Exceptions and limits¶

All SDK exceptions derive from GradientDynamicsError.

exception gradientdynamics.exceptions.GradientDynamicsError¶: Base exception raised by the gradientdynamics package.

exception gradientdynamics.exceptions.AuthenticationError¶: Credentials are missing, expired or not permitted to access the resource.

exception gradientdynamics.exceptions.ValidationError¶

Configuration validation failed before compute allocation.

errors: Sequence[FieldError]¶: Structured field paths, messages and rejected values.

exception gradientdynamics.exceptions.NotFoundError¶: The requested resource or completed asset does not exist in the authenticated organisation.

exception gradientdynamics.exceptions.ResourceNotReadyError¶: A result was requested before its job completed.

exception gradientdynamics.exceptions.CapacityError¶

Current meshing or GPU capacity is full.

retry_after: float | None¶: Suggested delay in seconds before another request.

exception gradientdynamics.exceptions.JobFailedError¶

Remote execution reached a terminal failure.

stage: str | None¶

job_id: str¶

exception gradientdynamics.exceptions.ServiceUnavailableError¶: The service cannot safely execute the operation at present.

Handling temporary failures¶

import time

from gradientdynamics.exceptions import CapacityError

try:
    job = simulation.run(idempotency_key="road-car-baseline-v1")
except CapacityError as exc:
    time.sleep(exc.retry_after or 30.0)
    job = simulation.run(idempotency_key="road-car-baseline-v1")

Retry temporary network, capacity and availability failures with bounded backoff. Do not automatically retry schema, topology or boundary-name errors. Read the original job before retrying after a lost response, and reuse the same idempotency key only for the same intended operation.

Organisation limits¶

Meshing concurrency, CPU memory, GPU capacity, project quotas, storage and retention are organisation- and environment-specific. The client exposes enabled capabilities and quota summaries through gradientdynamics.Client.organisation; scripts should handle queueing without creating duplicate work.