Manual transport comparison scripts for the Runloop Python SDK. These hit the real API and are not part of pytest; run on demand.

All scripts require RUNLOOP_API_KEY. Set RUNLOOP_BASE_URL to override the default endpoint (https://api.runloop.ai).

Every script sends devboxes.create requests against a deliberately nonexistent blueprint (bp_nonexistent_loadtest_00000), so every request fails fast server-side (HTTP 400) and no devboxes are created — isolating client + server request handling from provisioning.

The raw-transport probes compare httpx HTTP/2 multiplexing against HTTP/1.1 directly — the same comparison that motivates the SDK's shared HTTP/2 pool. They're kept so the comparison stays reproducible.

pool_check.py is a lightweight sanity check (no API key, no real requests) that confirms multiple sync Runloop instances reuse a single underlying transport object — preventing file-descriptor exhaustion when many clients are instantiated.

# Install the SDK from this checkout (editable):
cd /path/to/api-client-python && uv sync

# SDK: HTTP/2 (default) vs HTTP/1.1, 2000-request burst
source ~/env && REQUEST_COUNT=2000 uv run python loadtest/loadtest.py            # HTTP/2
source ~/env && REQUEST_COUNT=2000 USE_HTTP2=0 uv run python loadtest/loadtest.py # HTTP/1.1

# Raw httpx HTTP/2 vs HTTP/1.1 comparison
source ~/env && uv run python loadtest/h2_test.py
source ~/env && uv run python loadtest/raw_fetch_test.py

# Single-connection burst
source ~/env && uv run python loadtest/h2_single_conn.py

# ALPN check
source ~/env && uv run python loadtest/alpn_check.py

HTTP/1.1 opens a socket per in-flight request; for large bursts raise the file-descriptor limit (ulimit -n 65536) or keep REQUEST_COUNT small.