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