From 6024f9e45e4b9dcdc1880a6b67a5d92f09fdb3aa Mon Sep 17 00:00:00 2001 From: Krrish Dholakia Date: Fri, 7 Jun 2024 11:27:48 -0700 Subject: [PATCH] test(test_lowest_latency_routing.py): add time.sleep to reduce test flakiness --- litellm/tests/test_lowest_latency_routing.py | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/litellm/tests/test_lowest_latency_routing.py b/litellm/tests/test_lowest_latency_routing.py index f9f510673d..d83cdef9ad 100644 --- a/litellm/tests/test_lowest_latency_routing.py +++ b/litellm/tests/test_lowest_latency_routing.py @@ -275,7 +275,7 @@ async def _deploy(lowest_latency_logger, deployment_id, tokens_used, duration): } start_time = time.time() response_obj = {"usage": {"total_tokens": tokens_used}} - time.sleep(duration) + await asyncio.sleep(duration) end_time = time.time() lowest_latency_logger.log_success_event( response_obj=response_obj, @@ -325,6 +325,7 @@ def test_get_available_endpoints_tpm_rpm_check_async(ans_rpm): d1 = [(lowest_latency_logger, "1234", 50, 0.01)] * non_ans_rpm d2 = [(lowest_latency_logger, "5678", 50, 0.01)] * non_ans_rpm asyncio.run(_gather_deploy([*d1, *d2])) + time.sleep(3) ## CHECK WHAT'S SELECTED ## d_ans = lowest_latency_logger.get_available_deployments( model_group=model_group, healthy_deployments=model_list