sdk/cwl/arvados_cwl/arvcontainer.py

   1 import logging
   2 import json
   3 import os
   4 import urllib
   5
   6 import ruamel.yaml as yaml
   7
   8 from cwltool.errors import WorkflowException
   9 from cwltool.process import get_feature, UnsupportedRequirement, shortname
  10 from cwltool.pathmapper import adjustFileObjs, adjustDirObjs
  11 from cwltool.utils import aslist
  12
  13 import arvados.collection
  14
  15 from .arvdocker import arv_docker_get_image
  16 from . import done
  17 from .runner import Runner, arvados_jobs_image, packed_workflow, trim_anonymous_location
  18 from .fsaccess import CollectionFetcher
  19 from .pathmapper import NoFollowPathMapper, trim_listing
  20 from .perf import Perf
  21
  22 logger = logging.getLogger('arvados.cwl-runner')
  23 metrics = logging.getLogger('arvados.cwl-runner.metrics')
  24
  25 class ArvadosContainer(object):
  26     """Submit and manage a Crunch container request for executing a CWL CommandLineTool."""
  27
  28     def __init__(self, runner):
  29         self.arvrunner = runner
  30         self.running = False
  31         self.uuid = None
  32
  33     def update_pipeline_component(self, r):
  34         pass
  35
  36     def run(self, dry_run=False, pull_image=True, **kwargs):
  37         container_request = {
  38             "command": self.command_line,
  39             "owner_uuid": self.arvrunner.project_uuid,
  40             "name": self.name,
  41             "output_path": self.outdir,
  42             "cwd": self.outdir,
  43             "priority": 1,
  44             "state": "Committed",
  45             "properties": {}
  46         }
  47         runtime_constraints = {}
  48
  49         resources = self.builder.resources
  50         if resources is not None:
  51             runtime_constraints["vcpus"] = resources.get("cores", 1)
  52             runtime_constraints["ram"] = resources.get("ram") * 2**20
  53
  54         mounts = {
  55             self.outdir: {
  56                 "kind": "tmp",
  57                 "capacity": resources.get("outdirSize", 0) * 2**20
  58             },
  59             self.tmpdir: {
  60                 "kind": "tmp",
  61                 "capacity": resources.get("tmpdirSize", 0) * 2**20
  62             }
  63         }
  64         scheduling_parameters = {}
  65
  66         rf = [self.pathmapper.mapper(f) for f in self.pathmapper.referenced_files]
  67         rf.sort(key=lambda k: k.resolved)
  68         prevdir = None
  69         for resolved, target, tp, stg in rf:
  70             if not stg:
  71                 continue
  72             if prevdir and target.startswith(prevdir):
  73                 continue
  74             if tp == "Directory":
  75                 targetdir = target
  76             else:
  77                 targetdir = os.path.dirname(target)
  78             sp = resolved.split("/", 1)
  79             pdh = sp[0][5:]   # remove "keep:"
  80             mounts[targetdir] = {
  81                 "kind": "collection",
  82                 "portable_data_hash": pdh
  83             }
  84             if len(sp) == 2:
  85                 if tp == "Directory":
  86                     path = sp[1]
  87                 else:
  88                     path = os.path.dirname(sp[1])
  89                 if path and path != "/":
  90                     mounts[targetdir]["path"] = path
  91             prevdir = targetdir + "/"
  92
  93         with Perf(metrics, "generatefiles %s" % self.name):
  94             if self.generatefiles["listing"]:
  95                 vwd = arvados.collection.Collection(api_client=self.arvrunner.api,
  96                                                     keep_client=self.arvrunner.keep_client,
  97                                                     num_retries=self.arvrunner.num_retries)
  98                 generatemapper = NoFollowPathMapper([self.generatefiles], "", "",
  99                                                     separateDirs=False)
 100
 101                 with Perf(metrics, "createfiles %s" % self.name):
 102                     for f, p in generatemapper.items():
 103                         if not p.target:
 104                             pass
 105                         elif p.type in ("File", "Directory"):
 106                             source, path = self.arvrunner.fs_access.get_collection(p.resolved)
 107                             vwd.copy(path, p.target, source_collection=source)
 108                         elif p.type == "CreateFile":
 109                             with vwd.open(p.target, "w") as n:
 110                                 n.write(p.resolved.encode("utf-8"))
 111
 112                 with Perf(metrics, "generatefiles.save_new %s" % self.name):
 113                     vwd.save_new()
 114
 115                 for f, p in generatemapper.items():
 116                     if not p.target:
 117                         continue
 118                     mountpoint = "%s/%s" % (self.outdir, p.target)
 119                     mounts[mountpoint] = {"kind": "collection",
 120                                           "portable_data_hash": vwd.portable_data_hash(),
 121                                           "path": p.target}
 122
 123         container_request["environment"] = {"TMPDIR": self.tmpdir, "HOME": self.outdir}
 124         if self.environment:
 125             container_request["environment"].update(self.environment)
 126
 127         if self.stdin:
 128             sp = self.stdin[6:].split("/", 1)
 129             mounts["stdin"] = {"kind": "collection",
 130                                 "portable_data_hash": sp[0],
 131                                 "path": sp[1]}
 132
 133         if self.stderr:
 134             mounts["stderr"] = {"kind": "file",
 135                                 "path": "%s/%s" % (self.outdir, self.stderr)}
 136
 137         if self.stdout:
 138             mounts["stdout"] = {"kind": "file",
 139                                 "path": "%s/%s" % (self.outdir, self.stdout)}
 140
 141         (docker_req, docker_is_req) = get_feature(self, "DockerRequirement")
 142         if not docker_req:
 143             docker_req = {"dockerImageId": "arvados/jobs"}
 144
 145         container_request["container_image"] = arv_docker_get_image(self.arvrunner.api,
 146                                                                      docker_req,
 147                                                                      pull_image,
 148                                                                      self.arvrunner.project_uuid)
 149
 150         api_req, _ = get_feature(self, "http://arvados.org/cwl#APIRequirement")
 151         if api_req:
 152             runtime_constraints["API"] = True
 153
 154         runtime_req, _ = get_feature(self, "http://arvados.org/cwl#RuntimeConstraints")
 155         if runtime_req:
 156             if "keep_cache" in runtime_req:
 157                 runtime_constraints["keep_cache_ram"] = runtime_req["keep_cache"] * 2**20
 158             if "outputDirType" in runtime_req:
 159                 if runtime_req["outputDirType"] == "local_output_dir":
 160                     # Currently the default behavior.
