stephen

Deployed this and it looks great! Thanks again!

Amazing, thank you!

Great, thank you 🙏

The motivation for the request is that most of these logs are low value to us (see first screenshot, 66k logs in 12h). I now realize that some of the other non-/healthz + non-/graphql logs could maybe be interesting (second screenshot, 51 logs in the same 12h period) but we don't currently have a use for them, so happy to sacrifice them in order to cut down the noise. If PHOENIX_LOGGING_LEVEL also applied to these, that would work.

That'd be awesome, thanks!

Yeah, we already have PHOENIX_LOGGING_LEVEL=WARNING set but these logs are coming from uvicorn afaict. This seems to silence them:

diff --git a/src/phoenix/server/main.py b/src/phoenix/server/main.py
index 1e219418..b31086a1 100644
--- a/src/phoenix/server/main.py
+++ b/src/phoenix/server/main.py
@@ -459,7 +459,9 @@ def main() -> None:
         host=host,  # type: ignore[arg-type]
         port=port,
         root_path=host_root_path,
+        access_log=False,
     )

     if tls_enabled_for_http:
         assert tls_config

Yeah, we are running Phoenix in prod so /healthz gets hit frequently. We can filter them out at query time, but would prefer to not be paying our logging provider for them. Would you accept a PR to configure this via env var, similar to:

"""
The logging level ('debug', 'info', 'warning', 'error', 'critical') for the Phoenix backend. For
database logging see ENV_DB_LOGGING_LEVEL. Defaults to info.
"""
ENV_DB_LOGGING_LEVEL = "PHOENIX_DB_LOGGING_LEVEL"

RunLLM The docs say you can pass environment variables as flags 🤷