Take the proposed resolution with a pinch of salt — drafted quickly with AI assistance. The diagnosis (sync websocket call blocking the event loop) is verified against live logs, but the exact fix shape deserves maintainer review.

Summary

send_nostr_dm() in lnbits/core/services/nostr.py uses the synchronous websocket.create_connection() inside an async function and iterates relays sequentially. One unreachable or slow relay blocks the asyncio event loop for up to ~30 seconds per attempt, making NWC Provider and other WebSocket-based extensions unresponsive during that time.

This is a separate event-loop-starvation bug from #3917 / #3918 (IN_FLIGHT payment polling). After applying the #3918 fix, this became the next visible bottleneck.

Offending code

lnbits/core/services/nostr.py, line ~30:

ws_connections: list[WebSocket] = []
for relay in relays:
    try:
        ws = create_connection(relay, timeout=2)   # sync call in async fn
        ws.send(notification)
        ws_connections.append(ws)
    except Exception as e:
        logger.warning(f"Error sending notification to relay {relay}: {e}")
await asyncio.sleep(1)

Issues:

1. create_connection() is synchronous — it blocks the event loop on TCP connect + TLS handshake
2. The timeout=2 is not always honored (we observe 30s blocks in production against wss://relay.snort.social and wss://relay.nostr.band)
3. Relays are tried sequentially — 3 dead relays = 3 × blocking time added up
4. ws.send() is also sync

With 3 dead relays in the user's configured list, a single send_nostr_dm() call can block the event loop for 60–90 seconds.

Symptoms

• NWC list_transactions, get_balance, pay_invoice intermittently time out
• docker logs repeatedly shows:

  WARNING | Error sending notification to relay wss://nostr.wine: Handshake status 403 Forbidden
  WARNING | Error sending notification to relay wss://relay.snort.social: Connection timed out
  WARNING | Error sending notification to relay wss://relay.nostr.band: timed out
  

• Web UI works (HTTP served separately), masking the issue
• Only recovery from a pile-up is docker restart

Proposed fix

Two options (either or both):

A. Offload sync calls via asyncio.to_thread and run concurrently:

async def _publish(relay: str):
    try:
        ws = await asyncio.wait_for(
            asyncio.to_thread(create_connection, relay, timeout=2),
            timeout=3,
        )
        await asyncio.to_thread(ws.send, notification)
        return ws
    except Exception as e:
        logger.warning(f"Error sending notification to relay {relay}: {e}")
        return None

results = await asyncio.gather(*(_publish(r) for r in relays))
ws_connections = [ws for ws in results if ws]

B. Switch to an async websocket library (websockets or aiohttp) — larger change but removes the sync/async footgun entirely.

Option A is minimally invasive and fixes the starvation. Happy to open a PR with that approach.

Related

• [BUG] IN_FLIGHT payment polling starves asyncio event loop, breaking NWC and WebSocket extensions #3917 — IN_FLIGHT payment polling starves asyncio event loop (sibling bug)
• fix: exponential backoff for IN_FLIGHT payment polling #3918 — draft fix for [BUG] IN_FLIGHT payment polling starves asyncio event loop, breaking NWC and WebSocket extensions #3917 (currently in production testing)
• nostr client extension becomes temporarily unresponsive nostrclient#40 — nostrclient temporarily unresponsive (same class of symptom, different trigger)
• [BUG] Publishing to unreachable relays blocks asyncio event loop (NWC timeouts) nostrclient#69 — initially misfiled there; moved here after tracing to lnbits/core/services/nostr.py

Environment

• LNbits v1.5.3
• LNDRest backend
• Python 3.x, asyncio