[QTI] Load assessment items from the server while the cache may be stale

Offen
#6,340 0 Kommentare 1 Reaktion 0 zugewiesene Personen Auf GitHub ansehen

Maintainer antworten meist innerhalb von 1 Tag

Dieses Issue hat noch niemand übernommen.

Bewertung

Schwierigkeit
4/5
Geschätzter Aufwand
1-2 Tage
Anfängerfreundlichkeit
25/100
Issue-Typ
Bug
Klarheit
Klar beschrieben
Aktivitätsstatus
Aktiv
Tech-Stack
javascript
Bereich
backend, database

Rechercherichtung

Start with initializeDB() in shared/data/index.js and the startup flow in shared/app.js; trace how IndexedDB is opened before data is read. Check the EditModal loading and self-healing path, then add Jest coverage for cleanup errors, preserved changes, and failed loads. Done when the acceptance criteria hold and syncing still starts if cleanup fails.

Vom Indexierungsmodell aus dem Issue-Text verfasst.

Beschreibung

DEV: frontend

Overview

After the QTI release, the cache can show users on the new assets the wrong content:

  • Items cached before the release are legacy, and the new editor shows them as unsupported. The background refresh never reaches an open exercise, so they stay that way until it's reopened. Users who see their questions as unsupported may think they're broken, and delete or redo them.
  • Items edited in an old tab can be cached as QTI, with raw_data from before the edit, while the database row is still legacy and holds the edit. The new editor shows the old content, and the next edit there writes it back over the old tab's edit.

Until #6341, load an exercise's items from the server instead of the cache the first time it's opened after a page load, and whenever any cached item is neither QTI nor Perseus. If that fetch fails, show the cached items, as today.

This also fixes #6148, where exercises created with the old editor show as incomplete on unstable until they're opened a second time.

Part of #6337. Must be in unstable before the QTI release is cut.

Complexity: Low
Target branch: unstable

Context
  • loadAssessmentItems (channelEdit/vuex/assessmentItem/actions.js) is the only read of assessment items. It calls AssessmentItem.where and commits the result to the store.
  • Resource.where returns cached rows and refreshes them in the background through conditionalFetch. It only waits for the server when nothing is cached.
  • The refresh writes the fetched rows to IndexedDB, overwriting the cached ones and reapplying unsynced changes (setData). Updates from IndexedDB never reach the store, though: the plugin emits the key under "contentnode,assessment_id", and UPDATE_ASSESSMENTITEM_FROM_INDEXEDDB reads id.
  • A tab on the old assets that opens an exercise with legacy items cached doesn't reload, because #6338 only reloads on QTI items. Its background refresh writes QTI rows to the shared IndexedDB. An edit made then syncs, because the database row is still legacy, while the cached row keeps the raw_data from before the edit. The tab reloads into the new assets on the next open.
  • Edits from old tabs, and from collaborators on the old assets, reach the cache as changes to legacy fields, which leave a QTI row's raw_data as it was.
  • The store keeps an exercise's items across opens until the page reloads.
  • Unsynced changes sync once no change has been made for 2 seconds.
  • Type values: 'QTI' and 'perseus_question'. Anything else is legacy.
  • consolidate() leaves items it can't convert as legacy.
The Change
  • In loadAssessmentItems, wait for the server's items and commit those instead of the cached ones when an exercise being loaded hasn't been loaded from the server since the page loaded, or when any cached item is neither QTI nor perseus_question.
  • If that fetch fails, commit the cached items.
  • Keep the checks self-contained, so #6341 can remove them.
Decisions

Choices made while planning this, and why.

  • Why wait in loadAssessmentItems, rather than clearing the cached items on startup? An earlier plan cleared the whole assessment item table on every startup while any row was legacy. Because updates and deletes from IndexedDB never reach the store, other open tabs would keep showing the cleared questions, and edits made there would record no change and be lost. Waiting deletes nothing, only affects the exercise being opened, and needs no startup scan.
  • Why also wait on the first open after a page load? It covers cached QTI rows whose raw_data is older than the database, however they got there: an edit made in an old tab before it reloaded, or a change from an old tab or a collaborator on the old assets. A tab that reloads into the new assets starts a new page load, so its first open of each exercise reads the server. The cost is one round trip per exercise per page load, until #6341. An earlier plan only waited when cached items were legacy, which missed these.
  • Why wait for all of an exercise's items, and not only the legacy ones? Showing the QTI items first and the rest later would show an exercise with only some of its questions.
  • Why fall back to the cached items when the fetch fails? They show as they do today, rather than as an empty exercise.
  • Why not clear the cache with a Dexie schema version bump? A version bump runs once per browser, but cached legacy items come back from tabs still on the old assets, which share the same IndexedDB. Dexie also closes the database in every open tab on a version change, which would break those tabs.
  • Why not fix the store bug instead? Updates from IndexedDB never reach the store, because the plugin emits the key under "contentnode,assessment_id" while UPDATE_ASSESSMENTITEM_FROM_INDEXEDDB reads id. That bug predates QTI, affects other tables, and is tracked separately. Even with it fixed, the cached items would show until the refresh lands, and an edit made before then would still start from them, so the wait is still needed.
Known gaps
  • The first open of each exercise after a page load, and any open with legacy items cached, waits for the server, as opening one with nothing cached does today.
  • Exercises with items the server can't convert wait on every open, because those items stay legacy.
  • When the fetch fails, the cached items show until a later open succeeds: legacy ones as unsupported, and stale ones with their old content.
  • Reopening an exercise later in the same page load uses the cache. If an old tab or a collaborator on the old assets has edited a question since, its raw_data is stale, and an edit in the new editor writes the old content back over theirs.
  • A question created but not yet synced isn't among the server's items. If the page reloaded in between, it's missing from the editor until the next open, and if it was the exercise's only question, the completeness check marks the exercise incomplete.
Acceptance Criteria
  • The first open of an exercise after a page load shows the server's items, even when every cached item is QTI or Perseus
  • Opening an exercise whose cached items include a legacy one shows the server's items, without first showing any as unsupported or showing only some of them
  • Reopening an exercise in the same page load, with every cached item QTI or Perseus, shows the cached items and refreshes them in the background, as today
  • If the fetch fails, the cached items are shown
  • No rows are deleted and no changes are recorded
  • #6148 no longer reproduces
  • Jest tests cover the first-open, legacy, reopen and failed-fetch cases

AI usage

I outlined the approach and made the decisions above. Claude traced the read path on unstable and drafted the issue, which I then checked and corrected for accuracy.

Vorherrschende Sprache
Python
Sterne
212
Forks
321
Ø Merge
1 T. 7 Std.
Gemergte PRs (30 T.)
14

Entwicklungsumgebung

Erste Schritte

  1. Lesen Sie das ganze Issue und danach den Beitragsleitfaden des Projekts.
  2. Schreiben Sie ins Issue, dass Sie es übernehmen — das erspart doppelte Arbeit.
  3. Forken Sie das Repository und arbeiten Sie in einem Branch.
  4. Öffnen Sie einen Pull Request, der die Issue-Nummer nennt.

Mehr aus learningequality/studio

Alle Issues in learningequality/studio

Ähnliche Issues

Weitere Issues zu Python

Neue Issues direkt in Ihr Postfach

Eine kurze Übersicht über anfängerfreundliche GitHub-Issues.