RunPodClient configures a retry policy for 408/429 responses but mounts it only for http://. The normal endpoint base URL uses HTTPS, so those requests use Requests' default adapter with retries disabled.
Reproduced on the current main branch without a RunPod account or endpoint:
from runpod.endpoint.runner import RunPodClient
client = RunPodClient(api_key="test-key")
try:
print(client.rp_session.get_adapter("http://example.test").max_retries.total) # 5
print(client.rp_session.get_adapter("https://example.test").max_retries.total) # 0
finally:
client.rp_session.close()A local server returning 429 once and 200 next confirms the behavioral difference: the HTTP GET retries and succeeds, while the HTTPS GET raises HTTPError after the first response. The HTTPS test uses a locally generated certificate trusted explicitly by the test client.
Expected: the configured retry policy applies to HTTPS endpoint requests too. AI assistance was used for investigation and local verification; no RunPod resources were provisioned.
RunPodClientconfigures a retry policy for 408/429 responses but mounts it only forhttp://. The normal endpoint base URL uses HTTPS, so those requests use Requests' default adapter with retries disabled.Reproduced on the current main branch without a RunPod account or endpoint:
A local server returning 429 once and 200 next confirms the behavioral difference: the HTTP GET retries and succeeds, while the HTTPS GET raises
HTTPErrorafter the first response. The HTTPS test uses a locally generated certificate trusted explicitly by the test client.Expected: the configured retry policy applies to HTTPS endpoint requests too. AI assistance was used for investigation and local verification; no RunPod resources were provisioned.