Skip to content

v1.9.0

Latest

Choose a tag to compare

@legraina legraina released this 19 Nov 03:47
815923c

New images deploy themselves (#316)

The cluster now picks up a newly published rmni/<service>:main image within 5 minutes, without minikube-helper.sh -r. The new image-update CronJob (deployment/image-update.yml):

  • rolls out server, socketio and webapp when their :main tag points at another image on Docker Hub than the one running, once per image;
  • wakes an executor when rmni/executor:main changes, so that its large image is already on the node when a task comes;
  • posts each rollout to the sentinel's Slack channel.

Its role may only patch those three Deployments, list pods, read the executor's ScaledJob and annotate its own CronJob. Docker Hub or the server out of reach is retried at the next run; without an admin-api-key the executor is left alone. To see what it would do: deployment/image-update.sh --dry-run --executor job-executor server socketio webapp.

A v* release tag publishes :<version> only: the cluster deploys a release once the "Publish All Docker images" workflow has run on main.

PDF viewer 32 (#319)

The grading screen moves to ngx-extended-pdf-viewer 32.0.2. Its default bundle (pdf.js 6.5) now has the eraser and undo/redo, and the fixes we sent upstream (stephanrauh/pdf.js #16):

  • the eraser works on rotated scanned pages;
  • an erased drawing keeps its rotation;
  • a restored drawing is one undo step;
  • changing copy while pages render no longer throws.

The bleeding-edge bundle and the postinstall patch of the viewer are gone. On a scanned copy shown at 270°, the eraser cuts under the pointer, undo/redo work, and drawings come back as saved.

Phone layout (#317)

  • In the cards of the dashboard, task list and admin, the label stays on the left and the content together on the right, instead of spread over the line.
  • The guide's chapter dropdown names the open chapter even for an admin acting as a user.
  • A shared or read-only dashboard no longer ends each card with an empty line.
  • "No data matching the filter" and "Aucun compte" are full-width, centered lines.

Fixes

  • Health sentinel (#318): a CronJob created less than its success window ago (48h by default) is no longer reported as having no successful run. That was the warning sent right after image-update was deployed. A failing run is still reported.

After deploy

kubectl apply -k deployment/
kubectl create job image-update-now --from=cronjob/image-update && kubectl logs -f job/image-update-now

Full changelog: v1.8.5...v1.9.0