X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/da766e40d5119f23689a3d1859fd1473b4d7ad5f..97e40209600c094eac15085627e49da52ab0f517:/sdk/python/arvados/commands/arv_copy.py?ds=sidebyside diff --git a/sdk/python/arvados/commands/arv_copy.py b/sdk/python/arvados/commands/arv_copy.py index 10fe9d7024..6c7d873af4 100755 --- a/sdk/python/arvados/commands/arv_copy.py +++ b/sdk/python/arvados/commands/arv_copy.py @@ -36,6 +36,9 @@ import logging import tempfile import urllib.parse import io +import json +import queue +import threading import arvados import arvados.config @@ -43,9 +46,9 @@ import arvados.keep import arvados.util import arvados.commands._util as arv_cmd import arvados.commands.keepdocker +import arvados.http_to_keep import ruamel.yaml as yaml -from arvados.api import OrderedJsonModel from arvados._version import __version__ COMMIT_HASH_RE = re.compile(r'^[0-9a-f]{1,40}$') @@ -106,6 +109,11 @@ def main(): copy_opts.add_argument( '--storage-classes', dest='storage_classes', help='Comma separated list of storage classes to be used when saving data to the destinaton Arvados instance.') + copy_opts.add_argument("--varying-url-params", type=str, default="", + help="A comma separated list of URL query parameters that should be ignored when storing HTTP URLs in Keep.") + + copy_opts.add_argument("--prefer-cached-downloads", action="store_true", default=False, + help="If a HTTP URL is found in Keep, skip upstream URL freshness check (will not notice if the upstream has changed, but also not error if upstream is unavailable).") copy_opts.add_argument( 'object_uuid', @@ -126,7 +134,7 @@ def main(): else: logger.setLevel(logging.INFO) - if not args.source_arvados: + if not args.source_arvados and arvados.util.uuid_pattern.match(args.object_uuid): args.source_arvados = args.object_uuid[:5] # Create API clients for the source and destination instances @@ -138,28 +146,39 @@ def main(): # Identify the kind of object we have been given, and begin copying. t = uuid_type(src_arv, args.object_uuid) - if t == 'Collection': - set_src_owner_uuid(src_arv.collections(), args.object_uuid, args) - result = copy_collection(args.object_uuid, - src_arv, dst_arv, - args) - elif t == 'Workflow': - set_src_owner_uuid(src_arv.workflows(), args.object_uuid, args) - result = copy_workflow(args.object_uuid, src_arv, dst_arv, args) - elif t == 'Group': - set_src_owner_uuid(src_arv.groups(), args.object_uuid, args) - result = copy_project(args.object_uuid, src_arv, dst_arv, args.project_uuid, args) - else: - abort("cannot copy object {} of type {}".format(args.object_uuid, t)) + + try: + if t == 'Collection': + set_src_owner_uuid(src_arv.collections(), args.object_uuid, args) + result = copy_collection(args.object_uuid, + src_arv, dst_arv, + args) + elif t == 'Workflow': + set_src_owner_uuid(src_arv.workflows(), args.object_uuid, args) + result = copy_workflow(args.object_uuid, src_arv, dst_arv, args) + elif t == 'Group': + set_src_owner_uuid(src_arv.groups(), args.object_uuid, args) + result = copy_project(args.object_uuid, src_arv, dst_arv, args.project_uuid, args) + elif t == 'httpURL': + result = copy_from_http(args.object_uuid, src_arv, dst_arv, args) + else: + abort("cannot copy object {} of type {}".format(args.object_uuid, t)) + except Exception as e: + logger.error("%s", e, exc_info=args.verbose) + exit(1) # Clean up any outstanding temp git repositories. for d in listvalues(local_repo_dir): shutil.rmtree(d, ignore_errors=True) + if not result: + exit(1) + # If no exception was thrown and the response does not have an # error_token field, presume success - if 'error_token' in result or 'uuid' not in result: - logger.error("API server returned an error result: {}".format(result)) + if result is None or 'error_token' in result or 'uuid' not in result: + if result: + logger.error("API server returned an error result: {}".format(result)) exit(1) print(result['uuid']) @@ -191,7 +210,7 @@ def set_src_owner_uuid(resource, uuid, args): def api_for_instance(instance_name, num_retries): if not instance_name: # Use environment - return arvados.api('v1', model=OrderedJsonModel()) + return arvados.api('v1') if '/' in instance_name: config_file = instance_name @@ -215,7 +234,6 @@ def api_for_instance(instance_name, num_retries): host=cfg['ARVADOS_API_HOST'], token=cfg['ARVADOS_API_TOKEN'], insecure=api_is_insecure, - model=OrderedJsonModel(), num_retries=num_retries, ) else: @@ -309,21 +327,20 @@ def copy_workflow(wf_uuid, src, dst, args): # copy collections and docker images if args.recursive and wf["definition"]: - wf_def = yaml.safe_load(wf["definition"]) - if wf_def is not None: - locations = [] - docker_images = {} - graph = wf_def.get('$graph', None) - if graph is not None: - workflow_collections(graph, locations, docker_images) - else: - workflow_collections(wf_def, locations, docker_images) + env = {"ARVADOS_API_HOST": urllib.parse.urlparse(src._rootDesc["rootUrl"]).netloc, + "ARVADOS_API_TOKEN": src.api_token, + "PATH": os.environ["PATH"]} + try: + result = subprocess.run(["arvados-cwl-runner", "--quiet", "--print-keep-deps", "arvwf:"+wf_uuid], + capture_output=True, env=env) + except (FileNotFoundError, subprocess.CalledProcessError): + logger.error('Copying workflows requires