Issue Description

Note For context on how we use TypeORM, I'v written an article in August.

Expected Behavior

EntityMetadataNotFoundError should not happen.

Actual Behavior

For some unknown reason we have errors EntityMetadataNotFoundError happening only for some DataSource. All DataSource are initialized with the same options (except for the schema property)

Each time the error happen the request became Idle and consume one connection slot on the database. Combined with our WebSocket retry mechanism it consumed all available slots in few minutes.

Steps to Reproduce

Right now we are failing to reproduce. On our DEV & PREPROD the error almost never appear (only on Production where we have a lot of charges).

In Production we have close to 1,000 schemas (for each schema we open a DataSource and we keep it in memory). Since 0.3.x we use our own internal Map with the schema name as key (since there is no more getConnection() etc..).

My Environment

Additional Context

We had already identified the error a few months ago on some of our services that failed on the first request for a DataSource (at least this is what I was thinking at the time.. since other requests on the same datasource were ok).

However we had not much errors, so it was still not very serious and we didn't worry more than that.

However yesterday for a new Discussion service using WebSocket the issue became just terrible with dozen of No metadata when all clients were connecting at the same time on the WebSocket.

At some point there is so much errors and idle requests that the service go crazy (probably cause all sockets are used.. I'v seen EMFILE error at time)

Relevant Database Driver(s)

We only work with postgres

Are you willing to resolve this issue by submitting a Pull Request?

• ✅ Yes, I have the time, but I don't know how to start. I would need guidance.