Instability and slow response

[max-appno]

Date: 2026-03-11

---

Issues Observed

Two types of instability were identified:

1. Job stuck in PENDING indefinitely
   An extraction job was submitted successfully but never progressed. The results endpoint was polled multiple times over 20 minutes, receiving 400 Bad Request on every attempt with
   LlamaCloud reporting the extraction run status as PENDING. The job never completed.

• Job ID: ed729b7d-ef41-4be9-b1be-87519735d4f0
• LlamaCloud extraction run ID: 995d61ff-8b9d-4be1-9c0b-b3d3703282ba
• First poll: 2026-03-11 20:15:44 UTC
• Last observed poll: 2026-03-11 20:35:48 UTC
• Duration stuck: ~20 minutes
• LlamaCloud error: Extraction result not available. Status for extraction run 995d61ff-8b9d-4be1-9c0b-b3d3703282ba: PENDING
• Outcome: Job never completed

2. Slow processing
   On separate occasions, jobs did eventually complete but took significantly longer than the expected baseline. The extraction pipeline appeared to stall before returning results.

---

Root Cause Assessment

The failure appears to be on the LlamaCloud side. The job was submitted successfully and a valid job ID was returned. Polling began shortly after submission, but the underlying
extraction run never advanced from PENDING status despite repeated requests over ~20 minutes. This is consistent with intermittent instability on LlamaCloud's extraction workers.

---

Questions

1. Can you investigate what happened to extraction run 995d61ff-8b9d-4be1-9c0b-b3d3703282ba and why it never completed?
2. Is there a recommended timeout or retry strategy for jobs that get stuck in PENDING?
3. Is there a way to manually trigger a re-run of a stuck extraction job without re-uploading the file?