feat: backups for container persistent volumes

[alongubkin]

Tracks ALIEN-1232.

Containers with .persistentStorage() get one block volume per replica. Before this change nothing snapshotted those volumes, and deleting the container deleted them immediately. This branch adds the types, cloud clients, permissions and CLI for volume backups and restore on AWS, GCP and Azure.

Config

• persistentStorage(size, { backups }). Backups are on by default: every 24 hours, kept 7 days. Tune them with intervalHours (1, 2, 4, 6, 8, 12 or 24) and retentionDays (at most 365), or turn them off with backups: false.
• A compile-time preflight keeps the schedule within what AWS Data Lifecycle Manager, Compute Engine snapshot schedules and Azure Disk Backup all support, including at most 450 snapshots per volume.

Outputs

• ContainerOutputs.volumes: each replica's volume, zone, latest snapshot, and last restore.
• ContainerOutputs.volumeBackups: active, disabled or setupRequired, with a message.

Restore

• DeploymentConfig.volumeRestores carries restore requests (request ID, resource, ordinal, snapshot) to the controller. The manager's sync route passes them through.
• CLI: alien deployments volumes, alien deployments restore-volume <deployment> --resource <id> --ordinal <n> --snapshot <id>, and alien deployments cancel-volume-restore.

Controllers in this repository

• The types, clients and permissions here are what a container controller for AWS, GCP or Azure uses to schedule snapshots, report them and restore a volume.
• The Kubernetes and local container controllers don't snapshot volumes. They report no volumes and no volumeBackups status. A restore request can only target a volume listed in ContainerOutputs.volumes, so it can't target these controllers.

Cloud clients

• AWS: a Data Lifecycle Manager client, plus EC2 CreateSnapshot, DescribeSnapshots, DeleteSnapshot, CreateTags and DeleteTags. DescribeVolumes now reads attachment status.
• GCP: resource policies (snapshot schedules), add and remove policies on disks, and snapshots.
• Azure: Data Protection (Backup vaults, policies, backup instances, suspend and resume) and managed disk snapshots. The snapshot listing follows a nextLink only on the management endpoint's host, so the bearer token never goes to another host.

Permissions

• container/provision grants the schedule and snapshot calls. AWS grants are bound to the deployment's resource tags, and IAM role grants to the <prefix>-dlm-* name with PassRole limited to dlm.amazonaws.com.
• container/management grants only the status reads.
• Installations set up before this change don't have these grants. The container keeps running, and its outputs report setupRequired until the setup is updated, for example with a CloudFormation stack update or terraform apply. The remote stack management controllers now record a revision of the management permissions they applied, so an Alien direct setup re-applies them when the permission set changes. A running direct install has no command yet that re-runs setup: alien deploy waits on the deployment lock.

Tests

• Core, preflight, permission-set, cloud client and CLI unit tests.
• Results in the sandbox: alien-core 633, alien-preflights 399, AWS clients 91, Azure clients 77, GCP clients 43, alien-permissions 33, remote stack management 15. All passed.
• Scheduling, status, restore and delete were run live on AWS and GCP in a test account. Azure couldn't be run live: the test subscription has no VM sizes available in the regions with the machine image.

Generated files: packages/core/src/generated was regenerated with pnpm generate. The platform TypeScript SDK still needs regenerating with Speakeasy 1.680.11 (pnpm run generate:api-platform-sdk) before merge.

🤖 Generated with Claude Code

[enclave-ai]

Enclave skipped this draft pull request. It will run automatically when you mark the PR as Ready for review.

[alongubkin]

@greptileai review

[greptile-apps]

Confidence Score: 5/5

[High risk] Adds volume backup and restore operations across cloud providers.

The PR appears safe to merge based on this review; no new actionable issue or outstanding previous finding remains.

Summary

The PR adds persistent-volume backup configuration, cloud clients and permissions, restore plumbing, and CLI commands. Since the previous review, it also restricts Azure snapshot pagination to the management endpoint and clarifies which controllers report volumes.

_{Reviews (4) · Last reviewed commit: "fix: check the host of Azure snapshot ne..."}

[alongubkin]

@greptileai review

[alongubkin]

@greptileai review