Skip to content

ACI containers crash on startup: missing AZURE_UAMI_RESOURCE_ID env var + failed containers leak quota #355

Description

@jathavaan

Problem

Two bugs cause every ACI benchmark run to fail:

  1. Missing env var: AZURE_UAMI_RESOURCE_ID is required by Config at import time (_require_env), but _create_container_instance in main.py never forwards it to the container's --environment-variables. Containers crash immediately on startup with OSError: Required environment variable 'AZURE_UAMI_RESOURCE_ID' is not set.

  2. Quota cascade: When a container fails, _check_container_state raises and _delete_container_instance on the next line never executes. Failed containers keep holding ACI cores, and subsequent batches hit ContainerGroupQuotaReached (Usage: 42, Limit: 40).

Fix

  • Forward AZURE_UAMI_RESOURCE_ID to containers via --environment-variables
  • Wrap _check_container_state in try/finally so cleanup always runs

Metadata

Metadata

Assignees

Labels

bugSomething isn't working

Type

No type

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions