Polish bundle: perf + testing small items (asyncio import, MessageQueue._size, http2 test, iterator/dataframe bench) #271

Summary

Small perf + testing-coverage items bundled for a single PR.

Item 1 — import asyncio inside AsyncTransport.request hot path

A sys.modules lookup per async request. Negligible but in the per-request hot path and pointless since the module is already imported by KalshiAuth.sign_request_async.

Location: kalshi/_base_client.py:374.

Fix: Move import asyncio to the module's top-level imports.

Item 2 — MessageQueue._size maintained per put/get is redundant with len(deque)

Overflow check uses len(self._buffer) (O(1)), but _size is incremented/decremented on every put and popleft purely so qsize() excludes the single sentinel. Two extra int ops per message on the WS hot path with no functional gain over len(self._buffer) - (1 if self._closed else 0).

Location: kalshi/ws/backpressure.py:71-100,138-140.

Fix: Drop _size and compute qsize() from len(self._buffer) minus a sentinel adjustment when _closed.

Item 3 — No behavioural test or bench for http2=True multiplexing

The advertised h2 multiplexing benefit (#233) is never exercised end-to-end. A regression that silently disables ALPN h2 negotiation, or that mishandles concurrent requests over a single h2 connection, would land without any test failing.

Location: kalshi/_base_client.py:163,347, tests/test_config.py:265-303.

Fix: Add an integration test that hits the demo endpoint with http2=True and asserts response.http_version == "HTTP/2" on a concurrent gather; add a bench variant of bench_request_hot_path.py parameterized on http2.

Item 4 — No bench for high-level Orderbook iterator or DataFrame conversion paths

Both existing orderbook benches stop at the in-place mutation and never measure the mgr.get(ticker) / to_orderbook() materialization that high-level consumers pay. DataFrame conversion (Page.to_dataframe on large pages) and large-list model_validate are completely unbenched.

Location: scripts/bench_orderbook_delta.py, scripts/bench_ws_recv.py, scripts/bench_request_hot_path.py.

Fix: Add a bench that drives _OrderbookIterator.__anext__ end-to-end, and a bench_page_to_dataframe.py that times a 1000-item Page → DataFrame conversion.

Severity & category

low / polish, testing, performance