Description

There is currently no way to disable client-side event emission via configuration. The should_emit_events() function in src/prefect/events/worker.py hardcodes its logic based on PREFECT_API_URL and PREFECT_API_KEY:

def should_emit_events() -> bool:
    return (
        emit_events_to_cloud()
        or should_emit_events_to_running_server()
        or should_emit_events_to_ephemeral_server()
    )

When PREFECT_API_URL is set (which it always is for self-hosted Prefect), should_emit_events_to_running_server() returns True unconditionally. There is no env var or setting to override this.

Use Case

We run long-running ingest jobs (10+ hours) on a self-hosted Prefect server. We don't use Prefect's event-driven features (automations, triggers) — we only use flows, tasks, and concurrency limits. During transient server outages, the EventsWorker queue accumulates events without bound (see related: the queue is queue.Queue() with no maxsize), contributing to memory pressure in our worker pods.

Our current workaround is monkey-patching:

import prefect.events.utilities
prefect.events.utilities.should_emit_events = lambda: False

This works but relies on internal module structure and could break on any upgrade.

Proposed Solution

Add a PREFECT_EMIT_EVENTS (or PREFECT_CLIENT_EMIT_EVENTS) boolean setting that should_emit_events() checks first:

def should_emit_events() -> bool:
    if not PREFECT_EMIT_EVENTS.value():
        return False
    return (
        emit_events_to_cloud()
        or should_emit_events_to_running_server()
        or should_emit_events_to_ephemeral_server()
    )

Default: True (no behavior change). Setting PREFECT_EMIT_EVENTS=false would cleanly disable all client-side event emission.

Additional Context

The existing PREFECT_EVENTS_* settings (PREFECT_EVENTS_MAXIMUM_SIZE_BYTES, PREFECT_EVENTS_RETENTION_PERIOD, etc.) are all server-side settings. There are no client-side event settings today.

Happy to submit a PR for this.