Probe behaviorRuntime safety
Runtime safety
Interpret normalized safety states, enforced limits, runtime-specific reason codes, and cooldown behavior.
Safety overview
get_safety_overview returns the broker-derived online state, latest agent state, raw detail, enforced limits, probe status counts, and caveats for each service. Green means the agent has not reported a LiveProbe safeguard trip. It is not a general application health signal.
Canonical limits
| Environment variable | Default | Runtimes |
|---|---|---|
LIVEPROBE_MAX_PROBE_HITS_PER_SECOND | 10 | Node, Python, JVM |
LIVEPROBE_MAX_PROBE_PAUSE_MS_PER_SECOND | 20 ms | Python |
LIVEPROBE_MAX_TELEMETRY_BYTES_PER_SECOND | 204,800 | Node, Python |
LIVEPROBE_MAX_BUFFERED_EVENT_BYTES | 5 seconds of telemetry, minimum 64 KiB | Node |
LIVEPROBE_MAX_EVENT_LOOP_LAG_MS | 50 ms p95 | Node |
LIVEPROBE_SAFETY_COOLDOWN_MS | 10,000 ms | Node, Python |
Agents report only safeguards they actually enforce. An omitted limit means unsupported, not zero utilization.
Reason codes
| Code | Meaning |
|---|---|
event_loop_lag | Node event-loop p95 exceeded the configured threshold |
pause_budget | Python callbacks exceeded their per-second time budget |
rate_limited | JVM breakpoint requests were suspended by hit-rate protection |
instrumentation_failure | The runtime could not install or maintain instrumentation |
agent_worker_failure | An internal agent worker failed |
Runtime semantics
A red state describes LiveProbe's own runtime activity. It does not measure CPU load, memory pressure, garbage collection, or overall service availability. Reduce probe rate, remove hot-path snapshots, and wait for cooldown before retrying.