sdk/python/arvados/commands/keepdocker.py

   1 #!/usr/bin/env python
   2
   3 import argparse
   4 import datetime
   5 import errno
   6 import json
   7 import os
   8 import subprocess
   9 import sys
  10 import tarfile
  11 import tempfile
  12 import _strptime
  13
  14 from collections import namedtuple
  15 from stat import *
  16
  17 import arvados
  18 import arvados.commands._util as arv_cmd
  19 import arvados.commands.put as arv_put
  20
  21 STAT_CACHE_ERRORS = (IOError, OSError, ValueError)
  22
  23 DockerImage = namedtuple('DockerImage',
  24                          ['repo', 'tag', 'hash', 'created', 'vsize'])
  25
  26 keepdocker_parser = argparse.ArgumentParser(add_help=False)
  27 keepdocker_parser.add_argument(
  28     '-f', '--force', action='store_true', default=False,
  29     help="Re-upload the image even if it already exists on the server")
  30
  31 _group = keepdocker_parser.add_mutually_exclusive_group()
  32 _group.add_argument(
  33     '--pull', action='store_true', default=False,
  34     help="Try to pull the latest image from Docker registry")
  35 _group.add_argument(
  36     '--no-pull', action='store_false', dest='pull',
  37     help="Use locally installed image only, don't pull image from Docker registry (default)")
  38
  39 keepdocker_parser.add_argument(
  40     'image', nargs='?',
  41     help="Docker image to upload, as a repository name or hash")
  42 keepdocker_parser.add_argument(
  43     'tag', nargs='?', default='latest',
  44     help="Tag of the Docker image to upload (default 'latest')")
  45
  46 # Combine keepdocker options listed above with run_opts options of arv-put.
  47 # The options inherited from arv-put include --name, --project-uuid,
  48 # --progress/--no-progress/--batch-progress and --resume/--no-resume.
  49 arg_parser = argparse.ArgumentParser(
  50         description="Upload or list Docker images in Arvados",
  51         parents=[keepdocker_parser, arv_put.run_opts, arv_cmd.retry_opt])
  52
  53 class DockerError(Exception):
  54     pass
  55
  56
  57 def popen_docker(cmd, *args, **kwargs):
  58     manage_stdin = ('stdin' not in kwargs)
  59     kwargs.setdefault('stdin', subprocess.PIPE)
  60     kwargs.setdefault('stdout', sys.stderr)
  61     try:
  62         docker_proc = subprocess.Popen(['docker.io'] + cmd, *args, **kwargs)
  63     except OSError:  # No docker.io in $PATH
  64         docker_proc = subprocess.Popen(['docker'] + cmd, *args, **kwargs)
  65     if manage_stdin:
  66         docker_proc.stdin.close()
  67     return docker_proc
  68
  69 def check_docker(proc, description):
  70     proc.wait()
  71     if proc.returncode != 0:
  72         raise DockerError("docker {} returned status code {}".
  73                           format(description, proc.returncode))
  74
  75 def docker_images():
  76     # Yield a DockerImage tuple for each installed image.
  77     list_proc = popen_docker(['images', '--no-trunc'], stdout=subprocess.PIPE)
  78     list_output = iter(list_proc.stdout)
  79     next(list_output)  # Ignore the header line
  80     for line in list_output:
  81         words = line.split()
  82         size_index = len(words) - 2
  83         repo, tag, imageid = words[:3]
  84         ctime = ' '.join(words[3:size_index])
  85         vsize = ' '.join(words[size_index:])
  86         yield DockerImage(repo, tag, imageid, ctime, vsize)
  87     list_proc.stdout.close()
  88     check_docker(list_proc, "images")
  89
  90 def find_image_hashes(image_search, image_tag=None):
  91     # Given one argument, search for Docker images with matching hashes,
  92     # and return their full hashes in a set.
  93     # Given two arguments, also search for a Docker image with the
  94     # same repository and tag.  If one is found, return its hash in a
  95     # set; otherwise, fall back to the one-argument hash search.
  96     # Returns None if no match is found, or a hash search is ambiguous.
  97     hash_search = image_search.lower()
  98     hash_matches = set()
  99     for image in docker_images():
 100         if (image.repo == image_search) and (image.tag == image_tag):
 101             return set([image.hash])
 102         elif image.hash.startswith(hash_search):
 103             hash_matches.add(image.hash)
 104     return hash_matches
 105
 106 def find_one_image_hash(image_search, image_tag=None):
 107     hashes = find_image_hashes(image_search, image_tag)
 108     hash_count = len(hashes)
 109     if hash_count == 1:
 110         return hashes.pop()
 111     elif hash_count == 0:
 112         raise DockerError("no matching image found")
 113     else:
 114         raise DockerError("{} images match {}".format(hash_count, image_search))
 115
 116 def stat_cache_name(image_file):
 117     return getattr(image_file, 'name', image_file) + '.stat'
 118
 119 def pull_image(image_name, image_tag):
 120     check_docker(popen_docker(['pull', '{}:{}'.format(image_name, image_tag)]),
 121                  "pull")
 122
 123 def save_image(image_hash, image_file):
 124     # Save the specified Docker image to image_file, then try to save its
 125     # stats so we can try to resume after interruption.
