feat: Use Psycopg3 COPY #451
Conversation
| @@ -33,10 +33,11 @@ packages = [ | |||
|
|
|||
| [tool.poetry.dependencies] | |||
| python = ">=3.8" | |||
| faker = {version = "~=30.0", optional = true} | |||
| psycopg2-binary = "2.9.9" | |||
| faker = {version = "~=29.0", optional = true} | |||
There was a problem hiding this comment.
Any reason to downgrade this?
There was a problem hiding this comment.
Nope! My branch was slightly outdated 😅
| psycopg = "^3.2.3" | ||
| psycopg-binary = "^3.2.3" |
There was a problem hiding this comment.
Is there a reason we both the source and binary packages?
There was a problem hiding this comment.
@edgarrmondragon This can be changed to just psycopg. Although now I am wondering if a user wants to use psycopg[c] or psycopg[binary] what would be the suggestion?
There was a problem hiding this comment.
dbt-labs/dbt-postgres#96 is probably a good case study. Most users in the data space can't or don't want to build C extensions, so we'll probably prefer psycopg[binary].
|
|
||
| # Use copy to run the copy statement. | ||
| # https://www.psycopg.org/psycopg3/docs/basic/copy.html | ||
| with connection.connection.cursor().copy(copy_statement) as copy: # type: ignore[attr-defined] |
There was a problem hiding this comment.
What happens at this point if someone sets postgresql+psycopg2 for dialect+driver?
There was a problem hiding this comment.
@edgarrmondragon It would raise an exception. In the current main branch I don't think using anything aside from postgresql+psycopg2 would work anyway, so this being configurable doesn't add much.
There was a problem hiding this comment.
I'm not 100% sure, but I don't think we use driver-specific APIs and rely on SQLAlchemy DDL/DML in all places, so I would expect most drivers to work. Maybe I'm wrong.
Co-authored-by: Edgar Ramírez Mondragón <[email protected]>
This PR is aimed at using the Psycopg3 COPY functionality to speed up batch inserts.
Psycopg3 has the benefit of not needing to be file-based for COPY operations, so it avoids many of the issues with escaping values mentioned here:
#370
It also drastically decreases memory usage since we don't need to create a file in-memory to then serve to the database. All binding is done on server side (unlike psycopg2 which does it client-side).
This would require that we switch to psycopg3 as mentioned here:
#433