Concurrency Issue: 'Server Busy' Errors After Updating Ollama

[mynhinguyentruong]

What is the issue?

Issue Summary:
Yesterday, I conducted a test where I spun up 30 concurrent goroutines and sent them as POST requests to Ollama locally. The process worked smoothly, and I received responses within approximately 2 minutes.

go ProcessPromptWithOllama() x30 times

Problem Description:
However, after updating Ollama today, I encountered the following errors:

\"error\":\"server busy, please try again. maximum pending requests exceeded\"}
\"error\":\"unexpected server status: llm busy - no slots available\"}
Reproducibility:
The issue is consistently reproducible after the Ollama update. It occurs regardless of the specific endpoint or payload used in the POST requests.

Expected Behavior:
I expected the updated Ollama to handle the concurrent requests as efficiently as it did before the update, without encountering any server overload issues.

OS

macOS

GPU

Apple

CPU

Apple

Ollama version

0.1.33

[jmorganca]

Hi @mynhinguyentruong, I'm so sorry about this issue – we'll get this fixed so more requests can be queued

[dhiltgen]

Dup of #4124