Merge branch '6844-py-mem-leak' closes #6844
[arvados.git] / sdk / python / arvados / commands / keepdocker.py
index c04885c44da3bec86ef9e0d0ddc005f37eaf29ea..e48a6d15472cc2c90cd4af7e35251b653aa87cce 100644 (file)
@@ -1,6 +1,8 @@
 #!/usr/bin/env python
 
 import argparse
+import collections
+import datetime
 import errno
 import json
 import os
@@ -8,42 +10,49 @@ import subprocess
 import sys
 import tarfile
 import tempfile
+import _strptime
 
-from collections import namedtuple
+from operator import itemgetter
 from stat import *
 
 import arvados
+import arvados.util
 import arvados.commands._util as arv_cmd
 import arvados.commands.put as arv_put
+import ciso8601
 
+EARLIEST_DATETIME = datetime.datetime(datetime.MINYEAR, 1, 1, 0, 0, 0)
 STAT_CACHE_ERRORS = (IOError, OSError, ValueError)
 
-DockerImage = namedtuple('DockerImage',
-                         ['repo', 'tag', 'hash', 'created', 'vsize'])
+DockerImage = collections.namedtuple(
+    'DockerImage', ['repo', 'tag', 'hash', 'created', 'vsize'])
 
-opt_parser = argparse.ArgumentParser(add_help=False)
-opt_parser.add_argument(
+keepdocker_parser = argparse.ArgumentParser(add_help=False)
+keepdocker_parser.add_argument(
     '-f', '--force', action='store_true', default=False,
     help="Re-upload the image even if it already exists on the server")
 
-_group = opt_parser.add_mutually_exclusive_group()
+_group = keepdocker_parser.add_mutually_exclusive_group()
 _group.add_argument(
-    '--pull', action='store_true', default=True,
-    help="Pull the latest image from Docker repositories first (default)")
+    '--pull', action='store_true', default=False,
+    help="Try to pull the latest image from Docker registry")
 _group.add_argument(
     '--no-pull', action='store_false', dest='pull',
-    help="Don't pull images from Docker repositories")
+    help="Use locally installed image only, don't pull image from Docker registry (default)")
 
-opt_parser.add_argument(
-    'image',
+keepdocker_parser.add_argument(
+    'image', nargs='?',
     help="Docker image to upload, as a repository name or hash")
-opt_parser.add_argument(
+keepdocker_parser.add_argument(
     'tag', nargs='?', default='latest',
     help="Tag of the Docker image to upload (default 'latest')")
 
+# Combine keepdocker options listed above with run_opts options of arv-put.
+# The options inherited from arv-put include --name, --project-uuid,
+# --progress/--no-progress/--batch-progress and --resume/--no-resume.
 arg_parser = argparse.ArgumentParser(
-        description="Upload a Docker image to Arvados",
-        parents=[opt_parser, arv_put.run_opts])
+        description="Upload or list Docker images in Arvados",
+        parents=[keepdocker_parser, arv_put.run_opts, arv_cmd.retry_opt])
 
 class DockerError(Exception):
     pass
@@ -82,29 +91,38 @@ def docker_images():
     list_proc.stdout.close()
     check_docker(list_proc, "images")
 
-def find_image_hash(image_search, image_tag=None):
-    # Given one argument, search for one Docker image with a matching hash,
-    # and return its full hash.
-    # Given two arguments, also search for a Docker image with the same
-    # repository and tag.  If one is found, return its hash; otherwise,
-    # fall back to the one-argument hash search.
+def find_image_hashes(image_search, image_tag=None):
+    # Given one argument, search for Docker images with matching hashes,
+    # and return their full hashes in a set.
+    # Given two arguments, also search for a Docker image with the
+    # same repository and tag.  If one is found, return its hash in a
+    # set; otherwise, fall back to the one-argument hash search.
     # Returns None if no match is found, or a hash search is ambiguous.
     hash_search = image_search.lower()
     hash_matches = set()
     for image in docker_images():
         if (image.repo == image_search) and (image.tag == image_tag):
-            return image.hash
+            return set([image.hash])
         elif image.hash.startswith(hash_search):
             hash_matches.add(image.hash)
-    if len(hash_matches) == 1:
-        return hash_matches.pop()
-    return None
+    return hash_matches
+
+def find_one_image_hash(image_search, image_tag=None):
+    hashes = find_image_hashes(image_search, image_tag)
+    hash_count = len(hashes)
+    if hash_count == 1:
+        return hashes.pop()
+    elif hash_count == 0:
+        raise DockerError("no matching image found")
+    else:
+        raise DockerError("{} images match {}".format(hash_count, image_search))
 
