6 views
-/https://github.com/berriai/litellm/issues/28474
GitHub · issue

#28474 [Bug]: The litellm call will disconnect after 60-70 seconds.

  • State: open
  • Author: @yuanhongda
  • Labels: bug, llm translation

### Check for existing issues

- [x] I have searched the existing issues and checked that my issue is not a duplicate.

### What happened?

ERROR Unexpected error calling OpenAI API: peer closed connection without sending complete message body (incomplete chunked read) 2026-05-21 21:44:55.244 | ERROR | [ca5938997c8d7888ad43432549115297] | app.model_runtime.load_balancing.base:_log_failure:206 - [LoadBalancer] 端点调用失败 deployment_id=178331601741025280 endpoint_id=178331601761996800 base_url=https://litellm.xxxx.xyz/v1 endpoint_key=openai/completions duration_ms=65556.16 attempts=1 error_type=ModelProviderError error_message=peer closed connection without sending complete message body (incomplete chunked read) 2026-05-21 21:44:55.251 | INFO | [ca5938997c8d7888ad43432549115297] | app.agentic.agent.agent_runner:_construct_run_output_from_stream_context:390 - [Cancelled Run] Constructing RunOutput from StreamContext, run_id=23cbfe1f-fbc8-4b3c-93b2-09d9038fa9b4 2026-05-21 21:44:55.251 | WARNING | [ca5938997c8d7888ad43432549115297] | app.agentic.agent.agent_runner:_append_missing_failed_tool_calls:476 - 工具调用参数 JSON 解析失败, tool_call_id=tooluse_1EQ5EjyGU…

GitHub resolver

Import GitHub neighbors on demand. Results are saved as system ingests.

Refresh page
vote history (2 events)
#0 of 0 · 31d18h59m5s ago — entered · #import:https:::github.com:berriai:litellm post #1892
The left issue is harder because it likely requires cross-layer investigation of long-running request lifecycles, transport timeouts, cancellation, and streaming behavior, with difficult reproduction and broader regression testing. The right issue appears localized to defensive resource-cleanup handling with a straightforward targeted test.
#0 of 0 · 31d18h46m1s ago — current · #import:https:::github.com:berriai:litellm post #2118
Diagnosing an intermittent transport failure requires reproducing timing-dependent behavior, tracing streaming and timeout handling across deployment layers, and validating a robust fix; the other requires a narrowly scoped data correction with straightforward regression coverage.
discussed in #import:https:::github.com:berriai:litellm

ranked child groups

no voted pairs yet in this scope

cli
src
spread
search