A status code -9 means the OS sent SIGKILL. In Docker this is most often OOM (out of memory), not a Prefect error. Upping Docker Desktop’s memory usually fixes it, but there are a few angles to check: Quick checks - Was the container OOM-killed? Run:

docker ps -a
  docker inspect <container_id> --format '{{.State.OOMKilled}} {{.State.ExitCode}}'

If

OOMKilled

is true, you need more memory (or lower usage). - Watch memory in real time:

docker stats

If you’re using a Docker work pool (Prefect 3.x) - Set a higher memory limit for the job container via

job_variables

on your deployment. Example:

from prefect import flow

  @flow
  def my_flow():
      ...

  if __name__ == "__main__":
      my_flow.deploy(
          name="mem-test",
          work_pool_name="my-docker-pool",
          image="your/image:tag",
          job_variables={
              "mem_limit": "4g",
              "memswap_limit": "6g"  # optional; total = mem + swap
          },
      )

- Or set it per-deployment in prefect.yaml:

deployments:
    - name: mem-test
      entrypoint: flows/my_flow.py:my_flow
      work_pool:
        name: my-docker-pool
        job_variables:
          mem_limit: "4g"
          memswap_limit: "6g"

- Or set defaults at the pool level by editing the base job template (add

"mem_limit": "4g"

in the job_configuration), via UI or CLI:

# See the default template
  prefect work-pool get-default-base-job-template --type docker

  # Update a pool with a JSON template file
  prefect work-pool update --base-job-template base-job-template.json my-docker-pool

If you’re running your flow inside a general-purpose Docker container (not a Docker work pool) - Increase Docker Desktop’s memory allocation (Preferences > Resources). - If you start the container yourself, run it with a higher limit:

docker run --memory=4g --memory-swap=6g ...

Other mitigations - Reduce memory usage in the flow (chunk data, stream instead of loading whole datasets, avoid large in-memory objects, write intermediates to disk). - If using parallelism, lower concurrency so fewer tasks run at once. Can you confirm: - Are you using a Docker work pool to execute your flow, or running Prefect inside a general Docker container? - Roughly how much memory your workload needs? With that, I can suggest exact

job_variables

or template changes.