July 2024 - Project Tanager Workgroup

Owned by Stephen Fuqua
Last updated: Jul 26, 2024
4 min read

July 25, 2024, 11:00 am - 11:45 am central. Contact @Stephen Fuqua for the meeting invitation.

Agenda

  • Demonstration

  • Review the roadmap

  • Discuss / provide input on architecture

Attendees: JF Guertin (EdWire), Max Paulson (Denver Public Schools), Geoff McElhanon (Edufied), Doug Loyo (Edufied), Joshua Impson (Resultant), Erik Joranlien (Education Analytics), Brad Banister (Doubleline Partners on behalf of Ed-Fi), Stephen Fuqua (Ed-Fi Alliance).

Demonstration

Rather than review milestone v0.1.0 release, let’s see what we have today, now that we have added reference validation.

Meeting notes:

  • Walked through execution of a demonstration script showing referential integrity checks.

  • Described that JSON validation checks are also in place at this time.

Roadmap

Milestone

Functional Goals

Milestone

Functional Goals

0.1 done

Compliant Discovery API, Descriptor API, and Resource API definition (except GET by query): able to run bulk upload, smoke test. Includes JSON validation based on API schema file. Fake OAuth (1).

0.2 in progress

Reference validation, Streaming, and Profiles: rejects POST, PUT, and DELETE requests that would violate referential integrity. Streaming data out. Build basic Profiles support (2).

0.3 by Summit

GET by query and cascading updates: use search engine or relational DB to fulfill GET by query requests. Support cascading updates on allowed resources.

0.4 need to accelerate

Namespace authorization: real OAuth; JWT inspection; duplicate ODS/API's namespace authorization. First release of the Configuration Service.

0.5

Data model flexibility and Concurrency: extensions, choosing between DS 4 and DS 5, swapping data standards at start up (not compile). Dynamic Discovery API definition, based on actual Data Standard/extensions. Full support for eTag-based concurrency.

0.6

Dynamic profiles and multitenancy: full-fledged support for XML-based dynamic profiles, and for ODS/API 7 style multi-tenant routing and database segmentation.

0.7 tech congress?

Ed-org based authorization. (3)

0.8

Change queries.

Meeting notes:

  • Milestone 0.2 under way now, and targeting at least milestone 0.3 complete by the time of the Ed-Fi Summit at the end of September.

  • Swap “get by query” and “basic profiles” support between milestones 0.2 and 0.3.

  • Need to try to accelerate for a good release candidate at Tech Congress 2025.

  • Who can help?

    • JF: perhaps help with multinenancy / routing? Suggested to write up design notes first in this GitHub discussion.

    • Max: Lambda function as alternate front end interface. We can also look into other areas of interest in the roadmap.

Architecture

Database Design

Project-Tanager/docs/DMS/PRIMARY-DATA-STORAGE at main · Ed-Fi-Alliance-OSS/Project-Tanager (github.com)

“Single table” design inspired by NoSQL document stores, with added benefit of RDBMS-managed foreign key references.

  • document holds the Resource and Descriptor JSON documents

  • alias helps with referential integrity in inheritance structures: e.g. a School is an EducationOrganization. If JSON payload has an educationOrganizationId, then this table solves the problem of not knowing which type of Education Organization to query.

  • reference further helps with referential integrity checks, for example not able to delete a document that is referenced by another document.

Each of these three tables is partitioned out of the box for better performance. Default: 16. Easily supports up to 256 partitions; could be more with a very small tweak.

image-20240725-014052.png

Bulk Load Performance

Grand Bend data set (“populated template”). Running in Docker containers on localhost.

System

Data Set

Timing

Row count

System

Data Set

Timing

Row count

DMS

Descriptors only

0:19.9 minutes

3,201 in document

+ 3,201 in alias

ODS/API 7.1

Descriptors only

6:45.6 minutes

3,201 in descriptor

+ 3,201 across all of the descriptor tables

Isolation Level

Read Committed vs. Repeatable Read vs. Snapshot vs. Serialized

Ex: read before write when a delete fails because the deleted item is referenced by something else. Read committed could allow a change to the “something else”, making it impossible to report the problem with the delete. Repeatable read solves this, but can cause a transaction rollback for another “concurrent” (but second-in) transaction. We think this conflict is unlikely to occur frequently. If using read committed, would probably want to do more manual locking.

Search Database

image-20240725-025521.png

Project-Tanager/docs/DMS/CDC-STREAMING.md at main · Ed-Fi-Alliance-OSS/Project-Tanager (github.com)

 

Reading straight from dms.document, no outbox event table → open to suggestions / design for an additional outbox table.

Alternative

Additional query tables in PostgreSQL / MSSQL with painful indexing. States are likely to want this, and we have a design in mind already.