fix: Close streaming connection pool on FDv1 data source shutdown

[jsonbailey]

Summary

The legacy FDv1 streaming data source now retains and force-closes its urllib3 PoolManager on shutdown, so the streaming TCP socket is torn down promptly instead of lingering until GC. This mirrors the fix already present in the FDv2 data source (ldclient/impl/datasourcev2/streaming.py).

Root cause

StreamingUpdateProcessor (ldclient/impl/datasource/streaming.py) created its own PoolManager, passed it to the eventsource SSEClient, but never retained or closed it. SSEClient.close() only releases the active connection back to the pool and does not close a caller-supplied pool (eventsource only closes pools it created — correct ownership). Under urllib3 >= 1.26.16 / 2.x, PoolManager.clear() no longer closes sockets synchronously (teardown is deferred to weakref.finalize on GC), so the streaming socket lingered ESTABLISHED after LDClient.close().

Fix

• Retain the pool as self._sse_pool when building the SSEClient.
• Add a module-level _close_pool_manager(pool) helper that iterates pool.pools, closes each HTTPConnectionPool, then calls pool.clear() — wrapped in try/except with debug logging. Guarded on pool is None and idempotent, so it is safe to call from both the run thread and a stopping thread.
• Call it after self._sse.close() in both run() (tail) and __stop_with_error_info().

This mirrors the FDv2 data source's approach. A small local copy of the helper was chosen over importing from the FDv2 module to avoid backward coupling (legacy depending on v2); the helper is ~15 lines.

Test

Added test_stop_closes_underlying_pool to ldclient/testing/impl/datasource/test_streaming.py, modeled on the FDv2 test_streaming_closes_underlying_pool_on_fallback. It injects a tracking pool + fake SSE client, calls stop(), and asserts the SSE client is closed, the pool is cleared, and the underlying connection pool is closed.

Notes

Should ship bundled with the launchdarkly-eventsource 1.7.1 bump (the SDK already pins >=1.7.1 on main).

Closes SDK-2668

---

Note

Low Risk
Localized shutdown/cleanup change with idempotent pool teardown and unit test coverage; no auth, flag evaluation, or data-path changes.

Overview
FDv1 StreamingUpdateProcessor now keeps the urllib3 PoolManager used for SSE and force-closes it on shutdown, so the streaming TCP socket is torn down when the client stops instead of staying open until GC (urllib3 ≥ 1.26.16 / 2.x behavior when SSEClient.close() only returns the connection to the pool).

A local _close_pool_manager closes each HTTPConnectionPool in the manager then clear()s it; it runs after SSEClient.close() in both the run() loop exit and stop() / __stop_with_error_info(), and is safe to call from either thread. FDv2 streaming only bumps pool-close failure logs from debug to warning to match v1.

Adds test_stop_closes_underlying_pool to assert stop() closes the fake SSE client, clears the pool, and closes the underlying connection pool.

^{Reviewed by Cursor Bugbot for commit f45e997. Bugbot is set up for automated code reviews on this repo. Configure here.}

[jsonbailey]

Putting this in draft as I do some more testing and don't want this to get merged by accident.