X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/6ac59725fedc0a958d60d85740b47deb25addb5b..cf4c25824e15e3b9a7c1439ab3aff17d8be52630:/sdk/python/arvados/commands/run.py diff --git a/sdk/python/arvados/commands/run.py b/sdk/python/arvados/commands/run.py index f2bf0f353b..1e64eeb1da 100644 --- a/sdk/python/arvados/commands/run.py +++ b/sdk/python/arvados/commands/run.py @@ -1,5 +1,25 @@ -#!/usr/bin/env python - +# Copyright (C) The Arvados Authors. All rights reserved. +# Copyright (C) 2018 Genome Research Ltd. +# +# SPDX-License-Identifier: Apache-2.0 +# +# Licensed under the Apache License, Version 2.0 (the "License"); +# you may not use this file except in compliance with the License. +# You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, software +# distributed under the License is distributed on an "AS IS" BASIS, +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +# See the License for the specific language governing permissions and +# limitations under the License. + +from __future__ import print_function +from __future__ import absolute_import +from builtins import range +from past.builtins import basestring +from builtins import object import arvados import arvados.commands.ws as ws import argparse @@ -7,32 +27,32 @@ import json import re import os import stat -import put +from . import put import time import subprocess import logging +import sys +import errno import arvados.commands._util as arv_cmd +import arvados.collection +import arvados.config as config -logger = logging.getLogger('arvados.arv-run') +from arvados._version import __version__ -arvrun_parser = argparse.ArgumentParser(parents=[arv_cmd.retry_opt]) -arvrun_parser.add_argument('--dry-run', action="store_true", help="Print out the pipeline that would be submitted and exit") -arvrun_parser.add_argument('--local', action="store_true", help="Run locally using arv-run-pipeline-instance") -arvrun_parser.add_argument('--docker-image', type=str, default="arvados/jobs", help="Docker image to use, default arvados/jobs") -arvrun_parser.add_argument('--ignore-rcode', action="store_true", help="Commands that return non-zero return codes should not be considered failed.") -arvrun_parser.add_argument('--no-reuse', action="store_true", help="Do not reuse past jobs.") -arvrun_parser.add_argument('--no-wait', action="store_true", help="Do not wait and display logs after submitting command, just exit.") -arvrun_parser.add_argument('--project-uuid', type=str, help="Parent project of the pipeline") -arvrun_parser.add_argument('--git-dir', type=str, default="", help="Git repository passed to arv-crunch-job when using --local") -arvrun_parser.add_argument('--repository', type=str, default="arvados", help="repository field of component, default 'arvados'") -arvrun_parser.add_argument('--script-version', type=str, default="master", help="script_version field of component, default 'master'") -arvrun_parser.add_argument('args', nargs=argparse.REMAINDER) +logger = logging.getLogger('arvados.arv-run') +logger.setLevel(logging.INFO) class ArvFile(object): def __init__(self, prefix, fn): self.prefix = prefix self.fn = fn + def __hash__(self): + return (self.prefix+self.fn).__hash__() + + def __eq__(self, other): + return (self.prefix == other.prefix) and (self.fn == other.fn) + class UploadFile(ArvFile): pass @@ -51,7 +71,7 @@ def is_in_collection(root, branch): else: sp = os.path.split(root) return is_in_collection(sp[0], os.path.join(sp[1], branch)) - except IOError, OSError: + except (IOError, OSError): return (None, None) # Determine the project to place the output of this command by searching upward @@ -72,7 +92,7 @@ def determine_project(root, current_user): else: sp = os.path.split(root) return determine_project(sp[0], current_user) - except IOError, OSError: + except (IOError, OSError): return current_user # Determine if string corresponds to a file, and if that file is part of a @@ -80,111 +100,51 @@ def determine_project(root, current_user): # ArvFile() (file already exists in a collection), UploadFile() (file needs to # be uploaded to a collection), or simply returns prefix+fn (which yields the # original parameter string). -def statfile(prefix, fn): +def statfile(prefix, fn, fnPattern="$(file %s/%s)", dirPattern="$(dir %s/%s/)", raiseOSError=False): absfn = os.path.abspath(fn) - if os.path.exists(absfn): + try: st = os.stat(absfn) - if stat.S_ISREG(st.st_mode): - sp = os.path.split(absfn) - (pdh, branch) = is_in_collection(sp[0], sp[1]) - if pdh: - return ArvFile(prefix, "$(file %s/%s)" % (pdh, branch)) + sp = os.path.split(absfn) + (pdh, branch) = is_in_collection(sp[0], sp[1]) + if pdh: + if stat.S_ISREG(st.st_mode): + return ArvFile(prefix, fnPattern % (pdh, branch)) + elif stat.S_ISDIR(st.st_mode): + return ArvFile(prefix, dirPattern % (pdh, branch)) else: - # trim leading '/' for path prefix test later - return UploadFile(prefix, absfn[1:]) - if stat.S_ISDIR(st.st_mode): - sp = os.path.split(absfn) - (pdh, branch) = is_in_collection(sp[0], sp[1]) - if pdh: - return ArvFile(prefix, "$(dir %s/%s/)" % (pdh, branch)) - - return prefix+fn - -def main(arguments=None): - args = arvrun_parser.parse_args(arguments) - - if len(args.args) == 0: - arvrun_parser.print_help() - return - - starting_args = args.args - - reading_into = 2 - - # Parse the command arguments into 'slots'. - # All words following '>' are output arguments and are collected into slots[0]. - # All words following '<' are input arguments and are collected into slots[1]. - # slots[2..] store the parameters of each command in the pipeline. - # - # e.g. arv-run foo arg1 arg2 '|' bar arg3 arg4 '<' input1 input2 input3 '>' output.txt - # will be parsed into: - # [['output.txt'], - # ['input1', 'input2', 'input3'], - # ['foo', 'arg1', 'arg2'], - # ['bar', 'arg3', 'arg4']] - slots = [[], [], []] - for c in args.args: - if c.startswith('>'): - reading_into = 0 - if len(c) > 1: - slots[reading_into].append(c[1:]) - elif c.startswith('<'): - reading_into = 1 - if len(c) > 1: - slots[reading_into].append(c[1:]) - elif c == '|': - reading_into = len(slots) - slots.append([]) + raise Exception("%s is not a regular file or directory" % absfn) else: - slots[reading_into].append(c) - - if slots[0] and len(slots[0]) > 1: - logger.error("Can only specify a single stdout file (run-command substitutions are permitted)") - return - - if not args.dry_run: - api = arvados.api('v1') - if args.project_uuid: - project = args.project_uuid + # trim leading '/' for path prefix test later + return UploadFile(prefix, absfn[1:]) + except OSError as e: + if e.errno == errno.ENOENT and not raiseOSError: + pass else: - project = determine_project(os.getcwd(), api.users().current().execute()["uuid"]) - - # Identify input files. Look at each parameter and test to see if there is - # a file by that name. This uses 'patterns' to look for within - # command line arguments, such as --foo=file.txt or -lfile.txt - patterns = [re.compile("([^=]+=)(.*)"), - re.compile("(-[A-Za-z])(.+)")] - for j, command in enumerate(slots[1:]): - for i, a in enumerate(command): - if j > 0 and i == 0: - # j == 0 is stdin, j > 0 is commands - # always skip program executable (i == 0) in commands - pass - elif a.startswith('\\'): - # if it starts with a \ then don't do any interpretation - command[i] = a[1:] - else: - # See if it looks like a file - command[i] = statfile('', a) + raise - # If a file named command[i] was found, it would now be an - # ArvFile or UploadFile. If command[i] is a basestring, that - # means it doesn't correspond exactly to a file, so do some - # pattern matching. - if isinstance(command[i], basestring): - for p in patterns: - m = p.match(a) - if