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         intermediate_output_req, _ = get_feature(self, "http://arvados.org/cwl#IntermediateOutput")
 173         if intermediate_output_req:
 174             self.output_ttl = intermediate_output_req["outputTTL"]
 175         else:
 176             self.output_ttl = self.arvrunner.intermediate_output_ttl
 177
 178         if self.output_ttl < 0:
 179             raise WorkflowError("Invalid value %d for output_ttl, cannot be less than zero" % container_request["output_ttl"])
 180
 181         container_request["output_ttl"] = self.output_ttl
 182         container_request["mounts"] = mounts
 183         container_request["runtime_constraints"] = runtime_constraints
 184         container_request["use_existing"] = kwargs.get("enable_reuse", True)
 185         container_request["scheduling_parameters"] = scheduling_parameters
 186
 187         if kwargs.get("runnerjob", "").startswith("arvwf:"):
 188             wfuuid = kwargs["runnerjob"][6:kwargs["runnerjob"].index("#")]
 189             wfrecord = self.arvrunner.api.workflows().get(uuid=wfuuid).execute(num_retries=self.arvrunner.num_retries)
 190             if container_request["name"] == "main":
 191                 container_request["name"] = wfrecord["name"]
 192             container_request["properties"]["template_uuid"] = wfuuid
 193
 194         try:
 195             response = self.arvrunner.api.container_requests().create(
 196                 body=container_request
 197             ).execute(num_retries=self.arvrunner.num_retries)
 198
 199             self.uuid = response["uuid"]
 200             self.arvrunner.processes[self.uuid] = self
 201
 202             if response["state"] == "Final":
 203                 logger.info("%s reused container %s", self.arvrunner.label(self), response["container_uuid"])
 204                 self.done(response)
 205             else:
 206                 logger.info("%s %s state is %s", self.arvrunner.label(self), response["uuid"], response["state"])
 207         except Exception as e:
 208             logger.error("%s got error %s" % (self.arvrunner.label(self), str(e)))
 209             self.output_callback({}, "permanentFail")
 210
 211     def done(self, record):
 212         outputs = {}
 213         try:
 214             self.arvrunner.add_intermediate_output(record["output_uuid"])
 215
 216             container = self.arvrunner.api.containers().get(
 217                 uuid=record["container_uuid"]
 218             ).execute(num_retries=self.arvrunner.num_retries)
 219             if container["state"] == "Complete":
 220                 rcode = container["exit_code"]
 221                 if self.successCodes and rcode in self.successCodes:
 222                     processStatus = "success"
 223                 elif self.temporaryFailCodes and rcode in self.temporaryFailCodes:
 224                     processStatus = "temporaryFail"
 225                 elif self.permanentFailCodes and rcode in self.permanentFailCodes:
 226                     processStatus = "permanentFail"
 227                 elif rcode == 0:
 228                     processStatus = "success"
 229                 else:
 230                     processStatus = "permanentFail"
 231             else:
 232                 processStatus = "permanentFail"
 233
 234             if processStatus == "permanentFail":
 235                 logc = arvados.collection.CollectionReader(container["log"],
 236                                                            api_client=self.arvrunner.api,
 237                                                            keep_client=self.arvrunner.keep_client,
 238                                                            num_retries=self.arvrunner.num_retries)
 239                 done.logtail(logc, logger, "%s error log:" % self.arvrunner.label(self))
 240
 241             if container["output"]:
 242                 outputs = done.done_outputs(self, container, "/tmp", self.outdir, "/keep")
 243         except WorkflowException as e:
 244             logger.error("%s unable to collect output from %s:\n%s",
 245                          self.arvrunner.label(self), container["output"], e, exc_info=(e if self.arvrunner.debug else False))
 246             processStatus = "permanentFail"
 247         except Exception as e:
 248             logger.exception("%s while getting output object: %s", self.arvrunner.label(self), e)
 249             processStatus = "permanentFail"
 250         finally:
 251             self.output_callback(outputs, processStatus)
 252             if record["uuid"] in self.arvrunner.processes:
 253                 del self.arvrunner.processes[record["uuid"]]
 254
 255
 256 class RunnerContainer(Runner):
 257     """Submit and manage a container that runs arvados-cwl-runner."""
 258
 259     def arvados_job_spec(self, dry_run=False, pull_image=True, **kwargs):
 260         """Create an Arvados container request for this workflow.
 261
 262         The returned dict can be used to create a container passed as
 263         the +body+ argument to container_requests().create().
