Reconnect Bench

Controls

Recovery policy

Choose No backoff (one attempt per 500 ms model cycle) or capped exponential backoff for the synthetic client sessions.

Backoff base

Set the illustrative first delay used by the exponential schedule.

Spread retries with jitter

Apply a deterministic per-device spread to the example retry schedule; it is not a documented Seronic setting.

Device to inspect

Choose which synthetic device session is summarized above the fleet list.

Disconnect selected device

Interrupt only the chosen client connection. The shared synthetic service stays available, so its next scheduled retry can reconnect.

Trigger shared interruption

Disconnect all ten example sessions while keeping their generic app state visible.

End shared interruption

Allow due retries to reconnect to the available synthetic service.

Advance time by 500 ms

Available after you disconnect a client or trigger a shared interruption. Advances the synthetic clock and processes retries that are due.

Reset experiment

Restore ten connected sessions and the default No backoff example with 500 ms attempt cycles.

What to watch for

• Disconnect one selected client, then advance one 500 ms step. Its scheduled attempt reconnects because the shared service is still available.
• Trigger a shared interruption, then advance one 500 ms step. No backoff schedules every client for the same 500 ms cycle; after service returns, they can rejoin together and create a burst in this simplified model.
• Reset first, choose Exponential backoff, and trigger a shared interruption. Keep the service down while you advance time; compare the 500 ms, 1,000 ms, and 2,000 ms retry intervals. Restore service before retries run out.
• Enable jitter and compare the deterministic per-device schedule with the synchronized exponential schedule.
• Inspect a device while disconnected: the transport is down, while its separate generic app-state indicator remains retained.

Why it happens

A disconnected client must decide when to try again. This comparison schedules the No backoff policy on a fixed 500 ms model cycle; identical cycles can align. Exponential delays increase the wait after repeated failures, while jitter can spread clients that would otherwise share the same schedule. Real reconnect behavior also depends on detection, protocol state, subscriptions, server limits, platform lifecycle, and delivery semantics, which this client-side teaching model does not simulate.