Skip to main content
Version: master

Remote Update Cancel Test Plan

Covers the device honoring an owner-side cancel while a revision is QUEUED or DOWNLOADING, and the Hub precondition change that makes such a cancel possible in the first place.

1. What is being tested​

  • Hub: PUT /trails/{id}/steps/{rev}/cancel (owner token) is accepted while the step is NEW, QUEUED, DOWNLOADING or INPROGRESS (pantahub-base !415). It sets progress.status to CANCEL.
  • Device: while QUEUED/DOWNLOADING, Pantavisor re-reads its own step from the Hub on every download tick (6 s) and only then sends that tick's download progress PUT, so it never overwrites a cancel it has not read yet. On CANCEL it aborts in-flight object transfers, keeps the partial <sha>.tmp objects, reports CANCEL with status-msg Cancelled as requested by owner, and returns to idle on the previous revision.
  • Non-adherence stays visible: the device progress PUT is unconditional on the Hub, so a device that ignores the cancel overwrites CANCEL with DOWNLOADING/INPROGRESS. The test reads the step back from the Hub after the device went idle to prove that did not happen. The same unconditional PUT means a cancel landing in the short window between the device's poll and its progress PUT is overwritten; the owner (and the test) re-issues it.

Out of scope: a cancel (or the older wontgo) set on the Hub while INPROGRESS/ TESTING is never read by the device; it only stops a device that lost track of the step from retrying it, which cannot be driven from the appengine harness.

2. Forcing a wide cancel window​

Same recipe as download-resume-on-timeout: a 100 MB /dev/urandom object and PH_UPDATER_DOWNLOAD_RATE_LIMIT=1048576 give ~100 s of DOWNLOADING. The cancel is only sent once >5 MiB of the partial object is on disk, so the abort lands mid-transfer with real bytes to keep.

3. Test: remote/lifecycle/download-cancel-mid-transfer​

Steps and discriminators printed by resources/test:

  1. post the revision, wait for DOWNLOADING → partial-object-on-disk-before-cancel: 1
  2. owner cancel via curl with the bearer from $HOME/.pvr/auth.json → hub-cancel-http-code: 200 (a 400 here means the Hub still has the NEW-only precondition)
  3. pvcontrol steps show-progress $rev reaches CANCEL; if the Hub shows DOWNLOADING again after 15 s the cancel was overwritten by the progress PUT of the same tick and is re-issued (up to 4 attempts, logged to stderr)
  4. pantahub.state back to idle, pantavisor.status READY, pantavisor.revision unchanged → still-on-previous-revision: 1
  5. partial size after cancel >= size before → partial-object-kept: 1
  6. after a 10 s grace for any in-flight PUT, GET /trails/{id}/steps/{rev} as owner → hub-step-status: CANCEL

4. Manual variant on a real device​

pvr post -m "big" https://api.pantahub.com/trails/<device-id>
# once `pvr device status <nick>` shows DOWNLOADING:
curl -X PUT -H "Authorization: Bearer $(jq -r 'first(.tokens[])["access-token"]' ~/.pvr/auth.json)" \
    https://api.pantahub.com/trails/<device-id>/steps/<rev>/cancel
pvr device status <nick>     # expect CANCEL, then the device back at the old rev
ls /storage/objects/*.tmp    # partial object still there

5. Status​

  • Golden output is hand-written: it can only be regenerated once !415 is deployed on api.pantahub.com and the device image carries feature/remote-update-cancel.