divine-upload-server

Rust upload data plane for Divine Blossom media uploads.

Scope

divine-upload-server owns:

• resumable upload session lifecycle
• direct chunk uploads to GCS
• upload finalization and canonical blob writes
• audit-log ingress from the Fastly Blossom control plane
• thumbnail and media follow-up hooks

It does not own the Blossom control plane. divine-blossom remains the Fastly-facing service that answers client control-plane requests such as HEAD /upload, validates Blossom auth, and proxies short init and complete calls to this service.

In the approved Divine resumable flow:

• https://media.divine.video is the client-facing control plane
• https://upload.divine.video is the opaque resumable session data plane
• uploadUrl values returned to clients are server-issued session URLs and must be treated as opaque

Runtime Configuration

The service reads configuration from environment variables:

• GCS_BUCKET
• CDN_BASE_URL
• UPLOAD_BASE_URL
• PORT
• MIGRATION_NSEC when migration auth is required
• TRANSCODER_URL
• TRANSCRIBER_URL
• RESUMABLE_SESSION_TTL_SECS
• RESUMABLE_CHUNK_SIZE
• RESUMABLE_MAX_REQUEST_BODY_SIZE

RESUMABLE_CHUNK_SIZE is capped to RESUMABLE_MAX_REQUEST_BODY_SIZE before it is advertised in /upload/init, so the published chunk contract cannot exceed the actual request-body limit on upload.divine.video. The default route limit is 1048576 bytes, matching the current production nginx ingress behavior. UPLOAD_ROUTE_MAX_BODY_SIZE is also accepted as a compatibility alias.

Development

Run tests:

cargo test
python3 -m unittest discover -s tests -p 'test_*.py'

Run locally:

cargo run

The default local port is 8080.

Deployment

Phase 1 targets production GKE behind https://upload.divine.video.

The current media follow-up path still points at the existing transcoder endpoint through TRANSCODER_URL. GPU-backed transcoder migration is a separate phase.

The repository CI workflow builds and tests on pull requests, then publishes divine-upload-server to us-central1-docker.pkg.dev/dv-platform-prod/containers-production on pushes to main.

export-video-upload-hashes.py queries both the legacy Cloud Run audit logs and the GKE k8s_container audit logs so historic exports continue to work across the migration.