Hi, everyone! I'm new to prefect and have a question regarding some use cases that we saw whilst testing it in our environment: - Use of connection pool to connect to databases: - We actually saw this link here (https://github.com/PrefectHQ/prefect/issues/1876) and tried using it, but when we tried to register the flow to the server, it threw the error: " flow.register(project_name='name') TypeError: can't pickle _thread._local objects " - For us this is a major issue as within our flow, ideally we would have some light queries in a lot of tasks, using connection pooling is mandatory for this use case in particular. If anyone could point us in the right direction it would be very helpful! Thanks in advance, Luis

Hi @Luis Henrique, the error seems like you’re using a DaskExecutor. Is that right?

We're actually running it locally everything

I see, are you returning the connection in a task? What storage are you using?

We tried returning it in a task, when that didn't work we proceeded with the suggestion in the link above (using a global variable). And this worked fine for running the code directly, but when we tried to register the flow it threw the pickle error.

I see, can you try

from prefect.storage import Local
flow.storage = Local(stored_as_script=True)

We didn't try that! Thanks for the idea! We'll test it and see what happens. Many thanks!

@Luis Henrique - Any update? I am experiencing a very similar issue...

Hi, @Anastasia Belov! Yes, the solution provided by @Kevin Kho worked but it needed some extra work for deployment. What we'll do here is to actually get the database requests in an API so that we can bypass this on prefect

Note this only works for LocalExecutor @Anastasia Belov because the parallel executors would require serialization of Task outputs. You would need to make the connectionn inside those tasks