Batch API Job stuck in PENDING state for over 26 hours

Gemini API
1

Hi everyone,

I am currently experiencing an issue with the Gemini Batch API where my job has been stuck in the PENDING state for over 26 hours.

Typically, my batch jobs are completed within 2 to 3 hours. Therefore, this delay is quite unusual and concerning, especially since it has exceeded the 24-hour target.

Details of the stuck job:

  • Model: gemini-3-pro-preview

  • Job Creation Time: 2026-01-07T03:35:56Z

  • Job Resource Name (ID): batches/fl2n2q8e3kxgj3w48ikbx33y09aop34q6nxw

  • Region: asia-northeast3 (Seoul)

I have checked my quota and I am well within the limits. Is there currently a known outage or a delay with the Batch API queue in this region?

Any advice or updates from the Google team would be appreciated.

Thanks.

6 Likes
3

Hi @gmsim,

Welcome to the Forum!

Thanks for your report. We’ve pushed a fix that should resolve the problem. Please check and let us know if the problem persists.

5

gampang bin gampang

PENDING Status for more than 26 hours on a Batch API is not uncommon, especially since the auto-completion target is supposed to be 24 hours. However, based on the technical documentation and similar reports in the developer community, there are several systematic reasons why this happens.
Here are the details of the problem and the step-by-step solution for you:

  1. Understanding Queue Time Limits
    It is important to distinguish between Queue Time (when the status is still PENDING) and Processing Time (when the status is RUNNING).
  • Queue Time (PENDING): jobs can be held in line for up to 72 hours if demand in the region (in this case asia-northeast3) is very high. If within 72 hours has not started processing, the job will automatically expire (expired).
  • Processing Time (RUNNING): once the status changes to RUNNING, Google targets completion within 24 hours.
  1. Why Is Asia-Northeast3 (Seoul) Experiencing Obstacles?
    The Asia Region sometimes experiences higher computing load spikes for preview models such as the gemini-3-pro-preview. A history of disruptions to Vertex AI services in Asia (such as the December 2024 incident) shows that AI infrastructure capacity in the region is often full, leading to long PENDING status.
  2. Solutions you can do now
    A. Cancel and switch regions (top recommendation)
    If your job is urgent, don’t wait 72 hours. Cancel the current job and resubmit it to a region with greater capacity.
  • Alternate Region: use us-central1 (Iowa) or us-east4 (Northern Virginia). U.S. regions typically have much larger resource pools for the latest models.
    B. Use CLI to check Error details
    Sometimes the Dashboard UI does not display a specific error message. Use this command to view status events:
    fl2n2q8e3kxgj3w48ikbx33y09aop34q6nxw
    – location=asia-northeast3

Note the statusEvents section. If there is a message like “Resources unavailable” or “Deadline expired”, it confirms that the region is running out of capacity.
C. Use The Stable Version Model
You are using gemini-3-pro-preview. If possible, try using the Stable version (if already available for Series 3) or gemini-1.5-pro which usually has a more stable resource allocation priority than the preview version.
4. Comparison Of Job Status
| Status | Meaning / Maximum Limit |
|—|—|—|
/ PENDING / waiting in line | 72 hours |
| RUNNING | being processed by model | 24 hours |
| SUCCEEDED / perfect finish | - |
| EXPIRED / failed to start due to full queue | - |