Export SQL for a migration #7

Activity

1. olirice

   added

   enhancementNew feature or requestNew feature or request

   on Jun 1, 2023
2. TemaDeveloper commented on Apr 20, 2026

   

   TemaDeveloper

   on Apr 20, 2026

   More actions

   Hi @olirice — happy to take this. Proposed API per the issue body:

   def export_migration(
       self,
       *,
       include_extension: bool = True,
       include_schema: bool = True,
       if_not_exists: bool = True,
   ) -> str:

   Returns the DDL needed to recreate the collection's table and metadata index, suitable for pasting into Alembic / sqitch / raw migration files. Sample output for create_collection(name="docs", dimension=3):

   create extension if not exists vector;
   create schema if not exists vecs;
   CREATE TABLE IF NOT EXISTS vecs.docs (
       id VARCHAR NOT NULL,
       vec VECTOR(3) NOT NULL,
       metadata JSONB DEFAULT '{}'::jsonb NOT NULL,
       PRIMARY KEY (id)
   );
   create index if not exists ix_meta_docs
     on vecs."docs" using gin (metadata jsonb_path_ops);

   Two design choices I'd like to confirm before I open a PR:

   1. Index naming. _create currently uses ix_meta_<uuid> for the metadata GIN index to avoid collisions. For a migration this needs to be deterministic, so I'd switch to ix_meta_<collection_name>. Sound right, or would you prefer the caller pass an explicit name?
   2. Vector indexes excluded. create_index picks n_lists from the current row count and chooses HNSW vs IVFFlat from the live pgvector version, so I'd leave it out of export_migration. Users would still call collection.create_index() after data is loaded. Agree, or do you want a separate export_index_migration(method=..., n_lists=...)?

   I have an implementation + tests ready locally; happy to push as soon as you ack the approach. PR will follow the Conventional Commits format (feat(collection): add export_migration for SQL DDL export) and link back to this issue.