 126     check_docker(popen_docker(['save', image_hash], stdout=image_file),
 127                  "save")
 128     image_file.flush()
 129     try:
 130         with open(stat_cache_name(image_file), 'w') as statfile:
 131             json.dump(tuple(os.fstat(image_file.fileno())), statfile)
 132     except STAT_CACHE_ERRORS:
 133         pass  # We won't resume from this cache.  No big deal.
 134
 135 def prep_image_file(filename):
 136     # Return a file object ready to save a Docker image,
 137     # and a boolean indicating whether or not we need to actually save the
 138     # image (False if a cached save is available).
 139     cache_dir = arv_cmd.make_home_conf_dir(
 140         os.path.join('.cache', 'arvados', 'docker'), 0o700)
 141     if cache_dir is None:
 142         image_file = tempfile.NamedTemporaryFile(suffix='.tar')
 143         need_save = True
 144     else:
 145         file_path = os.path.join(cache_dir, filename)
 146         try:
 147             with open(stat_cache_name(file_path)) as statfile:
 148                 prev_stat = json.load(statfile)
 149             now_stat = os.stat(file_path)
 150             need_save = any(prev_stat[field] != now_stat[field]
 151                             for field in [ST_MTIME, ST_SIZE])
 152         except STAT_CACHE_ERRORS + (AttributeError, IndexError):
 153             need_save = True  # We couldn't compare against old stats
 154         image_file = open(file_path, 'w+b' if need_save else 'rb')
 155     return image_file, need_save
 156
 157 def make_link(api_client, num_retries, link_class, link_name, **link_attrs):
 158     link_attrs.update({'link_class': link_class, 'name': link_name})
 159     return api_client.links().create(body=link_attrs).execute(
 160         num_retries=num_retries)
 161
 162 def ptimestamp(t):
 163     s = t.split(".")
 164     if len(s) == 2:
 165         t = s[0] + s[1][-1:]
 166     return datetime.datetime.strptime(t, "%Y-%m-%dT%H:%M:%SZ")
 167
 168 def list_images_in_arv(api_client, num_retries):
 169     existing_links = api_client.links().list(
 170         filters=[['link_class', 'in', ['docker_image_hash', 'docker_image_repo+tag']]]
 171         ).execute(num_retries=num_retries)['items']
 172     images = {}
 173     for link in existing_links:
 174         collection_uuid = link["head_uuid"]
 175         if collection_uuid not in images:
 176             images[collection_uuid]= {"dockerhash": "<none>",
 177                       "repo":"<none>",
 178                       "tag":"<none>",
 179                       "timestamp": ptimestamp("1970-01-01T00:00:01Z")}
 180
 181         if link["link_class"] == "docker_image_hash":
 182             images[collection_uuid]["dockerhash"] = link["name"]
 183
 184         if link["link_class"] == "docker_image_repo+tag":
 185             r = link["name"].split(":")
 186             images[collection_uuid]["repo"] = r[0]
 187             if len(r) > 1:
 188                 images[collection_uuid]["tag"] = r[1]
 189
 190         if "image_timestamp" in link["properties"]:
 191             images[collection_uuid]["timestamp"] = ptimestamp(link["properties"]["image_timestamp"])
 192         else:
 193             images[collection_uuid]["timestamp"] = ptimestamp(link["created_at"])
 194
 195     st = sorted(images.items(), lambda a, b: cmp(b[1]["timestamp"], a[1]["timestamp"]))
 196
 197     fmt = "{:30}  {:10}  {:12}  {:29}  {:20}"
 198     print fmt.format("REPOSITORY", "TAG", "IMAGE ID", "COLLECTION", "CREATED")
 199     for i, j in st:
 200         print(fmt.format(j["repo"], j["tag"], j["dockerhash"][0:12], i, j["timestamp"].strftime("%c")))
 201
 202 def main(arguments=None):
 203     args = arg_parser.parse_args(arguments)
 204     api = arvados.api('v1')
 205
 206     if args.image is None or args.image == 'images':
 207         list_images_in_arv(api, args.retries)
 208         sys.exit(0)
 209
 210     # Pull the image if requested, unless the image is specified as a hash
 211     # that we already have.
 212     if args.pull and not find_image_hashes(args.image):
 213         pull_image(args.image, args.tag)
 214
 215     try:
 216         image_hash = find_one_image_hash(args.image, args.tag)
 217     except DockerError as error:
 218         print >>sys.stderr, "arv-keepdocker:", error.message
 219         sys.exit(1)
 220
 221     image_repo_tag = '{}:{}'.format(args.image, args.tag) if not image_hash.startswith(args.image.lower()) else None
 222
 223     if args.name is None:
 224         if image_repo_tag:
 225             collection_name = 'Docker image {} {}'.format(image_repo_tag, image_hash[0:12])
 226         else:
 227             collection_name = 'Docker image {}'.format(image_hash[0:12])
 228     else:
 229         collection_name = args.name
 230
 231     if not args.force:
 232         # Check if this image is already in Arvados.
