WS: order_group_updates sequence gap has no recovery path (consumer silently misses events) #205

Summary

SEQUENCED_CHANNELS includes order_group_updates, so gaps ARE detected on that channel — SequenceTracker.track() fires on_gap and returns False so dispatch is skipped. But _handle_seq_gap (in kalshi/ws/client.py) only has a recovery branch for sub.channel == 'orderbook_delta'. For an order_group_updates gap:

1. The gap-causing message is dropped (correctly — dispatch was skipped).
2. The seq tracker advances _last_seq[sid] to the new seq (sequence.py:70).
3. No resubscribe, no state reset, no signal to the consumer.

The consumer iterator silently misses one or more events. For order-group machinery tracking contracts_limit / event_type transitions, this leaves the consumer's view permanently diverged from the server.

This is a sibling of the orderbook-resync issue but on a different channel and with no recovery code path at all (orderbook at least clears local state).

Location

• kalshi/ws/client.py:393-403 — branch only for orderbook_delta
• kalshi/ws/sequence.py:68-75 — gap path advances watermark before returning False

Evidence

# kalshi/ws/client.py:393-403
if sub and sub.channel == "orderbook_delta":
    tickers = sub.params.get("market_tickers", [])
    ...
# ← order_group_updates falls off the end; no recovery

# kalshi/ws/sequence.py:68-75
gap = SequenceGap(sid=sid, expected=expected, received=seq)
logger.warning("Sequence gap: ...")
self._last_seq[sid] = seq   # advance, so future gaps won't fire here either

if self._on_gap is not None:
    await self._on_gap(gap)
return False

Recommended fix

Route non-orderbook sequenced channels through a generic resubscribe helper. For order_group_updates:

1. Unsubscribe and re-subscribe to the affected sid (the server will re-emit the current order-group state).
2. Until the resubscribe completes, stash incoming frames (the resubscribe-window mechanism from WS: data frames arriving during reconnect+resubscribe are dropped instead of stashed #176 already exists in SubscriptionManager — reuse it).
3. If resubscribe fails, surface KalshiSequenceGapError through the consumer iterator (see separate "WS error context" issue) with channel='order_group_updates', sid, last_seq, next_seq.

Add a regression test that injects a gap on order_group_updates (e.g., via direct SequenceTracker._last_seq[sid] = 5 then send seq=10) and asserts a recovery action fires.

This issue and the orderbook-resync issue (critical) share the same _handle_seq_gap body — fix them together so the per-channel routing is unified.

Severity & category

medium / correctness, ws