fix: read Learn's Fastly service ID from MIT_LEARN_FASTLY_SERVICE_ID by ChristopherChudzicki · Pull Request #3794 · mitodl/mitxonline

ChristopherChudzicki · 2026-07-28T18:33:37Z

What are the relevant tickets?

Refs https://github.com/mitodl/hq/issues/12381

Description (What does it do?)

Renames the Fastly token and URL settings to read the env vars set by fix: Set the proper env var for MITx Online to pick up the Fastly settings ol-infrastructure#5119 switched to setting (MITX_ONLINE_FASTLY_*),
- ☝️ renamed env vars somewhat erroneously, but the new names are more consistent with the rest of the env vars used in the repo.
Introduces MIT_LEARN_FASTLY_SERVICE_ID, replacing MITX_ONLINE_FASTLY_SERVICE_ID. The value is Learn's service ID — Learn is what tags its product pages with the mitxonline:* surrogate keys, so purging MITxOnline's own service does nothing. The old name invited exactly that mistake.
- No live code was using MITX_ONLINE_FASTLY_SERVICE_ID in a way that worked.
Makes service_id an explicit argument to queue_fastly_surrogate_key_purge, so a purge's target is visible at the call site. Defaults to None so messages enqueued by the previous release skip rather than raise mid-deploy.

Also reads these via django.conf.settings instead of importing from main.settings at module import time, which is what hid them from Django's settings fixture in tests.

How can this be tested?

Tests should pass. Not easily testable locally otherwise.

Additional Context

Safe to deploy alone:

the service ID is unset before (due to mismatch in name introduced by fix: Set the proper env var for MITx Online to pick up the Fastly settings ol-infrastructure#5119)
Even before name mismatch, the purge was non-functional because project id was wrong.

It starts working when the infra change below lands.

Soft → hard purge is deliberately excluded, so this infra change can be validated in RC while soft purge is still in place.

Checklist:

Add MIT_LEARN_FASTLY_SERVICE_ID to mitxonline/k8s_secrets.py, per environment. Values already exist as learn_service_id_{ci,qa,production} under infrastructure/fastly in src/bridge/secrets/concourse/operations.production.yaml. Not secret-mitxonline's fastly.service_id — that's MITxOnline's own service, and using it is what made the original purge a no-op.
Don't restore FASTLY_SERVICE_ID; nothing reads it now.
Confirm the token is authorized for Learn's service — purge-by-key needs purge_select on the target service, and this token comes from MITxOnline's Vault path.

github-actions · 2026-07-28T18:34:02Z

OpenAPI Changes

Show/hide changes

## Changes for v0.yaml:
No changes detected

## Changes for v1.yaml:
No changes detected

## Changes for v2.yaml:
No changes detected

Unexpected changes? Ensure your branch is up-to-date with main (consider rebasing).

The surrogate key purge targets MIT Learn's Fastly service, because MIT Learn is what tags its product page responses with the mitxonline:* surrogate keys. The setting holding that service ID was named MITX_ONLINE_FASTLY_SERVICE_ID and read from a bare FASTLY_SERVICE_ID env var, which reads as "MITxOnline's own service" and inverts this repo's convention -- elsewhere the app-prefixed name is the env var (EMAIL_BACKEND is read from MITX_ONLINE_EMAIL_BACKEND, and 16 others). That inversion is how the env vars came to be renamed to the setting names in mitodl/ol-infrastructure#5119, leaving both settings unset in production and the purge a silent no-op. Rename to MIT_LEARN_FASTLY_SERVICE_ID, matching the existing MIT_LEARN_* namespace, whose seven settings all use the env var name unchanged -- the prefix names the external system, so no app prefix applies. The token and API URL now follow the app-prefixed convention instead, reading MITX_ONLINE_FASTLY_AUTH_TOKEN and MITX_ONLINE_FASTLY_URL, which is what the deployment already sets. So the infra follow-up only has to add the service ID; nothing has to be reverted. Pass the service ID into queue_fastly_surrogate_key_purge explicitly rather than reading it from module state, so the target of a purge is visible at the call site instead of being an unqualified global in the MITxOnline repo. It defaults to None so messages enqueued by the previous release skip the purge rather than raising during a rolling deploy. Read the token and URL through django.conf.settings, removing the module-import-time coupling that made these values invisible to Django's settings fixture in tests. Adds cms/tasks_test.py and cms/signals_test.py; neither module had tests. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

