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