-
Notifications
You must be signed in to change notification settings - Fork 27
docs: add ADR for non-nullable tag external_id #658
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Changes from 1 commit
File filter
Filter by extension
Conversations
Jump to
Diff view
Diff view
There are no files selected for viewing
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,101 @@ | ||
| .. _openedx-tagging-adr-0012: | ||
|
|
||
| 12. Non-nullable tag ``external_id`` | ||
| ===================================== | ||
|
|
||
| Status | ||
| ------ | ||
|
|
||
| Proposed | ||
|
|
||
| Context | ||
| ------- | ||
|
|
||
| `ADR 0010 <0010-mutable-tag-external-id.rst>`_ made ``Tag.external_id`` mutable through an | ||
| import-only rename pathway, to serve as the institution-editable "Competency ID" for | ||
| Competency-Based Education (CBE) tags. That ADR left ``external_id`` nullable, scoped to | ||
| competency taxonomies. | ||
|
|
||
| Building the CBE competency workflow surfaced a stronger requirement: every tag needs a | ||
| value for ``external_id`` in order to be able to associate it with objects as part of | ||
| competency criteria. It's also anticipated that non-CBE taxonomies will want a mandatory, | ||
| institution-assigned identifier in the future, particularly when there becomes a desire to | ||
| convert regular taxonomies into Competency Taxonomies. | ||
|
|
||
| Today, ``external_id`` is nullable specifically to work around its own | ||
| ``unique_together(taxonomy, external_id)`` constraint: NULL is the only value Postgres, | ||
| MySQL, and SQLite all treat as exempt from a uniqueness constraint, so it's the | ||
| mechanism that lets multiple tags in the same taxonomy have "no external_id" at once. | ||
| Making the field mandatory means every existing NULL row in every current Open edX | ||
| instance needs a real value before the constraint can be added, since this is a | ||
| published library whose migrations run unmodified against arbitrary downstream data. | ||
|
|
||
| Decision | ||
| -------- | ||
|
|
||
| Make ``Tag.external_id`` ``NOT NULL`` at the database level, for every taxonomy, in one | ||
| migration. | ||
|
|
||
| - **Backfill for existing rows.** A ``RunPython`` step generates a value for every tag | ||
|
Contributor
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. Nit. Sounds fine, but curious why not two migrations - a data migration for backfill followed by a migration with this constraint? |
||
| whose ``external_id`` is currently NULL, following the same | ||
| backfill-then-constrain shape as the ``depth``/``lineage`` migration | ||
| (``0020_tag_depth_and_lineage.py``): populate real values first, then add the | ||
| constraint in the same migration once every row satisfies it. | ||
| - **Backfill algorithm.** Take the first ~7 characters of the tag's ``value`` (matching | ||
|
Contributor
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. Why the first 7 characters, not all of the value? |
||
| ``external_id``'s existing case-insensitive collation). If that prefix collides with | ||
| another tag's ``external_id`` already in the same taxonomy, append a numeric counter | ||
| to the prefix and keep incrementing it until it's unique. This is deterministic and | ||
| bounded: it never needs an unbounded retry loop, since appending an incrementing | ||
| counter to an already-short prefix is guaranteed to terminate within the taxonomy's | ||
| tag count. | ||
| - **New tags going forward.** The REST API and import format keep treating | ||
| ``external_id`` as optional for the caller, unchanged from today. When a tag is | ||
| created without one, the same prefix-plus-collision-fallback algorithm generates one | ||
| automatically, rather than rejecting the request. No existing integration that omits | ||
| ``external_id`` today has to change. | ||
| - **Institutions can replace an auto-generated value.** If an institution doesn't want | ||
| the auto-generated ``external_id``, they can change it to their own value through | ||
| ADR 0010's rename pathway. | ||
|
|
||
| This is not treated as a breaking change to the public REST API, and doesn't go through | ||
| the DEPR process. The request side (``external_id`` on tag creation) stays optional, so | ||
| no existing caller has to change what it sends. The response side narrows from | ||
| "nullable string" to "always a string," which is the safe direction for consumers: | ||
| code written to expect a possible ``null`` simply never takes that branch anymore. | ||
|
|
||
| Future considerations | ||
| ~~~~~~~~~~~~~~~~~~~~~~ | ||
|
|
||
| If an institution converts an existing regular taxonomy to a CBE taxonomy, its tags | ||
| will already have a non-null ``external_id`` (auto-generated or backfilled) from this | ||
| decision, with no separate migration needed for that conversion. | ||
|
|
||
| Rejected Alternatives | ||
| ---------------------- | ||
|
|
||
| Derive backfilled values from the tag's primary key | ||
| ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ | ||
|
|
||
| Still needs a collision fallback: a tag could already have a hand-assigned | ||
| ``external_id`` that happens to match another tag's ``id``. Also rejected because | ||
| ``id`` is meant to stay internal (per the README's model conventions, ``uuid`` is the | ||
| documented identifier for external reference), and because it produces a value with no | ||
| relationship to the tag (e.g. ``482913``), which reads as meaningless to institutions. | ||
|
|
||
| Require callers to supply ``external_id`` on tag creation | ||
| ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ | ||
|
|
||
| Simpler mechanically than auto-generation, but rejected because it's a breaking change | ||
| for the REST API and any import/UI flow that omits ``external_id`` today. | ||
|
|
||
| Scope this decision to CBE/competency taxonomies only, for now | ||
| ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ | ||
|
|
||
| Rejected because taxonomies later converted to Competency Taxonomies will want this too. | ||
|
|
||
| Changelog | ||
| --------- | ||
|
|
||
| 2026-07-10: | ||
|
|
||
| * Proposed. | ||
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Nit. For clarification: this just means that if we don't have nullable
external_id, then everyexternal_idin the same taxonomy must be unique and we can't use a workaround like a string to avoid external_id if it's not available, correct?And this is does not pose a problem for us because the "value" field already is unique per taxonomy, so just backfilling to external_id will have external_id be unique.