arvados-cwl-runner 2.7.1 or later to be installed in PATH.') + return - if locations: - copy_collections(locations, src, dst, args) + locations = json.loads(result.stdout) - for image in docker_images: - copy_docker_image(image, docker_images[image], src, dst, args) + if locations: + copy_collections(locations, src, dst, args) # copy the workflow itself del wf['uuid'] @@ -567,6 +584,125 @@ def copy_collection(obj_uuid, src, dst, args): else: progress_writer = None + # go through the words + # put each block loc into 'get' queue + # 'get' threads get block and put it into 'put' queue + # 'put' threads put block and then update dst_locators + # + # after going through the whole manifest we go back through it + # again and build dst_manifest + + lock = threading.Lock() + + # the get queue should be unbounded because we'll add all the + # block hashes we want to get, but these are small + get_queue = queue.Queue() + + threadcount = 4 + + # the put queue contains full data blocks + # and if 'get' is faster than 'put' we could end up consuming + # a great deal of RAM if it isn't bounded. + put_queue = queue.Queue(threadcount) + transfer_error = [] + + def get_thread(): + while True: + word = get_queue.get() + if word is None: + put_queue.put(None) + get_queue.task_done() + return + + blockhash = arvados.KeepLocator(word).md5sum + with lock: + if blockhash in dst_locators: + # Already uploaded + get_queue.task_done() + continue + + try: + logger.debug("Getting block %s", word) + data = src_keep.get(word) + put_queue.put((word, data)) + except e: + logger.error("Error getting block %s: %s", word, e) + transfer_error.append(e) + try: + # Drain the 'get' queue so we end early + while True: + get_queue.get(False) + get_queue.task_done() + except queue.Empty: + pass + finally: + get_queue.task_done() + + def put_thread(): + nonlocal bytes_written + while True: + item = put_queue.get() + if item is None: + put_queue.task_done() + return + + word, data = item + loc = arvados.KeepLocator(word) + blockhash = loc.md5sum + with lock: + if blockhash in dst_locators: + # Already uploaded + put_queue.task_done() + continue + + try: + logger.debug("Putting block %s (%s bytes)", blockhash, loc.size) + dst_locator = dst_keep.put(data, classes=(args.storage_classes or [])) + with lock: + dst_locators[blockhash] = dst_locator + bytes_written += loc.size + if progress_writer: + progress_writer.report(obj_uuid, bytes_written, bytes_expected) + except e: + logger.error("Error putting block %s (%s bytes): %s", blockhash, loc.size, e) + try: + # Drain the 'get' queue so we end early + while True: + get_queue.get(False) + get_queue.task_done() + except queue.Empty: + pass + transfer_error.append(e) + finally: + put_queue.task_done() + + for line in manifest.splitlines(): + words = line.split() + for word in words[1:]: + try: + loc = arvados.KeepLocator(word) + except ValueError: + # If 'word' can't be parsed as a locator, + # presume it's a filename. + continue + + get_queue.put(word) + + for i in range(0, threadcount): + get_queue.put(None) + + for i in range(0, threadcount): + threading.Thread(target=get_thread, daemon=True).start() + + for i in range(0, threadcount): + threading.Thread(target=put_thread, daemon=True).start() + + get_queue.join() + put_queue.join() + + if len(transfer_error) > 0: + return {"error_token": "Failed to transfer blocks"} + for line in manifest.splitlines(): words = line.split() dst_manifest.write(words[0]) @@ -580,16 +716,6 @@ def copy_collection(obj_uuid, src, dst, args): dst_manifest.write(word) continue blockhash = loc.md5sum - # copy this block if we haven't seen it before - # (otherwise, just reuse the existing dst_locator) - if blockhash not in dst_locators: - logger.debug("Copying block %s (%s bytes)", blockhash, loc.size) - if progress_writer: - progress_writer.report(obj_uuid, bytes_written, bytes_expected) - data = src_keep.get(word) - dst_locator = dst_keep.put(data, classes=(args.storage_classes or [])) - dst_locators[blockhash] = dst_locator - bytes_written += loc.size dst_manifest.write(' ') dst_manifest.write(dst_locators[blockhash]) dst_manifest.write("\n") @@ -758,6 +884,10 @@ def git_rev_parse(rev, repo): def uuid_type(api, object_uuid): if re.match(arvados.util.keep_locator_pattern, object_uuid): return 'Collection' + + if object_uuid.startswith("http:") or object_uuid.startswith("https:"): + return 'httpURL' + p = object_uuid.split('-') if len(p) == 3: type_prefix = p[1] @@ -767,6 +897,27 @@ def uuid_type(api, object_uuid): return k return None + +def copy_from_http(url, src, dst, args): + + project_uuid = args.project_uuid + varying_url_params = args.varying_url_params + prefer_cached_downloads = args.prefer_cached_downloads + + cached = arvados.http_to_keep.check_cached_url(src, project_uuid, url, {}, + varying_url_params=varying_url_params, + prefer_cached_downloads=prefer_cached_downloads) + if cached[2] is not None: + return copy_collection(cached[2], src, dst, args) + + cached = arvados.http_to_keep.http_to_keep(dst, project_uuid, url, + varying_url_params=varying_url_params, + prefer_cached_downloads=prefer_cached_downloads) + + if cached is not None: + return {"uuid": cached[2]} + + def abort(msg, code=1): logger.info("arv-copy: %s", msg) exit(code)