Short answer: - Bulk-cancel: there isn’t a single “cancel all” button, but you can cancel in bulk via a short script (Python or a shell loop). - Deleting the deployment: does NOT cancel existing flow runs (pending or otherwise). - Restarting the worker: does NOT cancel existing flow runs; pending runs remain pending and will be picked up again when the worker is back. Ways to cancel pending runs in bulk Option A — Python (most reliable) - Cancels all PENDING (and/or SCHEDULED) runs, optionally scoped to a specific deployment.

import asyncio
from prefect.client.orchestration import get_client
from prefect.client.schemas.filters import FlowRunFilter, FlowRunFilterStateType, FlowRunFilterDeploymentId
from prefect.client.schemas.objects import StateType
from prefect.states import Cancelled

DEPLOYMENT_ID = None  # or set to a specific deployment UUID string

async def main():
    async with get_client() as client:
        fr_filter = FlowRunFilter(
            state_type=FlowRunFilterStateType(any_=[StateType.PENDING, StateType.SCHEDULED])
        )
        if DEPLOYMENT_ID:
            fr_filter.deployment_id = FlowRunFilterDeploymentId(any_=[DEPLOYMENT_ID])

        runs = await client.read_flow_runs(flow_run_filter=fr_filter, limit=1000)
        for run in runs:
            await client.set_flow_run_state(run.id, Cancelled(name="Bulk cancel"))

asyncio.run(main())

Docs: - Cancel workflows in Prefect 3 Option B — CLI loop (quick and dirty) - List pending runs, grab IDs, and cancel them. You can narrow by flow name with

--flow-name

. - Note: output formatting can vary; test on a few runs first.

# List up to 1000 pending runs, extract IDs, cancel each
prefect flow-run ls --state PENDING --limit 1000 \
| awk 'NR>2 {print $1}' \
| xargs -n1 prefect flow-run cancel

CLI refs: -

prefect flow-run ls --help

(shows filtering by state/flow name) -

prefect flow-run cancel --help

Pausing work to stop new pickups (optional) - If you need to stop workers from pulling more runs while you clean up:

prefect work-queue pause <queue-name>
# later...
prefect work-queue resume <queue-name>

What happens if… - You delete the deployment: existing runs are unaffected; they will not be auto-cancelled. You must cancel them explicitly. - You restart the worker (Kubernetes worker): does not cancel pending runs. Already-launched Kubernetes pods keep running; runs that haven’t launched a pod yet remain pending and will be picked up again when the worker is back. If you share how you want to scope the cancellations (all pending, only for a deployment, only for a flow name, only older than X minutes), I can tailor the exact script/command.