WS: resubscribe_one stash is never drained — frames captured during gap recovery are dropped #254

Summary

On every gap recovery (_handle_seq_gap -> resubscribe_one), frames captured between unsubscribe-ack and subscribe-ack sit in self._stash until the next full reconnect (where resubscribe_all clears them). Bounded by stash_maxlen so not unbounded, but accumulates per gap and the captured frames — which legitimately arrived for the new sid window — are never replayed through _process_frame. Post-resubscribe deltas with seq>1 can hit the seq tracker before the snapshot they describe.

Location

• kalshi/ws/channels.py:452-535 — resubscribe_one stashes but never drains
• kalshi/ws/client.py:394-431 — _drain_resubscribe_stash is only called from _handle_reconnect

Evidence

# channels.py resubscribe_one:
prev_stashing = self._stashing
self._stashing = True
try:
    ...  # unsubscribe + subscribe via _wait_for_response (stashes by sid)
finally:
    self._stashing = prev_stashing
# No take_stash() / drain anywhere along this path.

take_stash() is only called from client.py:411 _drain_resubscribe_stash, which is only called from _handle_reconnect (full reconnect), not from _handle_seq_gap -> resubscribe_one.

Recommended fix

After resubscribe_one returns successfully, drain its stash through _process_frame (same shape as _drain_resubscribe_stash) — filter by new_sid only, drop the rest with a debug log. Add regression test that injects a delta during a resubscribe window and asserts it is replayed.

Severity & category

medium / reliability, ws