|Approver (project plan)||Kriti||Needs to approve project plan|
|Approver (backend)||Brent||Needs to approve back end spec|
|Contributor (requirements)||Brent||Creates product spec, requirements, GitHub issues|
|Contributor (requirements)||Dom||Creates product spec, requirements, GitHub issues|
|Contributor||Anish||Coding and reviewing|
|Contributor||Dom||Coding and reviewing|
|Contributor||Mukesh||Coding and reviewing|
SQLAlchemy is a fantastic piece of software, but it’s becoming less suitable to our needs as the project evolves. In particular, we’ve found it to be weak and inefficient for our use case which doesn’t assume a stable underlying data model at any point. This means we have to constantly reflect the database state in order to perform operations on it, and maintaining this reflection of the database has become more and more burdensome.
Our use of SQLAlchemy in this project is slowing down development, and indeed slowing down the performance of the app in a massive way.
We’ll remove SQLAlchemy from our codebase. Note that at the point we undertake this project, it should already be gone from all DDL, DML, and DQL operations. What remains should be only Data Definition Description Language (D3L) operations. D3L operations are those that amount to DQL operations on metadata. That is, they query for metadata about database objects. An example is “list the tables in
my_database”. Another is “List the columns in
mytable and their types”. Hopefully, most of these functions will already have been removed during previous phases, but there are some which actually pack API calls (e.g., to list the schemas in a database).
Many of these will already have been created during previous phases. This should just amount to batting cleanup. But, I expect some will still need to be created.
Do not replace or create functions that are no longer needed. Some common sense will be required here.
Note that this section may take more lateral thinking than for the DDL, DML, or DQL phases. The implementers will have to be careful to remove rather than create complexity.
Refactor to remove SQLAlchemy from the codebase wherever it remains, and delete any unneeded functions (e.g., scaffolding).
After this part, the following should be true:
Generally, at this point, the identifiers should be OIDs and attnums, and the only reason to know any other metadata about a database object should be to return it to the API.
Note: Parts of this timeline are delayed, since they’re blocked by the removal of DDL, DML, DQL, and Brent’s parental leave.
|2023-03-20||Prototyping work starts|
|2023-05-12||Implementation spec and prototyping complete|
|2023-06-26||Implementation spec approved|
|2023-06-09||All needed D3L SQL Functions written|
|2023-06-16||All thin python wrappers written|
|2023-06-23||Refactor and clean up complete|