sdk/cwl/arvados_cwl/__init__.py

   1 #!/usr/bin/env python
   2
   3 # Implement cwl-runner interface for submitting and running work on Arvados, using
   4 # either the Crunch jobs API or Crunch containers API.
   5
   6 import argparse
   7 import logging
   8 import os
   9 import sys
  10 import threading
  11 import hashlib
  12 import copy
  13 import json
  14 from functools import partial
  15 import pkg_resources  # part of setuptools
  16
  17 from cwltool.errors import WorkflowException
  18 import cwltool.main
  19 import cwltool.workflow
  20 import schema_salad
  21
  22 import arvados
  23 import arvados.config
  24
  25 from .arvcontainer import ArvadosContainer, RunnerContainer
  26 from .arvjob import ArvadosJob, RunnerJob, RunnerTemplate
  27 from. runner import Runner, upload_instance
  28 from .arvtool import ArvadosCommandTool
  29 from .arvworkflow import ArvadosWorkflow, upload_workflow
  30 from .fsaccess import CollectionFsAccess
  31 from .perf import Perf
  32 from .pathmapper import FinalOutputPathMapper
  33 from ._version import __version__
  34
  35 from cwltool.pack import pack
  36 from cwltool.process import shortname, UnsupportedRequirement, getListing
  37 from cwltool.pathmapper import adjustFileObjs, adjustDirObjs
  38 from cwltool.draft2tool import compute_checksums
  39 from arvados.api import OrderedJsonModel
  40
  41 logger = logging.getLogger('arvados.cwl-runner')
  42 metrics = logging.getLogger('arvados.cwl-runner.metrics')
  43 logger.setLevel(logging.INFO)
  44
  45
  46 class ArvCwlRunner(object):
  47     """Execute a CWL tool or workflow, submit work (using either jobs or
  48     containers API), wait for them to complete, and report output.
  49
  50     """
  51
  52     def __init__(self, api_client, work_api=None, keep_client=None, output_name=None):
  53         self.api = api_client
  54         self.processes = {}
  55         self.lock = threading.Lock()
  56         self.cond = threading.Condition(self.lock)
  57         self.final_output = None
  58         self.final_status = None
  59         self.uploaded = {}
  60         self.num_retries = 4
  61         self.uuid = None
  62         self.stop_polling = threading.Event()
  63         self.poll_api = None
  64         self.pipeline = None
  65         self.final_output_collection = None
  66         self.output_name = output_name
  67         self.project_uuid = None
  68
  69         if keep_client is not None:
  70             self.keep_client = keep_client
  71         else:
  72             self.keep_client = arvados.keep.KeepClient(api_client=self.api, num_retries=self.num_retries)
  73
  74         for api in ["jobs", "containers"]:
  75             try:
  76                 methods = self.api._rootDesc.get('resources')[api]['methods']
  77                 if ('httpMethod' in methods['create'] and
  78                     (work_api == api or work_api is None)):
  79                     self.work_api = api
  80                     break
  81             except KeyError:
  82                 pass
  83         if not self.work_api:
  84             if work_api is None:
  85                 raise Exception("No supported APIs")
  86             else:
  87                 raise Exception("Unsupported API '%s'" % work_api)
  88
  89     def arv_make_tool(self, toolpath_object, **kwargs):
  90         kwargs["work_api"] = self.work_api
  91         if "class" in toolpath_object and toolpath_object["class"] == "CommandLineTool":
  92             return ArvadosCommandTool(self, toolpath_object, **kwargs)
  93         elif "class" in toolpath_object and toolpath_object["class"] == "Workflow":
  94             return ArvadosWorkflow(self, toolpath_object, **kwargs)
  95         else:
  96             return cwltool.workflow.defaultMakeTool(toolpath_object, **kwargs)
  97
  98     def output_callback(self, out, processStatus):
  99         if processStatus == "success":
 100             logger.info("Overall process status is %s", processStatus)
 101             if self.pipeline:
 102                 self.api.pipeline_instances().update(uuid=self.pipeline["uuid"],
 103                                                      body={"state": "Complete"}).execute(num_retries=self.num_retries)
 104         else:
 105             logger.warn("Overall process status is %s", processStatus)
 106             if self.pipeline:
 107                 self.api.pipeline_instances().update(uuid=self.pipeline["uuid"],
 108                                                      body={"state": "Failed"}).execute(num_retries=self.num_retries)
 109         self.final_status = processStatus
 110         self.final_output = out
 111
 112     def on_message(self, event):
 113         if "object_uuid" in event:
 114             if event["object_uuid"] in self.processes and event["event_type"] == "update":
 115                 if event["properties"]["new_attributes"]["state"] == "Running" and self.processes[event["object_uuid"]].running is False:
 116                     uuid = event["object_uuid"]
 117                     with self.lock:
 118                         j = self.processes[uuid]
 119                         logger.info("Job %s (%s) is Running", j.name, uuid)
 120                         j.running = True
 121                         j.update_pipeline_component(event["properties"]["new_attributes"])
 122                 elif event["properties"]["new_attributes"]["state"] in ("Complete", "Failed", "Cancelled"):
 123                     uuid = event["object_uuid"]
 124                     try:
 125                         self.cond.acquire()
 126                         j = self.processes[uuid]
 127                         txt = self.work_api[0].upper() + self.work_api[1:-1]
 128                         logger.info("%s %s (%s) is %s", txt, j.name, uuid, event["properties"]["new_attributes"]["state"])
 129                         with Perf(metrics, "done %s" % j.name):
 130                             j.done(event["properties"]["new_attributes"])
 131                         self.cond.notify()
 132                     finally:
 133                         self.cond.release()
 134
 135     def poll_states(self):
 136         """Poll status of jobs or containers listed in the processes dict.
 137
 138         Runs in a separate thread.
 139         """
 140
 141         while True:
 142             self.stop_polling.wait(15)
 143             if self.stop_polling.is_set():
 144                 break
 145             with self.lock:
 146                 keys = self.processes.keys()
 147             if not keys:
 148                 continue
 149
 150             if self.work_api == "containers":
 151                 table = self.poll_api.containers()
 152             elif self.work_api == "jobs":
 153                 table = self.poll_api.jobs()
 154
 155             try:
 156                 proc_states = table.list(filters=[["uuid", "in", keys]]).execute(num_retries=self.num_retries)
 157             except Exception as e:
 158                 logger.warn("Error checking states on API server: %s", e)
 159                 continue
 160
 161             for p in proc_states["items"]:
 162                 self.on_message({
 163                     "object_uuid": p["uuid"],
 164                     "event_type": "update",
 165                     "properties": {
 166                         "new_attributes": p
 167                     }
 168                 })
 169
 170     def get_uploaded(self):
 171         return self.uploaded.copy()
 172
 173     def add_uploaded(self, src, pair):
 174         self.uploaded[src] = pair
 175
 176     def check_writable(self, obj):
 177         if isinstance(obj, dict):
 178             if obj.get("writable"):
 179                 raise UnsupportedRequirement("InitialWorkDir feature 'writable: true' not supported")
 180             for v in obj.itervalues():
 181                 self.check_writable(v)
 182         if isinstance(obj, list):
 183             for v in obj:
 184                 self.check_writable(v)
 185
 186     def make_output_collection(self, name, outputObj, tagsString):
 187         outputObj = copy.deepcopy(outputObj)
 188
 189         files = []
 190         def capture(fileobj):
 191             files.append(fileobj)
 192
 193         adjustDirObjs(outputObj, capture)
 194         adjustFileObjs(outputObj, capture)
 195
 196         generatemapper = FinalOutputPathMapper(files, "", "", separateDirs=False)
 197
 198         final = arvados.collection.Collection(api_client=self.api,
 199                                               keep_client=self.keep_client,
 200                                               num_retries=self.num_retries)
 201
 202         srccollections = {}
 203         for k,v in generatemapper.items():
 204             sp = k.split("/")
 205             srccollection = sp[0][5:]
 206             if srccollection not in srccollections:
 207                 srccollections[srccollection] = arvados.collection.CollectionReader(
 208                     srccollection,
 209                     api_client=self.api,
 210                     keep_client=self.keep_client,
 211                     num_retries=self.num_retries)
 212             reader = srccollections[srccollection]
 213             try:
 214                 srcpath = "/".join(sp[1:]) if len(sp) > 1 else "."
 215                 final.copy(srcpath, v.target, source_collection=reader, overwrite=False)
 216             except IOError as e:
 217                 logger.warn("While preparing output collection: %s", e)
 218
 219         def rewrite(fileobj):
 220             fileobj["location"] = generatemapper.mapper(fileobj["location"]).target
 221             for k in ("basename", "size", "listing"):
 222                 if k in fileobj:
 223                     del fileobj[k]
 224
 225         adjustDirObjs(outputObj, rewrite)
 226         adjustFileObjs(outputObj, rewrite)
 227
 228         with final.open("cwl.output.json", "w") as f:
 229             json.dump(outputObj, f, sort_keys=True, indent=4, separators=(',',': '))
 230
 231         final.save_new(name=name, owner_uuid=self.project_uuid, ensure_unique_name=True)
 232
 233         logger.info("Final output collection %s \"%s\" (%s)", final.portable_data_hash(),
 234                     final.api_response()["name"],
 235                     final.manifest_locator())
 236
 237         final_uuid = final.manifest_locator()
 238         tags = tagsString.split(',')
 239         for tag in tags:
 240              self.api.links().create(body={"head_uuid": final_uuid, "link_class": "tag", "name": tag}).execute()
 241
 242         self.final_output_collection = final
 243
 244     def set_crunch_output(self):
 245         if self.work_api == "containers":
 246             try:
 247                 current = self.api.containers().current().execute(num_retries=self.num_retries)
 248                 self.api.containers().update(uuid=current['uuid'],
 249                                              body={
 250                                                  'output': self.final_output_collection.portable_data_hash(),
 251                                              }).execute(num_retries=self.num_retries)
 252             except Exception as e:
 253                 logger.info("Setting container output: %s", e)
 254         elif self.work_api == "jobs" and "TASK_UUID" in os.environ:
 255             self.api.job_tasks().update(uuid=os.environ["TASK_UUID"],
 256                                    body={
 257                                        'output': self.final_output_collection.portable_data_hash(),
 258                                        'success': self.final_status == "success",
 259                                        'progress':1.0
 260                                    }).execute(num_retries=self.num_retries)
 261
 262     def arv_executor(self, tool, job_order, **kwargs):
 263         self.debug = kwargs.get("debug")
 264
 265         tool.visit(self.check_writable)
 266
 267         useruuid = self.api.users().current().execute()["uuid"]
 268         self.project_uuid = kwargs.get("project_uuid") if kwargs.get("project_uuid") else useruuid
 269         self.pipeline = None
 270         make_fs_access = kwargs.get("make_fs_access") or partial(CollectionFsAccess,
 271                                                                  api_client=self.api,
 272                                                                  keep_client=self.keep_client)
 273         self.fs_access = make_fs_access(kwargs["basedir"])
 274
 275         if kwargs.get("create_template"):
 276             tmpl = RunnerTemplate(self, tool, job_order, kwargs.get("enable_reuse"))
 277             tmpl.save()
 278             # cwltool.main will write our return value to stdout.
 279             return tmpl.uuid
 280
 281         if kwargs.get("create_workflow") or kwargs.get("update_workflow"):
 282             return upload_workflow(self, tool, job_order, self.project_uuid, kwargs.get("update_workflow"))
 283
 284         self.ignore_docker_for_reuse = kwargs.get("ignore_docker_for_reuse")
 285
 286         kwargs["make_fs_access"] = make_fs_access
 287         kwargs["enable_reuse"] = kwargs.get("enable_reuse")
 288         kwargs["use_container"] = True
 289         kwargs["tmpdir_prefix"] = "tmp"
 290         kwargs["on_error"] = "continue"
 291         kwargs["compute_checksum"] = kwargs.get("compute_checksum")
 292
 293         if self.work_api == "containers":
 294             kwargs["outdir"] = "/var/spool/cwl"
 295             kwargs["docker_outdir"] = "/var/spool/cwl"
 296             kwargs["tmpdir"] = "/tmp"
 297             kwargs["docker_tmpdir"] = "/tmp"
 298         elif self.work_api == "jobs":
 299             kwargs["outdir"] = "$(task.outdir)"
 300             kwargs["docker_outdir"] = "$(task.outdir)"
 301             kwargs["tmpdir"] = "$(task.tmpdir)"
 302
 303         upload_instance(self, shortname(tool.tool["id"]), tool, job_order)
 304
 305         runnerjob = None
 306         if kwargs.get("submit"):
 307             if self.work_api == "containers":
 308                 if tool.tool["class"] == "CommandLineTool":
 309                     runnerjob = tool.job(job_order,
 310                                          self.output_callback,
 311                                          **kwargs).next()
 312                 else:
 313                     runnerjob = RunnerContainer(self, tool, job_order, kwargs.get("enable_reuse"), self.output_name)
 314             else:
 315                 runnerjob = RunnerJob(self, tool, job_order, kwargs.get("enable_reuse"), self.output_name)
 316
 317         if not kwargs.get("submit") and "cwl_runner_job" not in kwargs and not self.work_api == "containers":
 318             # Create pipeline for local run
 319             self.pipeline = self.api.pipeline_instances().create(
 320                 body={
 321                     "owner_uuid": self.project_uuid,
 322                     "name": shortname(tool.tool["id"]),
 323                     "components": {},
 324                     "state": "RunningOnClient"}).execute(num_retries=self.num_retries)
 325             logger.info("Pipeline instance %s", self.pipeline["uuid"])
 326
 327         if runnerjob and not kwargs.get("wait"):
 328             runnerjob.run(wait=kwargs.get("wait"))
 329             return runnerjob.uuid
 330
 331         self.poll_api = arvados.api('v1')
 332         self.polling_thread = threading.Thread(target=self.poll_states)
 333         self.polling_thread.start()
 334
 335         if runnerjob:
 336             jobiter = iter((runnerjob,))
 337         else:
 338             if "cwl_runner_job" in kwargs:
 339                 self.uuid = kwargs.get("cwl_runner_job").get('uuid')
 340             jobiter = tool.job(job_order,
 341                                self.output_callback,
 342                                **kwargs)
 343
 344         try:
 345             self.cond.acquire()
 346             # Will continue to hold the lock for the duration of this code
 347             # except when in cond.wait(), at which point on_message can update
 348             # job state and process output callbacks.
 349
 350             loopperf = Perf(metrics, "jobiter")
 351             loopperf.__enter__()
 352             for runnable in jobiter:
 353                 loopperf.__exit__()
 354                 if runnable:
 355                     with Perf(metrics, "run"):
 356                         runnable.run(**kwargs)
 357                 else:
 358                     if self.processes:
 359                         self.cond.wait(1)
 360                     else:
 361                         logger.error("Workflow is deadlocked, no runnable jobs and not waiting on any pending jobs.")
 362                         break
 363                 loopperf.__enter__()
 364             loopperf.__exit__()
 365
 366             while self.processes:
 367                 self.cond.wait(1)
 368
 369         except UnsupportedRequirement:
 370             raise
 371         except:
 372             if sys.exc_info()[0] is KeyboardInterrupt:
 373                 logger.error("Interrupted, marking pipeline as failed")
 374             else:
 375                 logger.error("Caught unhandled exception, marking pipeline as failed.  Error was: %s", sys.exc_info()[1], exc_info=(sys.exc_info()[1] if self.debug else False))
 376             if self.pipeline:
 377                 self.api.pipeline_instances().update(uuid=self.pipeline["uuid"],
 378                                                      body={"state": "Failed"}).execute(num_retries=self.num_retries)
 379             if runnerjob and runnerjob.uuid and self.work_api == "containers":
 380                 self.api.container_requests().update(uuid=runnerjob.uuid,
 381                                                      body={"priority": "0"}).execute(num_retries=self.num_retries)
 382         finally:
 383             self.cond.release()
 384             self.stop_polling.set()
 385             self.polling_thread.join()
 386
 387         if self.final_status == "UnsupportedRequirement":
 388             raise UnsupportedRequirement("Check log for details.")
