Errors

SDK errors expose code, message and detail. For support, keep the error, deployment ID and request ID.

Error	Next action
`Unauthorized`	Run `servo login`, then `servo whoami`
`RobotRoleRequired`	Ask a robot admin for the role named in the error
`GoogleIdentityRequired`	Link your Google account for group access
`DirectoryUnavailable`	Retry Google group lookup or use a direct grant
`PaymentRequired`	Check organization billing before allocating capacity
`NotFound`	Check the resource ID and your account’s access
`ModelNotServable`, `model_not_checkpoint_native`	Check checkpoint files and supported runtime
`ValidationError`, `ObservationInvalid`	Match the named input’s inspected shape and dtype; remove extra keys
`ServerNotReady`	Run `servo deployment wait ID`
`ServingDeadlineExceeded`	Inspect deployment status and request logs
`ServingUnavailable`, `ServingCapacityExceeded`	Check deployment status and available capacity
`StaleGeneration`	Open a new session after the intended deployment change
`PayloadTooLarge`	Check image dimensions and extra channels

A timeout does not stop allocation. Use servo deployment show ID before retrying or stopping. A transport failure does not prove whether inference ran. See logs and telemetry.

Catch a prediction error

Use this helper with the session and inputs created in the quickstart:

predict_with_errors.py

import servo


def predict_with_errors(session, inputs):
    try:
        return session.predict(inputs=inputs)
    except (servo.ValidationError, servo.ObservationInvalid) as error:
        print("Fix the model input:", error.message)
        print("Details:", error.detail)
        raise
    except servo.ServingDeadlineExceeded as error:
        print("Prediction deadline exceeded:", error.message)
        raise

Get started

Guides

Reference

Catch a prediction error

​Catch a prediction error

Catch a prediction error