The mitxonline Fastly surrogate-key purge must target MIT Learn's Fastly service (MIT Learn tags product-page responses with the mitxonline:* surrogate keys), not MITxOnline's own service. Add a new MIT_LEARN_FASTLY_SERVICE_ID env var, sourced per-environment from a new fastly.mit_learn_service_id key in each mitxonline SOPS secrets file (values copied from learn_service_id_{ci,qa,production} under infrastructure/fastly in operations.production.yaml). Supports mitodl/mitxonline#3794. Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

marslanabdulrauf

Added a couple of findings:

Rolling deploy is handled in one direction only
Minor cms/signals_test.py calls purge_fastly_cache_on_publish directly rather than firing page_published

marslanabdulrauf · 2026-07-30T10:27:00Z

+        service_id (str): The Fastly service ID whose cache should be purged.
+            Callers pass settings.MIT_LEARN_FASTLY_SERVICE_ID. Defaults to None
+            so that messages enqueued by an older release, before this argument
+            existed, skip the purge rather than raising during a rolling deploy.


This fix is one sided and will always work if the celery is the one to be deployed first.
The other way is if we started passing two positional arguments and the function only accept one.

Good point.

I do want to keep the service id passing explicit, since i think it's somewhat unexpected that mitxonline's queue_fastly_surrogate_key_purge purges a different service id. But I made the passing optional, and a follow PR, stacked on this one, makes it required.

Keeps the enqueued Celery message a single positional argument so a rolling deploy is safe in both directions: an old webapp's one-arg message runs on a new worker and still purges via the fallback, and a new webapp's one-arg message runs on an old worker unchanged. Passing service_id explicitly from the call sites is #3804, once every worker accepts the argument. Also drives cms/signals_test.py through Wagtail's real publish rather than calling the receiver directly, so the page_published.connect() wiring is covered. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

marslanabdulrauf

LGTM

ChristopherChudzicki added the Needs Review label Jul 29, 2026

ChristopherChudzicki force-pushed the fix/fastly-learn-service-id-env-var branch from 905c9a4 to 8a6457e Compare July 29, 2026 01:11

Ardiea self-assigned this Jul 29, 2026

Ardiea mentioned this pull request Jul 29, 2026

feat(mitxonline): add MIT_LEARN_FASTLY_SERVICE_ID to k8s secrets mitodl/ol-infrastructure#5158

Merged

2 tasks

ChristopherChudzicki unassigned Ardiea Jul 29, 2026

marslanabdulrauf self-assigned this Jul 30, 2026

marslanabdulrauf requested changes Jul 30, 2026

View reviewed changes

ChristopherChudzicki force-pushed the fix/fastly-learn-service-id-env-var branch 2 times, most recently from 6dea71e to 36fcddc Compare July 30, 2026 14:53

ChristopherChudzicki mentioned this pull request Jul 30, 2026

refactor: pass the Learn Fastly service ID at the call sites #3804

Open

ChristopherChudzicki force-pushed the fix/fastly-learn-service-id-env-var branch from 36fcddc to 221cae9 Compare July 30, 2026 14:57

marslanabdulrauf approved these changes Jul 31, 2026

View reviewed changes

ChristopherChudzicki merged commit f104673 into main Jul 31, 2026
14 checks passed

ChristopherChudzicki deleted the fix/fastly-learn-service-id-env-var branch July 31, 2026 13:13

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

fix: read Learn's Fastly service ID from MIT_LEARN_FASTLY_SERVICE_ID - #3794

fix: read Learn's Fastly service ID from MIT_LEARN_FASTLY_SERVICE_ID#3794
ChristopherChudzicki merged 2 commits into
mainfrom
fix/fastly-learn-service-id-env-var

ChristopherChudzicki commented Jul 28, 2026 •

edited

Loading

Uh oh!

github-actions Bot commented Jul 28, 2026 •

edited

Loading

Uh oh!

marslanabdulrauf left a comment

Uh oh!

marslanabdulrauf Jul 30, 2026

Uh oh!

ChristopherChudzicki Jul 30, 2026

Uh oh!

Uh oh!

marslanabdulrauf left a comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Uh oh!

Conversation

ChristopherChudzicki commented Jul 28, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

What are the relevant tickets?

Description (What does it do?)

How can this be tested?

Additional Context

Checklist:

Uh oh!

github-actions Bot commented Jul 28, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

OpenAPI Changes

Uh oh!

marslanabdulrauf left a comment

Choose a reason for hiding this comment

Uh oh!

marslanabdulrauf Jul 30, 2026

Choose a reason for hiding this comment

Uh oh!

ChristopherChudzicki Jul 30, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

marslanabdulrauf left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

ChristopherChudzicki commented Jul 28, 2026 •

edited

Loading

github-actions Bot commented Jul 28, 2026 •

edited

Loading