[jasonwang-runpod]

Problem

vLLM starts before the SDK, so unrecognized startup failures stay in worker logs instead of reaching queued requests.

Solution

Run startup under SDK prestart supervision while keeping jobs queued until startup settles. Preserve existing retries and startup_error responses for known configuration failures. Unrecognized failures now fail one queued job with prestart_failed, include vLLM output, and exit so the platform can restart the worker.

Requires runpod 2.0.0, which main already pins.

Testing

• 184 worker tests and lint passed with runpod==2.0.0 from PyPI.
• Live A40 / vLLM 0.32.0 / runpod==2.0.0: prompt, chat, and streaming jobs queued during a cold start, were claimed after startup finished, and completed within a 20-second execution limit.
• Earlier live run against SDK 2.0 main also covered known and unknown startup failure reporting. Existing local-input and realtime SDK limitations remain.

[promptless]

Promptless documentation updates

• Document vLLM worker startup error handling in Serverless troubleshooting explains the prestart_failed error that one queued job receives when vLLM fails to start for an unrecognized reason.