> ## Documentation Index
> Fetch the complete documentation index at: https://docs.dexploit.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# Reconnect & backpressure

> Production hardening: how to recover from drops, fill missed events, and not get disconnected for being slow.

Both [WebSocket](/streaming/websocket) and [gRPC](/streaming/grpc) eventually drop. This page is the playbook for keeping a stream healthy in production.

## Reconnect

The pattern that works:

1. **Detect the close.** WS: `'close'` event. gRPC: stream `'end'` or `'error'`.
2. **Wait with exponential backoff.** Start at 1 second, double each retry, cap at 60 seconds.
3. **Reconnect.** Reset the backoff to 1 second on a successful connect.
4. **Re-send your subscriptions.** They are not persisted across reconnects.
5. **Fill the gap with REST.** From the timestamp of your last good event, query [`/api/v1/candles?start=...`](/api-reference) or [`/swaps/range?start=...`](/api-reference) for the missing window. We do **not** replay missed events on the stream.

## Backpressure

If your client is slower than the message rate, the server's send buffer fills up. When it overflows, you get disconnected with an error indicating overflow.

To stay under:

* **Don't await heavy work in the receive loop.** Push events into an in-memory queue and process them on a separate worker.
* **Coalesce.** If you're rendering a chart, you usually only need the latest event per pool per frame, not every event.
* **Filter aggressively.** Subscribe with the narrowest filter you can — a `pools` filter is much cheaper than no filter.

## Keepalive

WebSocket: send `{ "type": "ping" }` every 30 seconds. The server replies with `{ "type": "pong" }`. If you don't see a pong within \~10 seconds, treat it like a disconnect.

gRPC: configure HTTP/2 keepalive on the channel:

```typescript theme={null}
const client = new proto.dexploit.v1.PriceStream(ENDPOINT, creds, {
  'grpc.keepalive_time_ms': 30_000,
  'grpc.keepalive_timeout_ms': 10_000,
  'grpc.keepalive_permit_without_calls': 1,
});
```

Without these, half-open TCP hangs (firewall idle-timeouts, NAT drops, ungraceful peer crashes) look like the stream stalled — your client thinks it's connected but no events arrive. The keepalive ping forces the OS to surface the dead connection so your reconnect logic can fire.

## Common failure modes

| Symptom                                       | Likely cause                        | Fix                                                                                             |
| --------------------------------------------- | ----------------------------------- | ----------------------------------------------------------------------------------------------- |
| Stream silently stalls, no events for minutes | Half-open TCP                       | Configure keepalive (above).                                                                    |
| Reconnect succeeds but no events              | Forgot to re-subscribe              | Re-send your subscribe message after every connect.                                             |
| Events stop after a key rotation              | Using the old key                   | Update your client to the new key; old key works for 7 days during grace, then 401s mid-stream. |
| Constant disconnects under load               | Backpressure overflow               | Move work off the receive loop or filter more narrowly.                                         |
| 429 immediately on connect                    | Free-tier rate limit on connections | Add backoff or upgrade.                                                                         |