Data WG 2022-02-10 Meeting notes

Abstraction Layer for Open edX Core Metrics

• Review the current list of core metrics

• We floated the idea of creating an abstraction layer for those core metrics. How should we go about implementing it?

  • Figures compat module

    • AppSembler are planning to follow Django release process

    • Events are a possible vanguard effort that is facing the same problems we have related to well specified data models.

      • Schema enforcement is possible with externalized definitions

      • Allows testing of contracts

      • The events team is using Avro schema

• Is there a way to demo Figures?

  • Open Question

    • Using the devsite

• Is there any sharable piece of the 2U DBT work?

  • Andy has been thinking about this, but 2U rely on a lot of upstream data cleansing in order to produce the final set of metrics.

  • Probably makes the most sense to have a separate community implementation

  • With an event stream you could fill the tables that edX analytics api uses

• Tracking logs

  • They are a mess

  • Our approach has been fix in the warehouse

  • Dave advocates for a approach that fixes the data upstream

Plugin for data API is being explored as an option by Tobias at MITx

@Tobias Macey

• Initial use case is related to exporting the course content.

• Another need is enriching block ids with human facing names and follow bread crumbs through course navigation

• Andy asks about the list of users for the course – this data is “toxic waste” and fill of PII that you would not want to get by mistake.

• Expected to use celery facilities for asynchronous

• Does this need to cover CCX courses?

  • Tobias does not think that it does.

  • Dave worries this may not work by default

• Would this plugin need to reach across boundaries and use ORM models from other apps?

  • V1 probably yes

  • Hope that this can help define stable APIs

• Is the long term vision batch forever or is streaming changes part of their future plans

  • See incremental updates over Kafka or Pulsar as a valuable future state

  • Focused on batch for now?

• Is there any design docs for the data API?

  • MIT are currently in the early discovery phase and the data that will be pulled hasn’t been fully defined.

  • API is focused on raw data now, not core metrics.

• MIT code will be in a public repo, and will publish details about design and progress in the Data Working Group channel.

• Is the plugin model baked enough to consider it a best practice for decomposing the monolith?

  • CI protections are still weak.

  • This was one of the goals of the events and signals work

  • Overall, stable APIs are emergent and not robust yet.