Data WG 2022-03-24 Meeting Notes

5M

Followup on Elasticsearch → OpenSearch plans

Edward Zarecor

• There’s a new Slack channel, #search-migration, for tracking plans and progress

• The current proposal is to migrate a number of our services to use Opensearch

• Discovery was done into abandoning *search in favor of MySQL full text search, but was deemed not currently feasible because of performance problems

10M

edx insights acceptance tests

Simon Chen

• Do any of the Open edX members run acceptance tests on edx-analytics-dashboard? If so, how do you do it?

• How valuable do you find these acceptance tests?

• Notes

  • The tests are not trivial to run locally

  • The use bokchoy which has/is being deprecated

  • Do they run in CI or via the deployment pipeline?

  • There do seem to be some acceptance tests running in GitHub

  • Cost of doing nothing is maintenance, but isn’t high

  • Jill Vogel any opinion on this matter?

10M

learner view deprecation

Andy Shultz (Deactivated)

• We’re deprecating the learner view because it is expensive and not analytics

• Where is the right place for this student info to live? Is this “data” even a concern for this WG? The only actual users we have found for it use it as a student directory!

• Notes

  • View was of detailed user data and modules they interacted with.

  • Was sparsely used

  • PII risk

  • Other alternatives exist for getting similar data, but maybe not to the module level

  • We can see how they are doing on questions, but not at as low a level

  • Will Insights by design avoid pulling in PII in the future? Yes. Will focus on aggregates over learner level data which can leak out in numerous ways.

  • Can we add an ADR specifying this decision

  • Already turned off at edX

  • Deprecation has been announced, end of March is the comment deadline

  • It would be removed in Nutmeg

  • If this data comes back, individual learner engagement, is that a matter for this group?

    • Yes and no, an inform of plans to build a student directory would be helpful however.

Exporting courses via an API

Tobias Macey

• ODL have a beta API for course export

• Package is on pipy, ol-openedx-course-export

• https://pypi.org/project/ol-openedx-course-export/

• https://github.com/mitodl/open-edx-plugins/tree/main/src/ol_openedx_course_export

• In order to inform analytics they pull down course content to figure out block ids to map progress to content

• Can post a list of course ids to export

• Has a get endpoint for status of export

• Interested in contributors?

  • Yes

  • In a plugin mono repo

  • Need to be familiar with the pants build tool

• How does it work?

  • Uses mostly the same machinery as the Studio export, but exposes it as a RESTful API

  • The format is OLx

  • Is there a RESTful endpoint for posting a tarball? Unsure. There’s a way to connect to a git repository.

  • The sysadmin functionality has also been moved into a plugin

  • API is asynch and uses celery for executing the work

  • Currently the API is opinionated about S3, but could be easily adapted to use Django storages

  • There is this version of an import API.

• Should we be deprecating other versions of export/import in favor of this or an extended version of this?

• The main need here is to re-hydrate block ids and there’s hope that there will be an easier way to do that in the future

  • Would the course block API enable this today? Maybe, would require some discovery.

  • With a block id you can get usage ids, titles, etc.

  • Not sure what happens for items that are not connected to the course tree like textbooks

Standard for ETL for Insights

Maria Fernanda Magallanes Z

• Starting a conversation among Insights users on standards and how to keep it going as 2U deprecates pipeline

• Insights is the only place that access control is done today in our instructor analytics system

• Insights currently does data transformation beyond what is done in the API, can we eliminate this?

• How can we move into a phase of converging on a design recommendation?

• Starting with ETL and defining our migration path in phases may be a way of doing this.