 233
 234         # Project where everything should be owned
 235         if args.project_uuid:
 236             parent_project_uuid = args.project_uuid
 237         else:
 238             parent_project_uuid = api.users().current().execute(
 239                 num_retries=args.retries)['uuid']
 240
 241         # Find image hash tags
 242         existing_links = api.links().list(
 243             filters=[['link_class', '=', 'docker_image_hash'],
 244                      ['name', '=', image_hash]]
 245             ).execute(num_retries=args.retries)['items']
 246         if existing_links:
 247             # get readable collections
 248             collections = api.collections().list(
 249                 filters=[['uuid', 'in', [link['head_uuid'] for link in existing_links]]],
 250                 select=["uuid", "owner_uuid", "name", "manifest_text"]
 251                 ).execute(num_retries=args.retries)['items']
 252
 253             if collections:
 254                 # check for repo+tag links on these collections
 255                 existing_repo_tag = (api.links().list(
 256                     filters=[['link_class', '=', 'docker_image_repo+tag'],
 257                              ['name', '=', image_repo_tag],
 258                              ['head_uuid', 'in', collections]]
 259                     ).execute(num_retries=args.retries)['items']) if image_repo_tag else []
 260
 261                 # Filter on elements owned by the parent project
 262                 owned_col = [c for c in collections if c['owner_uuid'] == parent_project_uuid]
 263                 owned_img = [c for c in existing_links if c['owner_uuid'] == parent_project_uuid]
 264                 owned_rep = [c for c in existing_repo_tag if c['owner_uuid'] == parent_project_uuid]
 265
 266                 if owned_col:
 267                     # already have a collection owned by this project
 268                     coll_uuid = owned_col[0]['uuid']
 269                 else:
 270                     # create new collection owned by the project
 271                     coll_uuid = api.collections().create(
 272                         body={"manifest_text": collections[0]['manifest_text'],
 273                               "name": collection_name,
 274                               "owner_uuid": parent_project_uuid},
 275                         ensure_unique_name=True
 276                         ).execute(num_retries=args.retries)['uuid']
 277
 278                 link_base = {'owner_uuid': parent_project_uuid,
 279                              'head_uuid':  coll_uuid }
 280
 281                 if not owned_img:
 282                     # create image link owned by the project
 283                     make_link(api, args.retries,
 284                               'docker_image_hash', image_hash, **link_base)
 285
 286                 if not owned_rep and image_repo_tag:
 287                     # create repo+tag link owned by the project
 288                     make_link(api, args.retries, 'docker_image_repo+tag',
 289                               image_repo_tag, **link_base)
 290
 291                 print(coll_uuid)
 292
 293                 sys.exit(0)
 294
 295     # Open a file for the saved image, and write it if needed.
 296     outfile_name = '{}.tar'.format(image_hash)
 297     image_file, need_save = prep_image_file(outfile_name)
 298     if need_save:
 299         save_image(image_hash, image_file)
 300
 301     # Call arv-put with switches we inherited from it
 302     # (a.k.a., switches that aren't our own).
 303     put_args = keepdocker_parser.parse_known_args(arguments)[1]
 304
 305     if args.name is None:
 306         put_args += ['--name', collection_name]
 307
 308     coll_uuid = arv_put.main(
 309         put_args + ['--filename', outfile_name, image_file.name]).strip()
 310
 311     # Read the image metadata and make Arvados links from it.
 312     image_file.seek(0)
 313     image_tar = tarfile.open(fileobj=image_file)
 314     json_file = image_tar.extractfile(image_tar.getmember(image_hash + '/json'))
 315     image_metadata = json.load(json_file)
 316     json_file.close()
 317     image_tar.close()
 318     link_base = {'head_uuid': coll_uuid, 'properties': {}}
 319     if 'created' in image_metadata:
 320         link_base['properties']['image_timestamp'] = image_metadata['created']
 321     if args.project_uuid is not None:
 322         link_base['owner_uuid'] = args.project_uuid
 323
 324     make_link(api, args.retries, 'docker_image_hash', image_hash, **link_base)
 325     if image_repo_tag:
 326         make_link(api, args.retries,
 327                   'docker_image_repo+tag', image_repo_tag, **link_base)
 328
 329     # Clean up.
 330     image_file.close()
 331     for filename in [stat_cache_name(image_file), image_file.name]:
 332         try:
 333             os.unlink(filename)
 334         except OSError as error:
 335             if error.errno != errno.ENOENT:
 336                 raise
 337
 338 if __name__ == '__main__':
 339     main()