The meaningful numbers to know when operating OpenFlows. Most of these are deliberate design choices rather than
settings you’d tune — they’re presented here in plain language as behaviors and ranges, not as configuration
constants.
Health & liveness
Number
Meaning
Beacon refresh
Workers refresh their liveness beacon on a short cadence.
Beacon lifetime
If a beacon stops refreshing, it expires on its own shortly after.
Staleness threshold
A worker silent past this point is declared stale and recoverable.
Net effect: OpenFlows detects a dead or unresponsive worker within roughly a couple of minutes, without needing to
poke every container.
Recovery
Number
Meaning
Recovery attempts
A bounded retry count per ticket.
After exhausting
The ticket escalates to a human instead of retrying forever.
Recovery is bounded, then escalated — the system stops and asks rather than retrying forever.
Notifications
Limit
Meaning
Per-channel cooldown
At most one notification per channel per ticket within a window — prevents alert floods.
Failure safety
A failing channel is logged, never fatal to the orchestration.
Delegated verification
During delegated verification, the reviewer asks the builder to run a safe command and return evidence:
Limit
Meaning
Command set
Only a fixed allowlist of safe commands can be run during delegated verification.
Timeout cap
A verification run has a hard maximum timeout.
Output bound
Captured output is bounded so large logs don’t overflow the exchange.
Orchestration pacing
Number
Meaning
Pass interval
The cadence at which the Controller re-reads the whole system and advances work.
Pass bound
A single pass is bounded so it can’t loop forever; it stops and retries next beat.
Self-healing
A bad pass is logged and retried — it never kills the controller.
Durability
Item
Meaning
Coordination store
Configured for durable storage so team state survives restarts.
Two-writer rule
Only the controller and the worker surface may write; writes are validated.