 264         """
 265
 266         adjustDirObjs(self.job_order, trim_listing)
 267         adjustFileObjs(self.job_order, trim_anonymous_location)
 268         adjustDirObjs(self.job_order, trim_anonymous_location)
 269
 270         container_req = {
 271             "owner_uuid": self.arvrunner.project_uuid,
 272             "name": self.name,
 273             "output_path": "/var/spool/cwl",
 274             "cwd": "/var/spool/cwl",
 275             "priority": 1,
 276             "state": "Committed",
 277             "container_image": arvados_jobs_image(self.arvrunner, self.jobs_image),
 278             "mounts": {
 279                 "/var/lib/cwl/cwl.input.json": {
 280                     "kind": "json",
 281                     "content": self.job_order
 282                 },
 283                 "stdout": {
 284                     "kind": "file",
 285                     "path": "/var/spool/cwl/cwl.output.json"
 286                 },
 287                 "/var/spool/cwl": {
 288                     "kind": "collection",
 289                     "writable": True
 290                 }
 291             },
 292             "runtime_constraints": {
 293                 "vcpus": 1,
 294                 "ram": 1024*1024 * self.submit_runner_ram,
 295                 "API": True
 296             },
 297             "properties": {}
 298         }
 299
 300         if self.tool.tool.get("id", "").startswith("keep:"):
 301             sp = self.tool.tool["id"].split('/')
 302             workflowcollection = sp[0][5:]
 303             workflowname = "/".join(sp[1:])
 304             workflowpath = "/var/lib/cwl/workflow/%s" % workflowname
 305             container_req["mounts"]["/var/lib/cwl/workflow"] = {
 306                 "kind": "collection",
 307                 "portable_data_hash": "%s" % workflowcollection
 308             }
 309         else:
 310             packed = packed_workflow(self.arvrunner, self.tool)
 311             workflowpath = "/var/lib/cwl/workflow.json#main"
 312             container_req["mounts"]["/var/lib/cwl/workflow.json"] = {
 313                 "kind": "json",
 314                 "content": packed
 315             }
 316             if self.tool.tool.get("id", "").startswith("arvwf:"):
 317                 container_req["properties"]["template_uuid"] = self.tool.tool["id"][6:33]
 318
 319         command = ["arvados-cwl-runner", "--local", "--api=containers", "--no-log-timestamps"]
 320         if self.output_name:
 321             command.append("--output-name=" + self.output_name)
 322             container_req["output_name"] = self.output_name
 323
 324         if self.output_tags:
 325             command.append("--output-tags=" + self.output_tags)
 326
 327         if kwargs.get("debug"):
 328             command.append("--debug")
 329
 330         if self.enable_reuse:
 331             command.append("--enable-reuse")
 332         else:
 333             command.append("--disable-reuse")
 334
 335         if self.on_error:
 336             command.append("--on-error=" + self.on_error)
 337
 338         if self.intermediate_output_ttl:
 339             command.append("--intermediate-output-ttl=%d" % self.intermediate_output_ttl)
 340
 341         command.extend([workflowpath, "/var/lib/cwl/cwl.input.json"])
 342
 343         container_req["command"] = command
 344
 345         return container_req
 346
 347
 348     def run(self, *args, **kwargs):
 349         kwargs["keepprefix"] = "keep:"
 350         job_spec = self.arvados_job_spec(*args, **kwargs)
 351         job_spec.setdefault("owner_uuid", self.arvrunner.project_uuid)
 352
 353         response = self.arvrunner.api.container_requests().create(
 354             body=job_spec
 355         ).execute(num_retries=self.arvrunner.num_retries)
 356
 357         self.uuid = response["uuid"]
 358         self.arvrunner.processes[self.uuid] = self
 359
 360         logger.info("%s submitted container %s", self.arvrunner.label(self), response["uuid"])
 361
 362         if response["state"] == "Final":
 363             self.done(response)
 364
 365     def done(self, record):
 366         try:
 367             container = self.arvrunner.api.containers().get(
 368                 uuid=record["container_uuid"]
 369             ).execute(num_retries=self.arvrunner.num_retries)
 370         except Exception as e:
 371             logger.exception("%s while getting runner container: %s", self.arvrunner.label(self), e)
 372             self.arvrunner.output_callback({}, "permanentFail")
 373         else:
 374             super(RunnerContainer, self).done(container)
 375         finally:
 376             if record["uuid"] in self.arvrunner.processes:
 377                 del self.arvrunner.processes[record["uuid"]]