docker-rotate 2.0

.gitignore

@@ -55,3 +55,4 @@ target/
 
 cover
 .eggs
+.venv*

README.md

@@ -1,24 +1,24 @@
 docker-rotate
 =============
 
-In a continuously deployed environment, old and unused docker images accumulate and use up space.
-`docker-rotate` helps remove the K oldest images of each type.
+In a continuously deployed environment, old and unused docker images and containers accumulate and use up space.
+`docker-rotate` helps remove the K oldest images of each type and remove non-running containers.
 
 [![Build Status](https://travis-ci.org/locationlabs/docker-rotate.png)](https://travis-ci.org/locationlabs/docker-rotate)
 
 Usage:
 
-    # delete all but the three oldest images of each type
-    docker-rotate --clean-images --keep 3
+    # delete all but the three most recent images of each type
+    docker-rotate images --keep 3
 
-    # only target one type of image (by name)
-    docker-rotate --clean-images --keep 3 --only organization/image
+    # only target one type of image but don't remove latest
+    docker-rotate images --keep 3 --image "organization/image" "~:latest"
 
     # don't actualy delete anything
-    docker-rotate --clean-images --keep 3 --dry-run
+    docker-rotate --dry-run images --keep 3
 
-    # also delete exited containers (except those with volumes)
-    docker-rotate --clean-images --clean-containers --keep 3
+    # delete containers exited more than an hour ago
+    docker-rotate containers --exited 1h
 
 By default, `docker-rotate` connects to the local Unix socket; the usual environment variables will
 be respected if the `--use-env` flag is given.

dockerrotate/containers.py

@@ -0,0 +1,82 @@
+from datetime import datetime, timedelta
+from dateutil import parser
+from dateutil.tz import tzutc
+import re
+
+from docker.errors import APIError
+
+from dockerrotate.filter import include_image
+
+
+TIME_REGEX = re.compile(r'((?P<days>\d+?)d)?((?P<hours>\d+?)h)?((?P<minutes>\d+?)m)?((?P<seconds>\d+?)s)?')  # noqa
+
+
+def parse_time(time_str):
+    """
+    Parse a human readable time delta string.
+    """
+    parts = TIME_REGEX.match(time_str)
+    if not parts:
+        raise Exception("Invalid time delta format '{}'".format(time_str))
+    parts = parts.groupdict()
+    time_params = {}
+    for (name, param) in parts.iteritems():
+        if param:
+            time_params[name] = int(param)
+    return timedelta(**time_params)
+
+
+def include_container(container, args):
+    """
+    Return truthy if container should be removed.
+    """
+    inspect_data = args.client.inspect_container(container["Id"])
+    status = inspect_data["State"]["Status"]
+
+    if status == "exited":
+        finished_at = parser.parse(inspect_data["State"]["FinishedAt"])
+        if (args.now - finished_at) < args.exited_ts:
+            return False
+    elif status == "created":
+        created_at = parser.parse(inspect_data["Created"])
+        if (args.now - created_at) < args.created_ts:
+            return False
+    else:
+        return False
+
+    return include_image([container["Image"]], args)
+
+
+def clean_containers(args):
+    """
+    Delete non-running containers.
+
+    Images cannot be deleted if in use. Deleting dead containers allows
+    more images to be cleaned.
+    """
+    args.exited_ts = parse_time(args.exited)
+    args.created_ts = parse_time(args.created)
+    args.now = datetime.now(tzutc())
+
+    containers = [
+        container for container in args.client.containers(all=True)
+        if include_container(container, args)
+    ]
+
+    for container in containers:
+        print "Removing container ID: {}, Name: {}, Image: {}".format(
+            container["Id"],
+            (container.get("Names") or ["N/A"])[0],
+            container["Image"],
+        )
+
+        if args.dry_run:
+            continue
+
+        try:
+            args.client.remove_container(container["Id"])
+        except APIError as error:
+            print "Unable to remove container: {}: {}".format(
+                container["Id"],
+                error,
+            )

dockerrotate/filter.py

@@ -0,0 +1,22 @@
+import re
+
+
+def include_image(image_tags, args):
+    """
+    Return truthy if image should be considered for removal.
+    """
+    if not args.images:
+        return True
+
+    return all(regex_match(pattern, tag)
+               for pattern in args.images
+               for tag in image_tags)
+
+
+def regex_match(pattern, tag):
+    """
+    Perform a regex match on the tag.
+    """
+    if pattern[0] == '~':
+        return not re.search(pattern[1:], tag)
+    return re.search(pattern, tag)

dockerrotate/images.py

@@ -0,0 +1,63 @@
+from collections import defaultdict
+
+from docker.errors import APIError
+
+from dockerrotate.filter import include_image
+
+
+def clean_images(args):
+    """
+    Delete old images keeping the most recent N images by tag.
+    """
+    # should not need to inspect all images; only intermediate images should appear
+    # when all is true; these should be deleted along with dependent images
+    images = [image
+              for image in args.client.images(all=False)
+              if include_image(image["RepoTags"], args)]
+
+    # index by id
+    images_by_id = {
+        image["Id"]: image for image in images
+    }
+
+    # group by name
+    images_by_name = defaultdict(set)
+    for image in images:
+        for tag in image["RepoTags"]:
+            image_name = normalize_tag_name(tag)
+            images_by_name[image_name].add(image["Id"])
+
+    for image_name, image_ids in images_by_name.items():
+        # sort/keep
+        images_to_delete = sorted([
+            images_by_id[image_id] for image_id in image_ids],
+            key=lambda image: -image["Created"],
+        )[args.keep:]
+
+        # delete
+        for image in images_to_delete:
+            print "Removing image ID: {}, Tags: {}".format(
+                image["Id"],
+                ", ".join(image["RepoTags"])
+            )
+
+            if args.dry_run:
+                continue
+
+            try:
+                args.client.remove_image(image["Id"], force=True, noprune=False)
+            except APIError as error:
+                print error.message
+
+
+def normalize_tag_name(tag):
+    """
+    docker-py provides image names with tags as a single string.
+
+    We want:
+
+       some.domain.com/organization/image:tag -> organization/image
+                       organization/image:tag -> organization/image
+                                    image:tag ->              image
+    """
+    return "/".join(tag.rsplit(":", 1)[0].split("/")[-2:])