Hi Team, I鈥檓 testing persistence with AWS Aurora Serverless from a Notebook and it still seems to send traces to SQLite instead of Postgres.
os.environ["PHOENIX_SQL_DATABASE_URL"] = "postgresql://<DB_STRING>"
px.launch_app()Hey Sean L. we decided not to support postgres from the notebook as we felt pg persistence was better suited for long running servers rather than notebook runtimes. Any hesitations with actually deploying a phoenix server? I think we definitely can make the postgres connection from the notebook work, we just didn't anticipate this pattern as we want to control the number of servers connected to the database and this sorta breaks that model.
You. can still connect to a postgres provisioned phoenix from the notebook.
Got it. The usecase is that since the free version of phoenix does not have authentication, I wanted to run a phoenix in a long running server for ingesting spans with direct access blocked via network policy (e.g. private subnet) and run a separate local phoenix from notebook for accessing/running evals on the data. I thought since the both fronts shares the same database, I thought the notebook version will act as an access control.
I see, that makes sense. You could spin up the server locally to talk to the postgres?
python -m phoenix.server.main servethat way you still have a secure postgres instance but the server is only running locally
right, that will work. I think I can run it locally as a docker container too.
Thank you for the insight!
