Failing to load catalog on startup using FastAPI lifespan

[ancientpi3]

Hello,
We are encountering an issue when trying to handle start up and shut down using a FastAPI lifespan function implementation which we based off an example in your documentation (linked below). It appears that connect_to_db() and register_collection_catalog() will fail quietly on startup. We suspect it is because our Postgres instance isn't yet ready when these functions are being called on startup.

Calling the /collections endpoint immediately after startup returns:
{"detail":"Could not find collections catalog."}

If we manually trigger the /register endpoint afterward, the collections load correctly.

Your example implementation:
https://developmentseed.org/tipg/advanced/customization/

Our implementation:
https://github.com/SkyTruth/cerulean-cloud/blob/5caa95482e261209ae4ee31d68a111322c112c21/cerulean_cloud/cloud_run_tipg/handler.py

Do you have any recommended changes for getting startup to work using FastAPI lifespan functions?

[vincentsarago]

Hi @ancientpi3

you could write a custom version of this middleware https://github.com/developmentseed/tipg/blob/main/tipg/middleware.py#L73-L118

☝ will raise an issue if the catalog doesn't exist, yours could just run the function instead

[ancientpi3]

Thanks Vincent, we'll give this solution a shot! It's not an urgent bug or anything. We mostly just wanted to call out that the implementation provided in the TiPG documentation fails for our case.