 def stat_cache_name(image_file):
     return getattr(image_file, 'name', image_file) + '.stat'
 
 def pull_image(image_name, image_tag):
-    check_docker(popen_docker(['pull', '-t', image_tag, image_name]), "pull")
+    check_docker(popen_docker(['pull', '{}:{}'.format(image_name, image_tag)]),
+                 "pull")
 
 def save_image(image_hash, image_file):
     # Save the specified Docker image to image_file, then try to save its
@@ -140,34 +158,226 @@ def prep_image_file(filename):
         image_file = open(file_path, 'w+b' if need_save else 'rb')
     return image_file, need_save
 
-def make_link(link_class, link_name, **link_attrs):
+def make_link(api_client, num_retries, link_class, link_name, **link_attrs):
     link_attrs.update({'link_class': link_class, 'name': link_name})
-    return arvados.api('v1').links().create(body=link_attrs).execute()
+    return api_client.links().create(body=link_attrs).execute(
+        num_retries=num_retries)
+
+def docker_link_sort_key(link):
+    """Build a sort key to find the latest available Docker image.
+
+    To find one source collection for a Docker image referenced by
+    name or image id, the API server looks for a link with the most
+    recent `image_timestamp` property; then the most recent
+    `created_at` timestamp.  This method generates a sort key for
+    Docker metadata links to sort them from least to most preferred.
+    """
+    try:
+        image_timestamp = ciso8601.parse_datetime_unaware(
+            link['properties']['image_timestamp'])
+    except (KeyError, ValueError):
+        image_timestamp = EARLIEST_DATETIME
+    return (image_timestamp,
+            ciso8601.parse_datetime_unaware(link['created_at']))
+
+def _get_docker_links(api_client, num_retries, **kwargs):
+    links = arvados.util.list_all(api_client.links().list,
+                                  num_retries, **kwargs)
+    for link in links:
+        link['_sort_key'] = docker_link_sort_key(link)
+    links.sort(key=itemgetter('_sort_key'), reverse=True)
+    return links
+
+def _new_image_listing(link, dockerhash, repo='<none>', tag='<none>'):
+    timestamp_index = 1 if (link['_sort_key'][0] is EARLIEST_DATETIME) else 0
+    return {
+        '_sort_key': link['_sort_key'],
+        'timestamp': link['_sort_key'][timestamp_index],
+        'collection': link['head_uuid'],
+        'dockerhash': dockerhash,
+        'repo': repo,
+        'tag': tag,
+        }
+
+def list_images_in_arv(api_client, num_retries, image_name=None, image_tag=None):
+    """List all Docker images known to the api_client with image_name and
+    image_tag.  If no image_name is given, defaults to listing all
+    Docker images.
+
+    Returns a list of tuples representing matching Docker images,
+    sorted in preference order (i.e. the first collection in the list
+    is the one that the API server would use). Each tuple is a
+    (collection_uuid, collection_info) pair, where collection_info is
+    a dict with fields "dockerhash", "repo", "tag", and "timestamp".
+
+    """
+    search_filters = []
+    repo_links = None
+    hash_links = None
+    if image_name:
+        # Find images with the name the user specified.
+        search_links = _get_docker_links(
+            api_client, num_retries,
+            filters=[['link_class', '=', 'docker_image_repo+tag'],
+                     ['name', '=',
+                      '{}:{}'.format(image_name, image_tag or 'latest')]])
+        if search_links:
+            repo_links = search_links
+        else:
+            # Fall back to finding images with the specified image hash.
+            search_links = _get_docker_links(
+                api_client, num_retries,
+                filters=[['link_class', '=', 'docker_image_hash'],
+                         ['name', 'ilike', image_name + '%']])
+            hash_links = search_links
+        # Only list information about images that were found in the search.
+        search_filters.append(['head_uuid', 'in',
+                               [link['head_uuid'] for link in search_links]])
+
+    # It should be reasonable to expect that each collection only has one
+    # image hash (though there may be many links specifying this).  Find
+    # the API server's most preferred image hash link for each collection.
+    if hash_links is None:
+        hash_links = _get_docker_links(
+            api_client, num_retries,
+            filters=search_filters + [['link_class', '=', 'docker_image_hash']])
+    hash_link_map = {link['head_uuid']: link for link in reversed(hash_links)}
+
+    # Each collection may have more than one name (though again, one name
+    # may be specified more than once).  Build an image listing from name
+    # tags, sorted by API server preference.
+    if repo_links is None:
+        repo_links = _get_docker_links(
+            api_client, num_retries,
+            filters=search_filters + [['link_class', '=',
+                                       'docker_image_repo+tag']])
+    seen_image_names = collections.defaultdict(set)
+    images = []
+    for link in repo_links:
+        collection_uuid = link['head_uuid']
+        if link['name'] in seen_image_names[collection_uuid]:
+            continue
+        seen_image_names[collection_uuid].add(link['name'])
+        try:
+            dockerhash = hash_link_map[collection_uuid]['name']
+        except KeyError:
+            dockerhash = '<unknown>'
+        name_parts = link['name'].split(':', 1)
+        images.append(_new_image_listing(link, dockerhash, *name_parts))
+
+    # Find any image hash links that did not have a corresponding name link,
+    # and add image listings for them, retaining the API server preference
+    # sorting.
+    images_start_size = len(images)
+    for collection_uuid, link in hash_link_map.iteritems():
+        if not seen_image_names[collection_uuid]:
+            images.append(_new_image_listing(link, link['name']))
+    if len(images) > images_start_size:
+        images.sort(key=itemgetter('_sort_key'), reverse=True)
+
+    # Remove any image listings that refer to unknown collections.
+    existing_coll_uuids = {coll['uuid'] for coll in arvados.util.list_all(
+            api_client.collections().list, num_retries,
+            filters=[['uuid', 'in', [im['collection'] for im in images]]],
+            select=['uuid'])}
+    return [(image['collection'], image) for image in images
+            if image['collection'] in existing_coll_uuids]
 