m: - command[i] = statfile(m.group(1), m.group(2)) - break + return prefix+fn - n = True - pathprefix = "/" - files = [c for command in slots[1:] for c in command if isinstance(c, UploadFile)] - if len(files) > 0: - # Find the smallest path prefix that includes all the files that need to be uploaded. - # This starts at the root and iteratively removes common parent directory prefixes - # until all file pathes no longer have a common parent. +def write_file(collection, pathprefix, fn, flush=False): + with open(os.path.join(pathprefix, fn), "rb") as src: + dst = collection.open(fn, "wb") + r = src.read(1024*128) + while r: + dst.write(r) + r = src.read(1024*128) + dst.close(flush=flush) + +def uploadfiles(files, api, dry_run=False, num_retries=0, + project=None, + fnPattern="$(file %s/%s)", + name=None, + collection=None, + packed=True): + # Find the smallest path prefix that includes all the files that need to be uploaded. + # This starts at the root and iteratively removes common parent directory prefixes + # until all file paths no longer have a common parent. + if files: + n = True + pathprefix = "/" while n: pathstep = None for c in files: @@ -208,109 +168,83 @@ def main(arguments=None): for c in files: c.fn = c.fn[len(pathstep):] - orgdir = os.getcwd() - os.chdir(pathprefix) - - print("Upload local files: \"%s\"" % '" "'.join([c.fn for c in files])) - - if args.dry_run: - print("$(input) is %s" % pathprefix.rstrip('/')) - pdh = "$(input)" - else: - files = sorted(files, key=lambda x: x.fn) - collection = arvados.CollectionWriter(api, num_retries=args.retries) - stream = None - for f in files: - sp = os.path.split(f.fn) - if sp[0] != stream: - stream = sp[0] - collection.start_new_stream(stream) - collection.write_file(f.fn, sp[1]) - item = api.collections().create(body={"owner_uuid": project, "manifest_text": collection.manifest_text()}).execute() - pdh = item["portable_data_hash"] - print "Uploaded to %s" % item["uuid"] - - for c in files: - c.fn = "$(file %s/%s)" % (pdh, c.fn) - - os.chdir(orgdir) - - for i in xrange(1, len(slots)): - slots[i] = [("%s%s" % (c.prefix, c.fn)) if isinstance(c, ArvFile) else c for c in slots[i]] - - component = { - "script": "run-command", - "script_version": args.script_version, - "repository": args.repository, - "script_parameters": { - }, - "runtime_constraints": { - "docker_image": args.docker_image - } - } - - task_foreach = [] - group_parser = argparse.ArgumentParser() - group_parser.add_argument('-b', '--batch-size', type=int) - group_parser.add_argument('args', nargs=argparse.REMAINDER) + logger.info("Upload local files: \"%s\"", '" "'.join([c.fn for c in files])) - for s in xrange(2, len(slots)): - for i in xrange(0, len(slots[s])): - if slots[s][i] == '--': - inp = "input%i" % (s-2) - groupargs = group_parser.parse_args(slots[2][i+1:]) - if groupargs.batch_size: - component["script_parameters"][inp] = {"value": {"batch":groupargs.args, "size":groupargs.batch_size}} - slots[s] = slots[s][0:i] + [{"foreach": inp, "command": "$(%s)" % inp}] + if dry_run: + logger.info("$(input) is %s", pathprefix.rstrip('/')) + pdh = "$(input)" + else: + files = sorted(files, key=lambda x: x.fn) + if collection is None: + collection = arvados.collection.Collection(api_client=api, num_retries=num_retries) + prev = "" + for f in files: + localpath = os.path.join(pathprefix, f.fn) + if prev and localpath.startswith(prev+"/"): + # If this path is inside an already uploaded subdirectory, + # don't redundantly re-upload it. + # e.g. we uploaded /tmp/foo and the next file is /tmp/foo/bar + # skip it because it starts with "/tmp/foo/" + continue + prev = localpath + if os.path.isfile(localpath): + write_file(collection, pathprefix, f.fn, not packed) + elif os.path.isdir(localpath): + for root, dirs, iterfiles in os.walk(localpath): + root = root[len(pathprefix):] + for src in iterfiles: + write_file(collection, pathprefix, os.path.join(root, src), not packed) + + pdh = None + if len(collection) > 0: + # non-empty collection + filters = [["portable_data_hash", "=", collection.portable_data_hash()]] + name_pdh = "%s (%s)" % (name, collection.portable_data_hash()) + if name: + filters.append(["name", "=", name_pdh]) + if project: + filters.append(["owner_uuid", "=", project]) + + # do the list / create in a loop with up to 2 tries as we are using `ensure_unique_name=False` + # and there is a potential race with other workflows that may have created the collection + # between when we list it and find it does not exist and when we attempt to create it. + tries = 2 + while pdh is None and tries > 0: + exists = api.collections().list(filters=filters, limit=1).execute(num_retries=num_retries) + + if exists["items"]: + item = exists["items"][0] + pdh = item["portable_data_hash"] + logger.info("Using collection %s (%s)", pdh, item["uuid"]) else: - component["script_parameters"][inp] = groupargs.args - slots[s] = slots[s][0:i] + ["$(%s)" % inp] - task_foreach.append(inp) - break - if slots[s][i] == '\--': - slots[s][i] = '--' - - if slots[0]: - component["script_parameters"]["task.stdout"] = slots[0][0] - if slots[1]: - task_foreach.append("stdin") - component["script_parameters"]["stdin"] = slots[1] - component["script_parameters"]["task.stdin"] = "$(stdin)" - - if task_foreach: - component["script_parameters"]["task.foreach"] = task_foreach - - component["script_parameters"]["command"] = slots[2:] - if args.ignore_rcode: - component["script_parameters"]["task.ignore_rcode"] = args.ignore_rcode - - pipeline = { - "name": "arv-run " + " | ".join([s[0] for s in slots[2:]]), - "description": "@" + " ".join(starting_args) + "@", - "components": { - "command": component - }, - "state": "RunningOnClient" if args.local else "RunningOnServer" - } + try: + collection.save_new(name=name_pdh, owner_uuid=project, ensure_unique_name=False) + pdh = collection.portable_data_hash() + logger.info("Uploaded to %s (%s)", pdh, collection.manifest_locator()) + except arvados.errors.ApiError as ae: + tries -= 1 + if pdh is None: + # Something weird going on here, probably a collection + # with a conflicting name but wrong PDH. We won't + # able to reuse it but we still need to save our + # collection, so so save it with unique name. + logger.info("Name conflict on '%s', existing collection has an unexpected portable data hash", name_pdh) + collection.save_new(name=name_pdh, owner_uuid=project, ensure_unique_name=True) + pdh = collection.portable_data_hash() + logger.info("Uploaded to %s (%s)", pdh, collection.manifest_locator()) + else: + # empty collection + pdh = collection.portable_data_hash() + assert (pdh == config.EMPTY_BLOCK_LOCATOR), "Empty collection portable_data_hash did not have expected locator, was %s" % pdh + logger.debug("Using empty collection %s", pdh) - if args.dry_run: - print(json.dumps(pipeline, indent=4)) - else: - pipeline["owner_uuid"] = project - pi = api.pipeline_instances().create(body=pipeline, ensure_unique_name=True).execute() - print "Running pipeline %s" % pi["uuid"] + for c in files: + c.keepref = "%s/%s" % (pdh, c.fn) + c.fn = fnPattern % (pdh, c.fn) - if args.local: - subprocess.call(["arv-run-pipeline-instance", "--instance", pi["uuid"], "--run-jobs-here"] + (["--no-reuse"] if args.no_reuse else [])) - elif not args.no_wait: - ws.main(["--pipeline", pi["uuid"]]) - pi = api.pipeline_instances().get(uuid=pi["uuid"]).execute() - print "Pipeline is %s" % pi["state"] - if "output_uuid" in pi["components"]["command"]: - print "Output is %s" % pi["components"]["command"]["output_uuid"] - else: - print "No output" +def main(arguments=None): + raise Exception("Legacy arv-run removed.") if __name__ == '__main__': main()