 161                     pass
 162                 elif runtime_req["outputDirType"] == "keep_output_dir":
 163                     mounts[self.outdir]= {
 164                         "kind": "collection",
 165                         "writable": True
 166                     }
 167
 168         partition_req, _ = get_feature(self, "http://arvados.org/cwl#PartitionRequirement")
 169         if partition_req:
 170             scheduling_parameters["partitions"] = aslist(partition_req["partition"])
 171
 172         container_request["mounts"] = mounts
 173         container_request["runtime_constraints"] = runtime_constraints
 174         container_request["use_existing"] = kwargs.get("enable_reuse", True)
 175         container_request["scheduling_parameters"] = scheduling_parameters
 176
 177         if kwargs.get("runnerjob", "").startswith("arvwf:"):
 178             wfuuid = kwargs["runnerjob"][6:kwargs["runnerjob"].index("#")]
 179             wfrecord = self.arvrunner.api.workflows().get(uuid=wfuuid).execute(num_retries=self.arvrunner.num_retries)
 180             if container_request["name"] == "main":
 181                 container_request["name"] = wfrecord["name"]
 182             container_request["properties"]["template_uuid"] = wfuuid
 183
 184         try:
 185             response = self.arvrunner.api.container_requests().create(
 186                 body=container_request
 187             ).execute(num_retries=self.arvrunner.num_retries)
 188
 189             self.uuid = response["uuid"]
 190             self.arvrunner.processes[self.uuid] = self
 191
 192             if response["state"] == "Final":
 193                 logger.info("%s reused container %s", self.arvrunner.label(self), response["container_uuid"])
 194                 self.done(response)
 195             else:
 196                 logger.info("%s %s state is %s", self.arvrunner.label(self), response["uuid"], response["state"])
 197         except Exception as e:
 198             logger.error("%s got error %s" % (self.arvrunner.label(self), str(e)))
 199             self.output_callback({}, "permanentFail")
 200
 201     def done(self, record):
 202         try:
 203             container = self.arvrunner.api.containers().get(
 204                 uuid=record["container_uuid"]
 205             ).execute(num_retries=self.arvrunner.num_retries)
 206             if container["state"] == "Complete":
 207                 rcode = container["exit_code"]
 208                 if self.successCodes and rcode in self.successCodes:
 209                     processStatus = "success"
 210                 elif self.temporaryFailCodes and rcode in self.temporaryFailCodes:
 211                     processStatus = "temporaryFail"
 212                 elif self.permanentFailCodes and rcode in self.permanentFailCodes:
 213                     processStatus = "permanentFail"
 214                 elif rcode == 0:
 215                     processStatus = "success"
 216                 else:
 217                     processStatus = "permanentFail"
 218             else:
 219                 processStatus = "permanentFail"
 220
 221             if processStatus == "permanentFail":
 222                 logc = arvados.collection.CollectionReader(container["log"],
 223                                                            api_client=self.arvrunner.api,
 224                                                            keep_client=self.arvrunner.keep_client,
 225                                                            num_retries=self.arvrunner.num_retries)
 226                 done.logtail(logc, logger, "%s error log:" % self.arvrunner.label(self))
 227
 228             outputs = {}
 229             if container["output"]:
 230                 outputs = done.done_outputs(self, container, "/tmp", self.outdir, "/keep")
 231         except WorkflowException as e:
 232             logger.error("%s unable to collect output from %s:\n%s",
 233                          self.arvrunner.label(self), container["output"], e, exc_info=(e if self.arvrunner.debug else False))
 234             processStatus = "permanentFail"
 235         except Exception as e:
 236             logger.exception("%s while getting output object: %s", self.arvrunner.label(self), e)
 237             processStatus = "permanentFail"
 238         finally:
 239             self.output_callback(outputs, processStatus)
 240             if record["uuid"] in self.arvrunner.processes:
 241                 del self.arvrunner.processes[record["uuid"]]
 242
 243
 244 class RunnerContainer(Runner):
 245     """Submit and manage a container that runs arvados-cwl-runner."""
