X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/a0c099f41a00785b6d28a105e49f40e713e78882..efee0049baf53efa3be10804abc9bc292f2a60ed:/sdk/python/arvados/commands/run.py diff --git a/sdk/python/arvados/commands/run.py b/sdk/python/arvados/commands/run.py index 475b2a9ba5..860fc8f805 100644 --- a/sdk/python/arvados/commands/run.py +++ b/sdk/python/arvados/commands/run.py @@ -1,6 +1,7 @@ #!/usr/bin/env python import arvados +import arvados.commands.ws as ws import argparse import json import re @@ -8,23 +9,40 @@ import os import stat import put import time -import arvados.commands.ws as ws import subprocess import logging +import sys +import arvados.commands._util as arv_cmd -logger = logging.getLogger('arvados.arv-run') +from arvados._version import __version__ -arvrun_parser = argparse.ArgumentParser() -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'") +logger = logging.getLogger('arvados.arv-run') +logger.setLevel(logging.INFO) + +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, + help="Docker image to use, otherwise use instance default.") +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('--version', action='version', + version="%s %s" % (sys.argv[0], __version__), + help='Print version and exit.') arvrun_parser.add_argument('args', nargs=argparse.REMAINDER) class ArvFile(object): @@ -32,9 +50,18 @@ class ArvFile(object): 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 +# Determine if a file is in a collection, and return a tuple consisting of the +# portable data hash and the path relative to the root of the collection. +# Return None if the path isn't with an arv-mount collection or there was is error. def is_in_collection(root, branch): try: if root == "/": @@ -47,9 +74,12 @@ 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: + except (IOError, OSError): return (None, None) +# Determine the project to place the output of this command by searching upward +# for arv-mount psuedofile indicating the project. If the cwd isn't within +# an arv-mount project or there is an error, return current_user. def determine_project(root, current_user): try: if root == "/": @@ -65,10 +95,15 @@ def determine_project(root, current_user): else: sp = os.path.split(root) return determine_project(sp[0], current_user) - except: + except (IOError, OSError): return current_user -def statfile(prefix, fn): +# Determine if string corresponds to a file, and if that file is part of a +# arv-mounted collection or only local to the machine. Returns one of +# 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, fnPattern="$(file %s/%s)", dirPattern="$(dir %s/%s/)"): absfn = os.path.abspath(fn) if os.path.exists(absfn): st = os.stat(absfn) @@ -76,12 +111,92 @@ def statfile(prefix, fn): sp = os.path.split(absfn) (pdh, branch) = is_in_collection(sp[0], sp[1]) if pdh: - return ArvFile(prefix, "$(file %s/%s)" % (pdh, branch)) + return ArvFile(prefix, fnPattern % (pdh, branch)) else: # trim leading '/' for path prefix test later return UploadFile(prefix, absfn[1:]) + if stat.S_ISDIR(st.st_mode): + sp = os.path.split(absfn) + (pdh, branch) = is_in_collection(sp[0], sp[1]) + if pdh: + return ArvFile(prefix, dirPattern % (pdh, branch)) + return prefix+fn +def uploadfiles(files, api, dry_run=False, num_retries=0, project=None, fnPattern="$(file %s/%s)", name=None): + # 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. + n = True + pathprefix = "/" + while n: + pathstep = None + for c in files: + if pathstep is None: + sp = c.fn.split('/') + if len(sp) < 2: + # no parent directories left + n = False + break + # path step takes next directory + pathstep = sp[0] + "/" + else: + # check if pathstep is common prefix for all files + if not c.fn.startswith(pathstep): + n = False + break + if n: + # pathstep is common parent directory for all files, so remove the prefix + # from each path + pathprefix += pathstep + for c in files: + c.fn = c.fn[len(pathstep):] + + orgdir = os.getcwd() + os.chdir(pathprefix) + + logger.info("Upload local files: \"%s\"", '" "'.join([c.fn for c in files])) + + if dry_run: + logger.info("$(input) is %s", pathprefix.rstrip('/')) + pdh = "$(input)" + else: + files = sorted(files, key=lambda x: x.fn) + collection = arvados.CollectionWriter(api, num_retries=num_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]) + + filters=[["portable_data_hash", "=", collection.portable_data_hash()], + ["name", "like", name+"%"]] + if project: + filters.append(["owner_uuid", "=", project]) + + exists = api.collections().list(filters=filters).execute(num_retries=num_retries) + + if