docs: Add how-to guide for SQLDatabaseLoader

[amotl]

• Description: Documentation about the generic document loader adapter for SQLAlchemy, SQLDatabaseLoader.
• Needed by: Add support for CrateDB to LangChain LLM framework crate-workbench/langchain#1
• Followup from: community[minor]: Add SQLDatabaseLoader document loader #18281
• Addressed to: @eyurtsev

Dear Eugene,

coming from GH-16246/GH-18281, this patch adds missing documentation that has still been included in crate-workbench#1. We wanted to break it out of that patch, in order to further minimise its size before submitting it.

With kind regards,
Andreas.

NB: Can I also humbly ask you to review/exercise this for validity, @simonprickett or @kneth? I think the demonstrated code should work with recent vanilla versions of LangChain, with all of SQLite, PostgreSQL, and CrateDB.

[vercel]

The latest updates on your projects. Learn more about Vercel for Git ↗︎

Name	Status	Preview	Comments	Updated (UTC)
langchain	✅ Ready (Inspect)	Visit Preview	💬 Add feedback	Oct 28, 2024 11:23pm

[amotl]

Submitting just a few comments from self-review.

[amotl]

Suggested change

	"# SQL Database\n",
	"# SQLDatabaseLoader\n",

[amotl]

Suggested change

	"For talking to the database, the document loader uses the [SQLDatabase]\n",
	"utility from the LangChain integration toolkit.\n",
	"For talking to the database, the document loader uses the [SQLDatabase]\n",
	"utility.\n",

[amotl]

Linking to the SQLDatabase Toolkit was probably wrong, so let's better link to the Utility instead?

Suggested change

	"[SQLDatabase]: https://python.langchain.com/docs/integrations/toolkits/sql_database\n",
	"[SQLDatabase]: https://python.langchain.com/api_reference/community/utilities/langchain_community.utilities.sql_database.SQLDatabase.html\n",

Is there a better piece of dedicated documentation about this component elsewhere in the documentation tree, i.e. are you advising to use a different target link here?

[eyurtsev]

Hi @amotl, thank you for taking the initiative to improve the how-to guides!

We don't have good guidelines yet for the how-to guides, but if you can browse through a few (https://python.langchain.com/docs/how_to/), you may get a bit of a sense of the format:

1. Use a notebook format only here (remove the mdx file).
2. Add a link to the notebook from the how-to guide index in an appropriate place
3. The title should be phrased as a how-to question (e.g., "How to ...")
4. If you can link to conceptual prerequisites (e.g., https://python.langchain.com/docs/how_to/passthrough/). Potentially OK to link to any SQL resources that make sense for folks to know about
5. Don't comment out the package install (maybe simply skip it if you already have the packages installed)
6. Review https://diataxis.fr/how-to-guides/ if you're not familiar with diataxis

[amotl]

Thank you for your guidance, Eugene. I will rework the patch according to your suggestions, and come back afterwards. The same guidelines will probably also apply to GH-27713.

Both pieces of documentation are coming from initially being conceived around LangChain <0.1 times, where blueprints from other how-to documents have not been so elaborate yet.

[kneth]

LGTM

I would appreciate additional reviewers

[amotl]

@kneth: The ingredients of this patch need further love, as outlined by @eyurtsev. That's why we toggled it into draft mode again. Relevant patches are welcome.

[eyurtsev]

Hi @efriis re-assigned to you. I'm not sure if this PR is being actively worked on right now (probably on the back burner). But I check the integrations page https://python.langchain.com/docs/integrations/document_loaders/ and don't think we have integration docs for this loader.

I suspect we'll want to match this pattern: https://python.langchain.com/docs/integrations/document_loaders/

[efriis]

ah yep I'll close this one and discuss with their team on #27710

I think they put up all the PRs at once hence the confusing sequencing, and the new integration process will be much better suited to this!

this one can be reopened after the integration is done/published/docs are in

[amotl]

Hi. I think this one can stay, as it improves existing functionality with corresponding documentation, that has been missed on the first iteration. I was keeping it in draft mode, to keep tracking it. However, it needs improvements like outlined by @eyurtsev.

[efriis]

ah I misunderstood. @amotl can we

• remove the changes to the docs/docs/how_to directory
• instead add a single ipynb page to docs/integrations/document_loaders/sql_database or something like that (matching the format eugene linked)

[amotl]

Hi Erick. Sure. As soon as I will resume working on this patch, I will commandeer it into the directions you are suggesting. Thank you very much.

[efriis]

great! if it's not in the next week, would recommend closing and reopening when you're ready for a review!

[amotl]

Hi Erick. Sure. Let's close, and re-open later.