 246
 247     def arvados_job_spec(self, dry_run=False, pull_image=True, **kwargs):
 248         """Create an Arvados container request for this workflow.
 249
 250         The returned dict can be used to create a container passed as
 251         the +body+ argument to container_requests().create().
 252         """
 253
 254         adjustDirObjs(self.job_order, trim_listing)
 255         adjustFileObjs(self.job_order, trim_anonymous_location)
 256         adjustDirObjs(self.job_order, trim_anonymous_location)
 257
 258         container_req = {
 259             "owner_uuid": self.arvrunner.project_uuid,
 260             "name": self.name,
 261             "output_path": "/var/spool/cwl",
 262             "cwd": "/var/spool/cwl",
 263             "priority": 1,
 264             "state": "Committed",
 265             "container_image": arvados_jobs_image(self.arvrunner, self.jobs_image),
 266             "mounts": {
 267                 "/var/lib/cwl/cwl.input.json": {
 268                     "kind": "json",
 269                     "content": self.job_order
 270                 },
 271                 "stdout": {
 272                     "kind": "file",
 273                     "path": "/var/spool/cwl/cwl.output.json"
 274                 },
 275                 "/var/spool/cwl": {
 276                     "kind": "collection",
 277                     "writable": True
 278                 }
 279             },
 280             "runtime_constraints": {
 281                 "vcpus": 1,
 282                 "ram": 1024*1024 * self.submit_runner_ram,
 283                 "API": True
 284             },
 285             "properties": {}
 286         }
 287
 288         if self.tool.tool.get("id", "").startswith("keep:"):
 289             sp = self.tool.tool["id"].split('/')
 290             workflowcollection = sp[0][5:]
 291             workflowname = "/".join(sp[1:])
 292             workflowpath = "/var/lib/cwl/workflow/%s" % workflowname
 293             container_req["mounts"]["/var/lib/cwl/workflow"] = {
 294                 "kind": "collection",
 295                 "portable_data_hash": "%s" % workflowcollection
 296             }
 297         else:
 298             packed = packed_workflow(self.arvrunner, self.tool)
 299             workflowpath = "/var/lib/cwl/workflow.json#main"
 300             container_req["mounts"]["/var/lib/cwl/workflow.json"] = {
 301                 "kind": "json",
 302                 "content": packed
 303             }
 304             if self.tool.tool.get("id", "").startswith("arvwf:"):
 305                 container_req["properties"]["template_uuid"] = self.tool.tool["id"][6:33]
 306
 307         command = ["arvados-cwl-runner", "--local", "--api=containers", "--no-log-timestamps"]
 308         if self.output_name:
 309             command.append("--output-name=" + self.output_name)
 310             container_req["output_name"] = self.output_name
 311
 312         if self.output_tags:
 313             command.append("--output-tags=" + self.output_tags)
 314
 315         if kwargs.get("debug"):
 316             command.append("--debug")
 317
 318         if self.enable_reuse:
 319             command.append("--enable-reuse")
 320         else:
 321             command.append("--disable-reuse")
 322
 323         if self.on_error:
 324             command.append("--on-error=" + self.on_error)
 325
 326         command.extend([workflowpath, "/var/lib/cwl/cwl.input.json"])
 327
 328         container_req["command"] = command
 329
 330         return container_req
 331
 332
 333     def run(self, *args, **kwargs):
 334         kwargs["keepprefix"] = "keep:"
 335         job_spec = self.arvados_job_spec(*args, **kwargs)
 336         job_spec.setdefault("owner_uuid", self.arvrunner.project_uuid)
 337
 338         response = self.arvrunner.api.container_requests().create(
 339             body=job_spec
 340         ).execute(num_retries=self.arvrunner.num_retries)
 341
 342         self.uuid = response["uuid"]
 343         self.arvrunner.processes[self.uuid] = self
 344
 345         logger.info("%s submitted container %s", self.arvrunner.label(self), response["uuid"])
 346
 347         if response["state"] == "Final":
 348             self.done(response)
 349
 350     def done(self, record):
 351         try:
 352             container = self.arvrunner.api.containers().get(
 353                 uuid=record["container_uuid"]
 354             ).execute(num_retries=self.arvrunner.num_retries)
 355         except Exception as e:
 356             logger.exception("%s while getting runner container: %s", self.arvrunner.label(self), e)
 357             self.arvrunner.output_callback({}, "permanentFail")
 358         else:
 359             super(RunnerContainer, self).done(container)
 360         finally:
 361             if record["uuid"] in self.arvrunner.processes:
 362                 del self.arvrunner.processes[record["uuid"]]