Skip to content

VLStream Cloud v1.2.6

Choose a tag to compare

@github-actions github-actions released this 30 Sep 14:42
· 74 commits to main since this release
Immutable release. Only release title and notes can be modified.

VLStream Cloud v1.2.6

Built from origin/main at dadb04a1. This release adds persistent training
artifacts and cloud-compute workflows while retaining the public v1.2.5 database
history for production upgrades.

Highlights

  • Persist complete annotation datasets and verified model artifacts in MinIO,
    with immutable dataset references, checksum validation, retry states, and
    compatibility with existing SSH-based training data.
  • Add tenant-scoped AutoDL compute nodes, encrypted credentials, background
    training lifecycle management, cancellation safeguards, and model/artifact
    return workflows.
  • Improve algorithm/model import and annotation type handling, plus prevent
    real-time previews for offline devices.
  • Extend IPC tunnel session handling. The backend control plane is enabled by
    default; the tunnel runtime remains an optional Compose profile.
  • Keep all production secrets environment-backed and default GPU/SSH endpoints
    unset in the released runtime configuration.

Compatibility and database upgrades

  • WVP is deployed separately using
    VLStream WVP Lite v1.0.8,
    image ghcr.io/oortcloudgroup/vlstream-cloud-lite:1.0.8. WVP remains the
    device center and owns ZLMediaKit.
  • Backend and frontend images:
    ghcr.io/oortcloudgroup/vlstream-backend:1.2.6 and
    ghcr.io/oortcloudgroup/vlstream-frontend:1.2.6.
  • Production upgrades continue the public v1.2.5 Flyway history and apply
    V1_2_0_024 through V1_2_0_028. prod selects
    db/migration/release; dev and local retain the separate mainline history.
    Never change migration locations for a database that already has Flyway
    history.
  • Fresh installations import the sanitized VLS schema and the reviewed WVP
    v1.0.8 bootstrap schema. The VLS Compose project does not run WVP or
    ZLMediaKit services.

Cloud training and AutoDL behavior depends on tenant configuration, network
access, and the target GPU runtime. The included unit, integration, and release
build checks do not establish model-quality or compatibility for every GPU
type.

Full Changelog: https://github.com/OortCloudGroup/VLStream-Cloud/commits/v1.2.6