Skip to content

Gateway hangs indefinitely — no response, no timeout error (Lambda-backed target, streaming disabled) #1388

@wenhaog7

Description

@wenhaog7

Description

AgentCore Gateway accepted an HTTP POST (tools/call) but never returned any HTTP response — no 200, no error, no timeout. The calling client has been hung for 50+ minutes waiting for a response. Response streaming is disabled (default). The gateway target is Lambda-backed (no container cold start involved). Other requests to the same gateway in the same time window completed successfully.

Steps to Reproduce

  1. Agent runtime sends concurrent tool calls through ACG Gateway (Lambda-backed target)
  2. Gateway accepts the HTTP POST request
  3. Gateway never returns any HTTP response
  4. Client httpx connection hangs indefinitely — no timeout error surfaced

This is intermittent and appears under concurrent load (3+ executions sharing the same agent container, all calling the same gateway simultaneously).

Expected Behavior

Gateway should return an HTTP response within its configured timeout (40s based on our observations), either with the tool result or a timeout error.

Actual Behavior

Gateway accepted the request and never responded. No HTTP response at all — not a timeout error, not a 5xx, nothing. The execution has been stuck for 50+ minutes with the httpx client waiting on an open connection.

CLI Version

N/A — this is a Gateway service-side issue, not CLI.

Operating System

Linux

Additional Context

  • Account: 272994241044
  • Region: us-west-2
  • Gateway: releaseautopilot-gateway-prod-internal-p03wrf71h7 (Lambda-backed)
  • Timestamp: 2026-05-27 01:55:03 UTC (request sent, never received response)
  • Response streaming: disabled (not configured, using default)
  • Evidence from CloudWatch logs: The agent container continued processing other executions successfully after this request hung. The last log line for the affected trace (6a164e821d6483d72bc4ada63789a8d9) is the outbound HTTP POST to the gateway URL — then zero activity for 50+ minutes.
  • Related: We also observe a pattern where the same gateway returns "Timed out waiting for tool execution after 900 seconds" errors after exactly ~40s wall-clock time (100 occurrences over 3 days). This hung case appears to be the same delivery failure but without the timeout error being surfaced.

Metadata

Metadata

Assignees

No one assigned

    Labels

    bugSomething isn't working

    Type

    No type
    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions