#!/usr/bin/env python
import arvados
+import arvados.commands.ws as ws
import argparse
import json
import re
import os
import stat
import put
-import arvados.events
import time
+import subprocess
+import logging
+import sys
+import arvados.commands._util as arv_cmd
-arvrun_parser = argparse.ArgumentParser()
-arvrun_parser.add_argument('--dry-run', action="store_true")
-arvrun_parser.add_argument('--docker-image', type=str, default="arvados/jobs")
-arvrun_parser.add_argument('command')
-arvrun_parser.add_argument('args', nargs=argparse.REMAINDER)
+from arvados._version import __version__
+
+logger = logging.getLogger('arvados.arv-run')
+logger.setLevel(logging.INFO)
-needupload_files = []
+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):
def __init__(self, prefix, fn):
self.prefix = prefix
self.fn = fn
-def statfile(prefix, fn, pattern):
+ 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 == "/":
+ return (None, None)
+ fn = os.path.join(root, ".arvados#collection")
+ if os.path.exists(fn):
+ with file(fn, 'r') as f:
+ c = json.load(f)
+ return (c["portable_data_hash"], branch)
+ else:
+ sp = os.path.split(root)
+ return is_in_collection(sp[0], os.path.join(sp[1], branch))
+ 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 == "/":
+ return current_user
+ fn = os.path.join(root, ".arvados#project")
+ if os.path.exists(fn):
+ with file(fn, 'r') as f:
+ c = json.load(f)
+ if 'writable_by' in c and current_user in c['writable_by']:
+ return c["uuid"]
+ else:
+ return current_user
+ else:
+ sp = os.path.split(root)
+ return determine_project(sp[0], current_user)
+ except (IOError, OSError):
+ return current_user
+
+# 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):
- fn = os.path.abspath(fn)
- st = os.stat(fn)
+ st = os.stat(absfn)
if stat.S_ISREG(st.st_mode):
- mount = os.path.dirname(fn)+"/.arvados#collection"
- if os.path.exists(mount):
- with file(mount, 'r') as f:
- c = json.load(f)
- return prefix+"$(file "+c["portable_data_hash"]+"/" + os.path.basename(fn) + ")"
+ sp = os.path.split(absfn)
+ (pdh, branch) = is_in_collection(sp[0], sp[1])
+ if pdh:
+ return ArvFile(prefix, fnPattern % (pdh, branch))
else:
- needupload_files.append(fn)
- return ArvFile(prefix, fn[1:])
+ # trim leading '/' for path prefix test later
+ return UploadFile(prefix, absfn[1:])
+ if stat.S_ISDIR(st.st_mode):
+ sp = os.path.split(absfn)
+ (pdh, branch) = is_in_collection(sp[0], sp[1])
+ if pdh:
+ return ArvFile(prefix, dirPattern % (pdh, branch))
+
return prefix+fn
+def uploadfiles(files, api, dry_run=False, num_retries=0, project=None, fnPattern="$(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)
- patterns = [re.compile("(--[^=]+=)(.*)"),
- re.compile("(-[^=]+=)(.*)"),
- re.compile("(-.)(.+)")]
+ if len(args.args) == 0:
+ arvrun_parser.print_help()
+ return
- commandargs = []
+ starting_args = args.args
- for a in args.args:
- if a[0] == '-':
- matched = False
- for p in patterns:
- m = p.match(a)
- if m:
- commandargs.append(statfile(m.group(1), m.group(2), p))
- matched = True
- break
- if not matched:
- commandargs.append(a)
+ 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([])
else:
- commandargs.append(statfile('', a, None))
+ slots[reading_into].append(c)
- n = True
- pathprefix = "/"
- files = [c for c in commandargs if isinstance(c, ArvFile)]
- if len(files) > 0:
- while n:
- pathstep = None
- for c in files:
- if pathstep is None:
- sp = c.fn.split('/')
- if len(sp) < 2:
- n = False
- break
- pathstep = sp[0] + "/"
- else:
- if not c.fn.startswith(pathstep):
- n = False
- break
- if n:
- pathprefix += pathstep
- for c in files:
- c.fn = c.fn[len(pathstep):]
-
- os.chdir(pathprefix)
-
- if args.dry_run:
- print("cd %s" % pathprefix)
- print("arv-put \"%s\"" % '" "'.join([c.fn for c in files]))
- pdh = "$(input)"
+ if slots[0] and len(slots[0]) > 1:
+ logger.error("Can only specify a single stdout file (run-command substitutions are permitted)")
+ return
+
+ if not args.dry_run:
+ api = arvados.api('v1')
+ if args.project_uuid:
+ project = args.project_uuid
else:
- pdh = put.main(["--portable-data-hash"]+[c.fn for c in files])
-
- commandargs = [("%s$(file %s/%s)" % (c.prefix, pdh, c.fn)) if isinstance(c, ArvFile) else c for c in commandargs]
-
- cut = None
- i = -1
- stdio = [None, None]
- for j in xrange(0, len(commandargs)):
- c = commandargs[j]
- if c == '<':
- stdio[0] = []
- i = 0
- cut = j if cut is None else cut
- elif c == '>':
- stdio[1] = []
- i = 1
- cut = j if cut is None else cut
- elif i > -1:
- stdio[i].append(c)
-
- if cut is not None:
- commandargs = commandargs[:cut]
+ 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)
+
+ # 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
+
+ files = [c for command in slots[1:] for c in command if isinstance(c, UploadFile)]
+ 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]]
component = {
"script": "run-command",
- "script_version": "bf243e064a7a2ee4e69a87dc3ba46e949a545150",
- "repository": "arvados",
+ "script_version": args.script_version,
+ "repository": args.repository,
"script_parameters": {
- "command": [args.command]+commandargs
},
- "runtime_constraints": {
- "docker_image": args.docker_image
- }
+ "runtime_constraints": {}
}
- if stdio[0]:
- component["script_parameters"]["task.stdin"] = stdio[0][0]
- if stdio[1]:
- component["script_parameters"]["task.stdout"] = stdio[1][0]
+ 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)
+ group_parser.add_argument('args', nargs=argparse.REMAINDER)
+
+ 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": "$(%s)" % inp}]
+ else:
+ component["script_parameters"][inp] = groupargs.args
+ slots[s] = slots[s][0:i] + ["$(%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": "",
+ "name": "arv-run " + " | ".join([s[0] for s in slots[2:]]),
+ "description": "@" + " ".join(starting_args) + "@",
"components": {
- args.command: component
+ "command": component
},
- "state":"RunningOnServer"
+ "state": "RunningOnClient" if args.local else "RunningOnServer"
}
if args.dry_run:
print(json.dumps(pipeline, indent=4))
else:
- api = arvados.api('v1')
- pi = api.pipeline_instances().create(body=pipeline).execute()
- ws = None
- def report(x):
- if "event_type" in x:
- print "\n"
- print x
- if x["event_type"] == "stderr":
- print x["properties"]["text"]
- elif x["event_type"] == "update" and x["properties"]["new_attributes"]["state"] in ["Complete", "Failed"]:
- ws.close_connection()
-
- ws = arvados.events.subscribe(api, [["object_uuid", "=", pi["uuid"]], ["event_type", "in", ["stderr", "update"]]], report)
- ws.run_forever()
+ pipeline["owner_uuid"] = project
+ 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 []))
+ elif not args.no_wait:
+ ws.main(["--pipeline", pi["uuid"]])
+
+ pi = api.pipeline_instances().get(uuid=pi["uuid"]).execute()
+ logger.info("Pipeline is %s", pi["state"])
+ if "output_uuid" in pi["components"]["command"]:
+ logger.info("Output is %s", pi["components"]["command"]["output_uuid"])
+ else:
+ logger.info("No output")
if __name__ == '__main__':
main()