While creating deployments on the BentoCloud, I have been seeing some unintuitive behavior regarding the status both through the web GUI and via CLI web responses:
When a deployment has failed and I redeploy, the status changes from “Updating” to “Image Build Failed” before going to “Image Building”. The “Image Build Failed” is transient and seems premature. I don’t see any evidence of the failure in the log before it immediately starts building and succeeds. Since I am trying to verify deployment status automatically, the seemingly false failure is making it difficult to maintain ground truth.
Sometimes my deployments will appear to succeed, but be stuck in the “Image Building” state for hours despite responding 200 to readyz. Attempting accurate automation is difficult when the status does not update in a timely manner to a success state.
The second case is less consistent to reproduce so I don’t have it on hand, but I checked the logs and they seemed like the succeeded like normal and responded to /readyz as well as the inference endpoint and returned valid responses despite being stuck in an “Image Building” state.
You have addressed it? I still see the issue unless I’m misunderstanding what you mean.
My test deployment horizon-contact-classification-preview-2204 has examples of both issues. The long/stale “Image Building” can be seen at timestamp (UTC): 2026-05-21 22:59:57
