OpenAI and Speech-to-Text blocks hang when running STT workflow in `dify-plugin-daemon` on Kubernetes

Environment

• Deployment: Kubernetes
• Runtime Version: 0.0.7-local

Problem Description

We are encountering an issue where OpenAI blocks become unresponsive when a workflow involving the Speech-to-Text (STT) block is executed. Additionally, the workflow containing the STT block itself hangs indefinitely and does not return any results.

Observations

• When the STT workflow is running, all concurrent or subsequent OpenAI block executions appear to be blocked.
• Pod metrics indicate that the relevant process within the dify-plugin-daemon pod consumes 1 full CPU core during this state.
• This behavior leads us to suspect that the daemon might be single-threaded or facing a bottleneck related to concurrent processing, particularly when handling STT operations.

Questions

1. Is there a known issue or potential reason why the Speech-to-Text (STT) block would cause other OpenAI blocks to hang and fail to complete itself?
2. What are the recommended practices or configurations for deploying and managing the dify-plugin-daemon on Kubernetes, potentially including considerations for scaling or concurrency to handle workloads like STT?

Thank you for your support.

Begin with the Kubernetes deployment and the STT workflow described in the report, then inspect how concurrent OpenAI and STT executions are handled by dify-plugin-daemon. Done means identifying the hang and CPU bottleneck and documenting a reproducible fix or deployment or concurrency recommendation.