Thanks for the detailed write-up — that helps a lot. A few things to check for Prefect 3.x: 1) Make sure the UI is not filtering out DEBUG - In the Flow Run’s Logs tab, use the Level filter and ensure DEBUG is included. It’s easy to miss that the UI can filter by level and sometimes defaults to showing INFO+. 2) Verify the flow run actually sees PREFECT_LOGGING_LEVEL=DEBUG - Add a quick check at the very top of your flow to confirm the settings inside the run process:

from prefect import flow, get_run_logger
from prefect.settings import get_current_settings

@flow
def debug_check():
    s = get_current_settings()
    logger = get_run_logger()
    <http://logger.info|logger.info>(f"Prefect logging level seen by flow: {s.logging.level}")
    logger.debug("This is a DEBUG message from get_run_logger()")

if __name__ == "__main__":
    debug_check()

- Run a new deployment-based run and confirm you see: - An INFO log line showing

Prefect logging level seen by flow: DEBUG

- The subsequent DEBUG log line 3) Confirm the ENV is attached to the flow runtime (not just the worker) - In Prefect 3.x,

deployments[].env

in

prefect.yaml

is applied to the flow run’s execution environment. You’ve confirmed the variable appears in the deployment details in the UI, which is good — but for Docker/Kubernetes work pools, a custom job template could accidentally omit or overwrite env. If you’re using Docker/K8s, double-check your work pool job template merges deployment env into the job/container environment. If you’re using a Process worker, it should “just work”. 4) Sanity check via CLI - After triggering a run, you can also confirm the backend has DEBUG logs with:

prefect flow-run logs <flow_run_id> --tail

If DEBUG lines show here but not in the UI, it’s definitely the UI filter. 5) No separate “server” log level - There isn’t a

PREFECT_LOGGING_SERVER_LEVEL

setting in Prefect 3.x. The main setting to raise verbosity is

PREFECT_LOGGING_LEVEL

. The API log submission is enabled by default; since you’re seeing INFO logs already, that’s working. 6) If you need more control - You can supply a custom logging config via

PREFECT_LOGGING_CONFIG_PATH

(maps to

logging.config_path

) to explicitly set handler levels (e.g., console and API handler) to DEBUG if you suspect a handler-level gate. This is generally not needed, but it’s available. If you try the snippet in #2 and don’t see

DEBUG

in the printed settings, that means the env var isn’t actually present in the run environment — in that case, let me know: - Which worker type you’re using (Process, Docker, Kubernetes)? - If Docker/K8s, please share whether you’ve customized the work pool job template. Helpful docs: - Add logging to flows - Advanced logging customization - prefect.yaml reference