agent-substrate / agent-substrate/env
Actor deletion stuck due to lacking permissions
- 主要語言
- Go
- 星號
- 2
- 分支
- 2
- 平均合併
- 21 小時 1 分鐘
- 30 天內合併 PR
- 28
描述
Deleting an actor fails when `ate-api-server` attempts to clean up snapshots from GCS:
```
rpc error: code = PermissionDenied desc = Caller does not have storage.objects.list access to the Google Cloud Storage bucket. Permission 'storage.objects.list' denied on resource '//storage.googleapis.com/projects/_/buckets/'
```
Because snapshot cleanup fails, the actor remains permanently stuck in `ACTOR_STATE_DELETING`.
#### **Cause**
On GKE, default node service accounts only have `read_only` storage scopes. The `ate-system/ate-api-server` Kubernetes service account is not bound to a Google Service Account (GSA) via Workload Identity, leaving it without permissions to list or delete objects in the snapshots bucket.
#### **Suggestions**
1. **Document Workload Identity setup:** Specify that `ate-api-server` requires `roles/storage.objectAdmin` on the snapshot bucket via Workload Identity (`ate-system/ate-api-server`).
2. **Handle cleanup failures gracefully:** Allow actor deletion to complete (or report a warning) instead of trapping the actor in `ACTOR_STATE_DELETING` forever when bucket cleanup errors occur.
貢獻指南
這個儲存庫沒有索引到貢獻指南
評估
這個 Issue 還沒有評估資料。