exists["items"]: + item = exists["items"][0] + logger.info("Using collection %s", item["uuid"]) + else: + body = {"owner_uuid": project, "manifest_text": collection.manifest_text()} + if name is not None: + body["name"] = name + item = api.collections().create(body=body, ensure_unique_name=True).execute() + logger.info("Uploaded to %s", item["uuid"]) + + pdh = item["portable_data_hash"] + + for c in files: + c.keepref = "%s/%s" % (pdh, c.fn) + c.fn = fnPattern % (pdh, c.fn) + + os.chdir(orgdir) + + def main(arguments=None): args = arvrun_parser.parse_args(arguments) @@ -93,6 +208,17 @@ def main(arguments=None): 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('>'): @@ -120,9 +246,11 @@ def main(arguments=None): 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("(-.)(.+)")] - + re.compile("(-[A-Za-z])(.+)")] for j, command in enumerate(slots[1:]): for i, a in enumerate(command): if j > 0 and i == 0: @@ -133,74 +261,23 @@ def main(arguments=None): # if it starts with a \ then don't do any interpretation command[i] = a[1:] else: - # Do some pattern matching - matched = False - for p in patterns: - m = p.match(a) - if m: - command[i] = statfile(m.group(1), m.group(2)) - matched = True - break - if not matched: - # parameter might be a file, so test it - command[i] = statfile('', a) + # See if it looks like a file + command[i] = statfile('', a) + + # 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 - 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. - while n: - pathstep = None - for c in files: - if pathstep is None: - sp = c.fn.split('/') - if len(sp) < 2: - # no parent directories left - n = False - break - # path step takes next directory - pathstep = sp[0] + "/" - else: - # check if pathstep is common prefix for all files - if not c.fn.startswith(pathstep): - n = False - break - if n: - # pathstep is common parent directory for all files, so remove the prefix - # from each path - pathprefix += pathstep - 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("cd %s" % pathprefix) - pdh = "$(input)" - else: - files = sorted(files, key=lambda x: x.fn) - collection = arvados.CollectionWriter(api, num_retries=3) - stream = None - for f in files: - sp = os.path.split(f.fn) - if sp[0] != stream: - stream = sp[0] - collection.start_new_stream(stream) - collection.write_file(f.fn, sp[1]) - item = api.collections().create(body={"owner_uuid": project, "manifest_text": collection.manifest_text()}).execute() - pdh = item["portable_data_hash"] - print "Uploaded to %s" % item["uuid"] - - for c in files: - c.fn = "$(file %s/%s)" % (pdh, c.fn) - - os.chdir(orgdir) + if files: + uploadfiles(files, api, dry_run=args.dry_run, num_retries=args.retries, project=project) 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]] @@ -211,11 +288,12 @@ def main(arguments=None): "repository": args.repository, "script_parameters": { }, - "runtime_constraints": { - "docker_image": args.docker_image - } + "runtime_constraints": {} } + if args.docker_image: + component["runtime_constraints"]["docker_image"] = args.docker_image + task_foreach = [] group_parser = argparse.ArgumentParser() group_parser.add_argument('-b', '--batch-size', type=int) @@ -242,7 +320,7 @@ def main(arguments=None): if slots[1]: task_foreach.append("stdin") component["script_parameters"]["stdin"] = slots[1] - component["script_parameters"]["task.stdin"] = "$(stdin)"\ + component["script_parameters"]["task.stdin"] = "$(stdin)" if task_foreach: component["script_parameters"]["task.foreach"] = task_foreach @@ -264,8 +342,8 @@ def main(arguments=None): print(json.dumps(pipeline, indent=4)) else: pipeline["owner_uuid"] = project - pi = api.pipeline_instances().create(body=pipeline).execute() - print "Running pipeline %s" % pi["uuid"] + pi = api.pipeline_instances().create(body=pipeline, ensure_unique_name=True).execute() + logger.info("Running pipeline %s", pi["uuid"]) if args.local: subprocess.call(["arv-run-pipeline-instance", "--instance", pi["uuid"], "--run-jobs-here"] + (["--no-reuse"] if args.no_reuse else [])) @@ -273,11 +351,11 @@ def main(arguments=None): ws.main(["--pipeline", pi["uuid"]]) pi = api.pipeline_instances().get(uuid=pi["uuid"]).execute() - print "Pipeline is %s" % pi["state"] + logger.info("Pipeline is %s", pi["state"]) if "output_uuid" in pi["components"]["command"]: - print "Output is %s" % pi["components"]["command"]["output_uuid"] + logger.info("Output is %s", pi["components"]["command"]["output_uuid"]) else: - print "No output" + logger.info("No output") if __name__ == '__main__': main()