Deployment Health Visibility in AI Catalog

The JFrog ML Registry surfaces real-time deployment health directly on the model page, so you can identify and act on deployment failures without leaving the JFrog Platform.

When a deployment never becomes available, the platform may remove it automatically. For that behavior and status values such as Initialization Failed, see Automatic Undeployment.

Deployment Status Banners

When a deployment encounters a problem, a status banner appears at the top of the model page. The following table summarizes the two severity levels.

Severity	When It Appears
Critical deployment failure	The deployment completely failed to start.
Degraded deployment state	Some replicas started successfully, others did not.

Critical Deployment Failure

The platform shows this banner when a deployment has completely failed to start.

Example message:

Critical deployment failure: Deployment of version v2.0.4 failed initialization. Reason: resource constraints (out of memory).

The following table lists actions on the banner.

Action	Description
View logs	Opens the Runtime tab with logs for the failed deployment.
Retry deployment	Re-triggers the deployment with the same configuration.

Degraded Deployment State

The platform shows this banner when a deployment is partially active.

Example message:

Degraded deployment state: Deployment of version v2.0.4 is partially active. Reason: 3/5 replicas failed to initialize due to resource constraints.

The following table lists actions on the banner.

Action	Description
View logs	Opens runtime logs filtered to the affected replicas.
Retry deployment	Attempts to recover the failed replicas.

Runtime Logs

The Runtime tab on the model page provides raw log output from the model's running instances.

Controls

The Runtime tab includes the following controls.

Control	Description
Show	Time range filter (for example, Last hour, Last 24 hours).
Display	Column selection: Timestamps, Replica ID.
Search	Free-text search across log messages.
Follow	Toggle to tail logs in real time.

Log Table

Each row in the log table represents a single log entry.

Column	Description
Instance ID	The replica that produced the log entry (for example, `b7ad8000-a...`).
Timestamp (UTC)	When the entry was recorded.
Message	Log level and content (for example, `INFO –`).

Each row includes an expand control (>) that opens the full log message.

Common Failure Reasons

The following table lists common failure reasons shown in status banners and logs.

Reason	Meaning	Recommended Action
Resource constraints (out of memory)	The instance ran out of video RAM or RAM during model load.	Choose a larger instance type from Instance Sizes & ML Credits or enable quantization.
Resource constraints (N/M replicas failed)	The cluster has insufficient capacity for the requested replica count.	Reduce the replica count or retry at a later time.

Frequently Asked Questions

This section provides answers to frequently asked questions.

plusFAQs

Q: How do I see why a model deployment failed on the model page?

A: Open the model in the JFrog ML Registry and read the status banner at the top of the page. Select View logs on the banner to open the Runtime logs tab for that deployment.

Q: What is the difference between critical deployment failure and degraded deployment state?

A: Critical deployment failure means the deployment did not start. Degraded deployment state means some replicas are running and others failed to initialize. For more information, see Deployment Status Banners.

Q: What should I do when failure reasons mention resource constraints?

A: For out-of-memory errors, choose a larger instance type or enable quantization. For replica capacity errors, reduce the replica count or retry later. For more information, see Common Failure Reasons.

Q: What do I need before I can review runtime logs for a deployment?

A: You need a model that has been deployed or attempted on the cluster so instances exist or failed during startup. If no deployment is active, deploy the model from Deploy Model Packages.

Q: Can I retry a failed deployment from the model page?

A: Yes. Select Retry deployment on the status banner to re-trigger the same configuration. If the platform removed the deployment automatically, deploy again after fixing the root cause. For more information, see Automatic Undeployment.