Assistants API runs taking 60–90s in long sessions — polling timeout fix & seeking better approach

API
, ,
1

Problem

We’re using the OpenAI Assistants API (Node.js) with a polling loop to wait for run completion. In production, we noticed that some runs — particularly in longer sessions with many back-and-forth messages — were taking 60–90 seconds to complete.

Our backend was configured with a 60-second polling timeout, which caused a run_timeout error and returned a 500 to the user even though the run eventually completed successfully on OpenAI’s side.

What we tried

We increased our polling timeout from 60s to 150s using an environment variable:

const TIMEOUT_MS = parseInt(process.env.OPENAI_RUN_TIMEOUT_MS) || 150000;

while (!TERMINAL.includes(runStatus.status)) {
    if (Date.now() - startTime >= TIMEOUT_MS) throw new Error('run_timeout');
    await new Promise(r => setTimeout(r, 1000));
    runStatus = await openai.beta.threads.runs.retrieve(threadId, run.id);
}

This solved the immediate issue. But we feel polling every 1 second for up to 150 seconds is not ideal.

Question

  1. Is runs.stream() a better long-term approach to avoid timeout issues entirely?
  2. What causes runs to take so much longer in sessions with more messages?
  3. Any production recommendations for handling slow runs gracefully?

Thanks

2

Hi and welcome to the community!

Since you’re looking for a long-term solution, it’s worth noting that the Assistants API is scheduled to be sunset in August. I’ve linked last year’s announcement as well as the migration guide below.

If you run into any issues during the migration, feel free to ask here in the community. We’re happy to help.