s we discussed in the thread on Spec thread for our new API, the idea
came up of making conn_id unique in Airflow

https://lists.apache.org/thread.html/rfb6f7d95b2754fda9dd09b08444214dfad12d10f143d32de0fcf4104%40%3Cdev.airflow.apache.org%3E

The people in that thread seemed to be in favour of it:

• It is often confusing to users (there have been a few bug reports about it over the years)
• It's questionable if it actually works well or not
• There are better/smarter tools for loadbalancing connections to a DB than picking one of a random list
• For Hive at least it has been implemented another way - allowing two host's in a single connection string [AIRFLOW-3888] HA for metastore connection #4708 (available in 1.10.6)
• It makes the HTTP API confusing (needing a integer connection ID, and a string conn_id field)

Given the downsides/work arounds, and the confusion this causes we should remove this (mis)feature from Airflow 2.0.

We should take some care in the migration that does this to capture the unique constraint error and report it a a meaninful way to the user.