 389
 390         if self.final_output is None:
 391             raise WorkflowException("Workflow did not return a result.")
 392
 393         if kwargs.get("submit") and isinstance(runnerjob, Runner):
 394             logger.info("Final output collection %s", runnerjob.final_output)
 395         else:
 396             if self.output_name is None:
 397                 self.output_name = "Output of %s" % (shortname(tool.tool["id"]))
 398             self.make_output_collection(self.output_name, self.final_output, kwargs.get("output_tags", ""))
 399             self.set_crunch_output()
 400
 401         if self.final_status != "success":
 402             raise WorkflowException("Workflow failed.")
 403
 404         if kwargs.get("compute_checksum"):
 405             adjustDirObjs(self.final_output, partial(getListing, self.fs_access))
 406             adjustFileObjs(self.final_output, partial(compute_checksums, self.fs_access))
 407
 408         return self.final_output
 409
 410
 411 def versionstring():
 412     """Print version string of key packages for provenance and debugging."""
 413
 414     arvcwlpkg = pkg_resources.require("arvados-cwl-runner")
 415     arvpkg = pkg_resources.require("arvados-python-client")
 416     cwlpkg = pkg_resources.require("cwltool")
 417
 418     return "%s %s %s, %s %s, %s %s" % (sys.argv[0], __version__, arvcwlpkg[0].version,
 419                                     "arvados-python-client", arvpkg[0].version,
 420                                     "cwltool", cwlpkg[0].version)
 421
 422
 423 def arg_parser():  # type: () -> argparse.ArgumentParser
 424     parser = argparse.ArgumentParser(description='Arvados executor for Common Workflow Language')
 425
 426     parser.add_argument("--basedir", type=str,
 427                         help="Base directory used to resolve relative references in the input, default to directory of input object file or current directory (if inputs piped/provided on command line).")
 428     parser.add_argument("--outdir", type=str, default=os.path.abspath('.'),
 429                         help="Output directory, default current directory")
 430
 431     parser.add_argument("--eval-timeout",
 432                         help="Time to wait for a Javascript expression to evaluate before giving an error, default 20s.",
 433                         type=float,
 434                         default=20)
 435     parser.add_argument("--version", action="store_true", help="Print version and exit")
 436
 437     exgroup = parser.add_mutually_exclusive_group()
 438     exgroup.add_argument("--verbose", action="store_true", help="Default logging")
 439     exgroup.add_argument("--quiet", action="store_true", help="Only print warnings and errors.")
 440     exgroup.add_argument("--debug", action="store_true", help="Print even more logging")
 441
 442     parser.add_argument("--metrics", action="store_true", help="Print timing metrics")
 443
 444     parser.add_argument("--tool-help", action="store_true", help="Print command line help for tool")
 445
 446     exgroup = parser.add_mutually_exclusive_group()
 447     exgroup.add_argument("--enable-reuse", action="store_true",
 448                         default=True, dest="enable_reuse",
 449                         help="")
 450     exgroup.add_argument("--disable-reuse", action="store_false",
 451                         default=True, dest="enable_reuse",
 452                         help="")
 453
 454     parser.add_argument("--project-uuid", type=str, metavar="UUID", help="Project that will own the workflow jobs, if not provided, will go to home project.")
 455     parser.add_argument("--output-name", type=str, help="Name to use for collection that stores the final output.", default=None)
 456     parser.add_argument("--output-tags", type=str, help="Tags for the final output collection separated by commas, e.g., '--output-tags tag0,tag1,tag2'.", default=None)
 457     parser.add_argument("--ignore-docker-for-reuse", action="store_true",
 458                         help="Ignore Docker image version when deciding whether to reuse past jobs.",
 459                         default=False)
 460
 461     exgroup = parser.add_mutually_exclusive_group()
 462     exgroup.add_argument("--submit", action="store_true", help="Submit workflow to run on Arvados.",
 463                         default=True, dest="submit")
 464     exgroup.add_argument("--local", action="store_false", help="Run workflow on local host (submits jobs to Arvados).",
 465                         default=True, dest="submit")
 466     exgroup.add_argument("--create-template", action="store_true", help="Create an Arvados pipeline template.")
 467     exgroup.add_argument("--create-workflow", action="store_true", help="Create an Arvados workflow.")
 468     exgroup.add_argument("--update-workflow", type=str, metavar="UUID", help="Update existing Arvados workflow with uuid.")
 469
 470     exgroup = parser.add_mutually_exclusive_group()
 471     exgroup.add_argument("--wait", action="store_true", help="After submitting workflow runner job, wait for completion.",
 472                         default=True, dest="wait")
 473     exgroup.add_argument("--no-wait", action="store_false", help="Submit workflow runner job and exit.",
 474                         default=True, dest="wait")
 475
 476     parser.add_argument("--api", type=str,
 477                         default=None, dest="work_api",
 478                         help="Select work submission API, one of 'jobs' or 'containers'. Default is 'jobs' if that API is available, otherwise 'containers'.")
 479
 480     parser.add_argument("--compute-checksum", action="store_true", default=False,
 481                         help="Compute checksum of contents while collecting outputs",
 482                         dest="compute_checksum")
 483
 484     parser.add_argument("workflow", type=str, nargs="?", default=None, help="The workflow to execute")
 485     parser.add_argument("job_order", nargs=argparse.REMAINDER, help="The input object to the workflow.")
 486
 487     return parser
 488
 489 def add_arv_hints():
 490     cache = {}
 491     res = pkg_resources.resource_stream(__name__, 'arv-cwl-schema.yml')
 492     cache["http://arvados.org/cwl"] = res.read()
 493     res.close()
 494     document_loader, cwlnames, _, _ = cwltool.process.get_schema("v1.0")
 495     _, extnames, _, _ = schema_salad.schema.load_schema("http://arvados.org/cwl", cache=cache)
 496     for n in extnames.names:
 497         if not cwlnames.has_name("http://arvados.org/cwl#"+n, ""):
 498             cwlnames.add_name("http://arvados.org/cwl#"+n, "", extnames.get_name(n, ""))
 499         document_loader.idx["http://arvados.org/cwl#"+n] = {}
 500
 501 def main(args, stdout, stderr, api_client=None, keep_client=None):
 502     parser = arg_parser()
 503
 504     job_order_object = None
 505     arvargs = parser.parse_args(args)
 506     if (arvargs.create_template or arvargs.create_workflow or arvargs.update_workflow) and not arvargs.job_order:
 507         job_order_object = ({}, "")
 508
 509     add_arv_hints()
 510
 511     try:
 512         if api_client is None:
 513             api_client=arvados.api('v1', model=OrderedJsonModel())
 514         runner = ArvCwlRunner(api_client, work_api=arvargs.work_api, keep_client=keep_client, output_name=arvargs.output_name)
 515     except Exception as e:
 516         logger.error(e)
 517         return 1
 518
 519     if arvargs.debug:
 520         logger.setLevel(logging.DEBUG)
 521
 522     if arvargs.quiet:
 523         logger.setLevel(logging.WARN)
 524         logging.getLogger('arvados.arv-run').setLevel(logging.WARN)
 525
 526     if arvargs.metrics:
 527         metrics.setLevel(logging.DEBUG)
 528         logging.getLogger("cwltool.metrics").setLevel(logging.DEBUG)
 529
 530     arvargs.conformance_test = None
 531     arvargs.use_container = True
 532
 533     return cwltool.main.main(args=arvargs,
 534                              stdout=stdout,
 535                              stderr=stderr,
 536                              executor=runner.arv_executor,
 537                              makeTool=runner.arv_make_tool,
 538                              versionfunc=versionstring,
 539                              job_order_object=job_order_object,
 540                              make_fs_access=partial(CollectionFsAccess, api_client=api_client))