 def main(arguments=None):
     args = arg_parser.parse_args(arguments)
+    api = arvados.api('v1')
+
+    if args.image is None or args.image == 'images':
+        fmt = "{:30}  {:10}  {:12}  {:29}  {:20}"
+        print fmt.format("REPOSITORY", "TAG", "IMAGE ID", "COLLECTION", "CREATED")
+        for i, j in list_images_in_arv(api, args.retries):
+            print(fmt.format(j["repo"], j["tag"], j["dockerhash"][0:12], i, j["timestamp"].strftime("%c")))
+        sys.exit(0)
 
     # Pull the image if requested, unless the image is specified as a hash
     # that we already have.
-    if args.pull and (find_image_hash(args.image) is None):
+    if args.pull and not find_image_hashes(args.image):
         pull_image(args.image, args.tag)
 
-    image_hash = find_image_hash(args.image, args.tag)
-    if image_hash is None:
-        print >>sys.stderr, "arv-keepdocker: No image found."
+    try:
+        image_hash = find_one_image_hash(args.image, args.tag)
+    except DockerError as error:
+        print >>sys.stderr, "arv-keepdocker:", error.message
         sys.exit(1)
-    elif not args.force:
-        # Abort if this image is already in Arvados.
-        existing_links = arvados.api('v1').links().list(
+
+    image_repo_tag = '{}:{}'.format(args.image, args.tag) if not image_hash.startswith(args.image.lower()) else None
+
+    if args.name is None:
+        if image_repo_tag:
+            collection_name = 'Docker image {} {}'.format(image_repo_tag, image_hash[0:12])
+        else:
+            collection_name = 'Docker image {}'.format(image_hash[0:12])
+    else:
+        collection_name = args.name
+
+    if not args.force:
+        # Check if this image is already in Arvados.
+
+        # Project where everything should be owned
+        if args.project_uuid:
+            parent_project_uuid = args.project_uuid
+        else:
+            parent_project_uuid = api.users().current().execute(
+                num_retries=args.retries)['uuid']
+
+        # Find image hash tags
+        existing_links = api.links().list(
             filters=[['link_class', '=', 'docker_image_hash'],
-                     ['name', '=', image_hash]]).execute()['items']
+                     ['name', '=', image_hash]]
+            ).execute(num_retries=args.retries)['items']
         if existing_links:
-            message = [
-                "arv-keepdocker: Image {} already stored in collection(s):".
-                format(image_hash)]
-            message.extend(link['head_uuid'] for link in existing_links)
-            print >>sys.stderr, "\n".join(message)
-            sys.exit(0)
+            # get readable collections
+            collections = api.collections().list(
+                filters=[['uuid', 'in', [link['head_uuid'] for link in existing_links]]],
+                select=["uuid", "owner_uuid", "name", "manifest_text"]
+                ).execute(num_retries=args.retries)['items']
+
+            if collections:
+                # check for repo+tag links on these collections
+                existing_repo_tag = (api.links().list(
+                    filters=[['link_class', '=', 'docker_image_repo+tag'],
+                             ['name', '=', image_repo_tag],
