Short answer: set those on your services container, not on the webserver started with

--no-services

. -

PREFECT_SERVER_SERVICES_EVENT_PERSISTER_BATCH_SIZE

and

PREFECT_SERVER_SERVICES_EVENT_PERSISTER_FLUSH_INTERVAL

are only read by the Event Persister service. When you start the API with

prefect server start --no-services

, that process does not run services, so these vars have no effect there. They take effect in the separate services process/container. - Event Persister implementation: event_persister.py - Related settings model: server services settings - Note: there are backward-compatible aliases like

PREFECT_API_SERVICES_EVENT_PERSISTER_*

as well. -

PREFECT_REDIS_MESSAGING_CONSUMER_TRIM_EVERY

is used by Redis-based consumers (e.g., the Event Persister’s Redis consumer). In

--no-services

mode the webserver process won’t use it; it will apply in the services container where those consumers run. - Redis consumer settings/logic: prefect_redis messaging If you’re splitting into an API/webserver container and a services container: - Put all three env vars on the services container. - It’s fine if they’re also present on the API container, but they won’t do anything there with

--no-services

. If helpful, you can confirm which services are enabled/active with:

prefect server services list-services

Want me to point to the exact place to set them in your Docker Compose? Feel free to share a snippet.

Given ~8k events/day (~5.5/min, ~1 every 11s), the defaults are already well‑suited. Here are practical starting points and when to tweak them: Recommended starting values (services container):

PREFECT_SERVER_SERVICES_EVENT_PERSISTER_BATCH_SIZE=20
PREFECT_SERVER_SERVICES_EVENT_PERSISTER_FLUSH_INTERVAL=5
PREFECT_REDIS_MESSAGING_CONSUMER_TRIM_EVERY=300

Why: - With your low, steady rate, batches won’t often fill; flushes will typically be driven by the interval. A 5s interval keeps UI/analytics reasonably up-to-date while keeping DB writes low. - Trimming Redis every 5 minutes further reduces background work without meaningful stream growth at your volume. If you want to tune: - Lower-latency persistence: set

FLUSH_INTERVAL=1-2

(more frequent smaller writes). - Fewer DB writes during occasional bursts: increase

BATCH_SIZE

to 50-100; keep

FLUSH_INTERVAL

at 5-10 to bound latency. - Even less Redis background work: increase

TRIM_EVERY

to 600s; just ensure your stream doesn’t grow unbounded if you have occasional spikes. Notes and source: - Event Persister (reads the

BATCH_SIZE

and

FLUSH_INTERVAL

settings): event_persister.py - Services settings model (env var names): server services settings - Redis consumer (uses

TRIM_EVERY

): prefect_redis messaging If you share your Docker Compose for the services container, I’m happy to drop these in the right place.