Dockerfile

FROM python:3.13-slim
RUN pip install boto3
COPY s3rmrf.py /usr/local/bin/s3rmrf.py

job.yaml

apiVersion: batch/v1
kind: Job
metadata:
  name: s3-delete
spec:
  parallelism: 1
  template:
    metadata:
      labels:
        app: deleter
    spec:
      restartPolicy: OnFailure
      containers:
        - name: rmrf
          image: ghcr.io/minrk/mybinder-s3rmrf
          command:
            - python3
            - /s3rmrf/s3rmrf.py
            - mybinder-2i2c-registry
            - docker/
          env:
            - name: PYTHONUNBUFFERED
              value: "1"
            - name: AWS_ENDPOINT_URL
              value: https://nbg1.your-objectstorage.com
            - name: AWS_ACCESS_KEY_ID
              valueFrom:
                secretKeyRef:
                  name: hetzner-2i2c-harbor-registry
                  key: REGISTRY_STORAGE_S3_ACCESSKEY
            - name: AWS_SECRET_ACCESS_KEY
              valueFrom:
                secretKeyRef:
                  name: hetzner-2i2c-harbor-registry
                  key: REGISTRY_STORAGE_S3_SECRETKEY
          volumeMounts:
            - mountPath: /s3rmrf
              name: s3rmrf
              readOnly: true
      volumes:
          - name: s3rmrf
            configMap:
              name: s3rmrf

s3rmrf.py

"""
delete all s3 files in a prefix

way faster that aws s3 rm --recursive, which doesn't seem to do bulk delete for some reason

2 api requests per 1000 files

add to job with

kubectl create configmap s3rmrf --from-file ./s3rmrf.py
"""

import pprint
import time
from datetime import timedelta

import boto3

def main(bucket, prefix):
    s3 = boto3.client("s3")
    start = time.perf_counter()
    total = 0
    while True:
        listing = s3.list_objects_v2(Bucket=bucket, Prefix=prefix)
        objects = listing["Contents"]
        for obj in objects:
            if not obj['Key'].startswith(prefix):
                raise ValueError(f"Not deleting! {obj}")
        to_delete = [{'Key': obj['Key']} for obj in objects]
        to_delete[-1]['Key'] = 'nosuch'
        response = s3.delete_objects(Bucket=bucket, Delete=dict(Objects=to_delete))
        total += len(response['Deleted'])
        if set(response.keys()) != {"Deleted", "ResponseMetadata"}:
            pprint.pprint(response)
        duration = time.perf_counter() - start
        td = timedelta(seconds=int(duration))
        print(f"{td} {total=}, rate={total/duration:.0f}/s")


if __name__ == "__main__":
    import argparse
    parser = argparse.ArgumentParser()
    parser.add_argument("bucket")
    parser.add_argument("prefix")
    opts = parser.parse_args()
    main(bucket=opts.bucket, prefix=opts.prefix)