+                             ['head_uuid', 'in', collections]]
+                    ).execute(num_retries=args.retries)['items']) if image_repo_tag else []
+
+                # Filter on elements owned by the parent project
+                owned_col = [c for c in collections if c['owner_uuid'] == parent_project_uuid]
+                owned_img = [c for c in existing_links if c['owner_uuid'] == parent_project_uuid]
+                owned_rep = [c for c in existing_repo_tag if c['owner_uuid'] == parent_project_uuid]
+
+                if owned_col:
+                    # already have a collection owned by this project
+                    coll_uuid = owned_col[0]['uuid']
+                else:
+                    # create new collection owned by the project
+                    coll_uuid = api.collections().create(
+                        body={"manifest_text": collections[0]['manifest_text'],
+                              "name": collection_name,
+                              "owner_uuid": parent_project_uuid},
+                        ensure_unique_name=True
+                        ).execute(num_retries=args.retries)['uuid']
+
+                link_base = {'owner_uuid': parent_project_uuid,
+                             'head_uuid':  coll_uuid }
+
+                if not owned_img:
+                    # create image link owned by the project
+                    make_link(api, args.retries,
+                              'docker_image_hash', image_hash, **link_base)
+
+                if not owned_rep and image_repo_tag:
+                    # create repo+tag link owned by the project
+                    make_link(api, args.retries, 'docker_image_repo+tag',
+                              image_repo_tag, **link_base)
+
+                print(coll_uuid)
+
+                sys.exit(0)
 
     # Open a file for the saved image, and write it if needed.
     outfile_name = '{}.tar'.format(image_hash)
@@ -177,7 +387,11 @@ def main(arguments=None):
 
     # Call arv-put with switches we inherited from it
     # (a.k.a., switches that aren't our own).
-    put_args = opt_parser.parse_known_args(arguments)[1]
+    put_args = keepdocker_parser.parse_known_args(arguments)[1]
+
+    if args.name is None:
+        put_args += ['--name', collection_name]
+
     coll_uuid = arv_put.main(
         put_args + ['--filename', outfile_name, image_file.name]).strip()
 
@@ -191,11 +405,13 @@ def main(arguments=None):
     link_base = {'head_uuid': coll_uuid, 'properties': {}}
     if 'created' in image_metadata:
         link_base['properties']['image_timestamp'] = image_metadata['created']
+    if args.project_uuid is not None:
+        link_base['owner_uuid'] = args.project_uuid
 
-    make_link('docker_image_hash', image_hash, **link_base)
-    if not image_hash.startswith(args.image.lower()):
-        make_link('docker_image_repository', args.image, **link_base)
-        make_link('docker_image_tag', args.tag, **link_base)
+    make_link(api, args.retries, 'docker_image_hash', image_hash, **link_base)
+    if image_repo_tag:
+        make_link(api, args.retries,
+                  'docker_image_repo+tag', image_repo_tag, **link_base)
 
     # Clean up.
     image_file.close()