OTLP collector.
The display is near real-time, not a billing or forensic ledger. Processes flush approximately every ten seconds; crashes and failed/ambiguous database writes can lose unconfirmed counts. A selected time range includes overlapping buckets. Use audit and job records to investigate a particular action.
For collector metrics, see Monitoring and Logging.