commit 9c57dd2b0c687e71ae18e6bb82c2705247340963
Author: chris <cg@zknt.org>
Date:   Sun Feb 7 20:49:47 2021 +0100

    docker-garbage-collect.md

diff --git a/docker-garbage-collect.md b/docker-garbage-collect.md
new file mode 100644
index 0000000..74e83e5
--- /dev/null
+++ b/docker-garbage-collect.md
@@ -0,0 +1,103 @@
+# Is the docker registries garbage collection garbage?
+
+Step 1) Run garbage collection on my registry
+
+    docker-compose exec registry bin/registry garbage-collect /etc/docker/registry/config.yml
+    ....
+    4005 blobs marked, 0 blobs and 0 manifests eligible for deletion
+
+So everything seems used in some image:tag.
+
+Step 2) Check size
+
+    du -sch $(docker volume inspect registry_registry -f "{{.Mountpoint}}")
+    50G     /var/lib/docker/volumes/registry_registry/_data
+
+This is big, for the amount of images in there.
+It's 73 images, all originating in the same base images.
+Everything is either alpine or debian.
+
+Step 3) Build a list of all images and all tags, download everything
+
+    python list-all-repos.py --registry ... --ask-pass > all-repos
+    for img in $(cat all-repos); do docker pull $img; done
+
+(See below for script source)
+
+Step 4) Throw away all registry data
+
+    docker-compose down -v && docker-compose up -d
+
+Step 5) Upload everything again
+
+    for img in $(cat all-repos); do docker push $img; done
+
+Now that everything is up again, I checked that all images are there again.
+Just ran the script again and diffed the output. No differences.
+
+Step 6) Check size again
+
+    du -sch $(docker volume inspect registry_registry -f "{{.Mountpoint}}")
+    6.4G    /var/lib/docker/volumes/registry_registry/_data
+
+    ...garbage-collect
+    473 blobs marked, 0 blobs and 0 manifests eligible for deletion
+
+So, what am I not getting? Shouldn't GC already have thrown away everthing
+not referenced by some image:tag?
+
+## The script to list all images
+
+This depends on [DXF](https://github.com/davedoesdev/dxf)
+
+```python
+"""Print all images and tags from given docker registry."""
+import os
+import argparse
+import getpass
+
+from dxf import DXF, DXFBase
+
+
+def _get_arguments():
+    parser = argparse.ArgumentParser(
+        description="Print all images and tags from registry."""
+    )
+    parser.add_argument('--registry', dest='registry')
+    parser.add_argument('--username', dest='username',
+                        help="Defaults to $REGISTRY_USERNAME")
+    parser.add_argument('--password', dest='password',
+                        help="Defaults to $REGISTRY_PASSWORD")
+    parser.add_argument('--ask-pass', dest='ask_pass', action='store_true',
+                        help='Interactively ask for password')
+    return parser.parse_args()
+
+
+def main(registry, username, password):
+    """Print all images and tag."""
+    regbase = DXFBase(registry)
+    regbase.authenticate(username, password)
+
+    for repo in regbase.list_repos():
+        regrepo = DXF(registry, repo)
+        regrepo.authenticate(username, password)
+        aliases = regrepo.list_aliases()
+        for alias in aliases:
+            print(f"{registry}/{repo}:{alias}")
+
+
+if __name__ == "__main__":
+    args = _get_arguments()
+    _username = (args.username if args.username else
+                 os.environ.get('REGISTRY_USERNAME'))
+    if args.ask_pass:
+        _password = getpass.getpass(f"{args.registry} password: ")
+    else:
+        _password = (args.password if args.password else
+                     os.environ.get('REGISTRY_PASSWORD'))
+    main(
+        args.registry,
+        _username,
+        _password,
+    )
+```