JingHao-Leon aeb88738e8 fix(router): sync streaming fallback re-ran the failing group instead of falling back ... _completion_streaming_iterator.stream_with_fallbacks() re-entered the fallback chain via function_with_fallbacks(), which tries the original group first. The retry returns a fresh stream that succeeds at creation time (HTTP 200) and only fails while being iterated; each nested Router.completion() wraps its own stream in this same iterator, so every retry fails again during iteration and re-enters the chain — recursing until the stack runs out (measured: 478 requests to the failing group, ~116 s, then InternalServerError, with a healthy fallback configured). Mirror the async twin: hand the MidStreamFallbackError to async_function_with_fallbacks_common_utils() via run_async_function, which cools down the failed deployment and walks the fallback list directly. Sync tests that patched function_with_fallbacks are updated to the new re-entry point, plus a regression test asserting the triggering error reaches the common utils and the original group is not re-run. Fixes #43945		2026-10-01 11:15:09 +08:00
..
__init__.py	test: add __init__.py to every tests/unit directory phase 16 touches	2026-09-20 11:52:38 +00:00
test_enforce_model_rate_limits.py	test: migrate phase 16 legacy tests to tests/unit	2026-09-20 10:59:02 +00:00
test_io_token_rate_limits.py	test: migrate phase 16 legacy tests to tests/unit	2026-09-20 10:59:02 +00:00
test_router.py	fix(router): sync streaming fallback re-ran the failing group instead of falling back	2026-10-01 11:15:09 +08:00