Workloads with long names never start

[scotwells]

Summary

A workload never gets running instances when its name, combined with the location it deploys to, runs past 63 characters. The user sees no error, and the workload's deployments never report any status.

Any customer who picks a descriptive name can hit this. A general-purpose workload named with 46 characters failed in both Dallas and San Jose in production today, while a unikernel workload in the same project, 6 characters shorter, ran in both.

Acceptance criteria

• A user can create a workload with any name the API accepts and get running instances in every location it targets
• A user whose workload cannot be placed sees the reason in the workload's status

[scotwells]

Evidence from production on 2026-09-25, project project-l6g7z, edge cell us-central-1-alice.

The edge compute-manager fails to create the Instance because it copies the WorkloadDeployment name into a label value, which Kubernetes caps at 63 characters. It logged 132 of these in 75 minutes and never wrote status back to the WorkloadDeployment.

failed executing instance control action: failed to create *v1alpha.Instance:
Instance.compute.datumapis.com "compute-lifecycle-general-purpose-dfw-1b5ba8bb-default-us-central-1-0" is invalid:
metadata.labels: Invalid value: "compute-lifecycle-general-purpose-dfw-1b5ba8bb-default-us-central-1": must be no more than 63 bytes

WorkloadDeployment name	Length	Result
compute-lifecycle-general-purpose-dfw-7b2d10c4-default-us-central-1	67	fails
compute-lifecycle-general-purpose-sjc-b2be1cd3-default-us-west-1	64	fails
compute-lifecycle-unikernel-dfw-801b1204-default-us-central-1	61	runs
compute-lifecycle-unikernel-sjc-37870f03-default-us-west-1	58	runs

Runtime class plays no part. The unikernel names are 4 characters shorter and land under the limit.