sdk/cwl/arvados_cwl/arvcontainer.py

   1 import logging
   2 import json
   3 import os
   4
   5 import ruamel.yaml as yaml
   6
   7 from cwltool.errors import WorkflowException
   8 from cwltool.process import get_feature, UnsupportedRequirement, shortname
   9 from cwltool.pathmapper import adjustFiles, adjustDirObjs
  10 from cwltool.utils import aslist
  11
  12 import arvados.collection
  13
  14 from .arvdocker import arv_docker_get_image
  15 from . import done
  16 from .runner import Runner, arvados_jobs_image, packed_workflow, trim_listing
  17 from .fsaccess import CollectionFetcher
  18
  19 logger = logging.getLogger('arvados.cwl-runner')
  20
  21 class ArvadosContainer(object):
  22     """Submit and manage a Crunch container request for executing a CWL CommandLineTool."""
  23
  24     def __init__(self, runner):
  25         self.arvrunner = runner
  26         self.running = False
  27         self.uuid = None
  28
  29     def update_pipeline_component(self, r):
  30         pass
  31
  32     def run(self, dry_run=False, pull_image=True, **kwargs):
  33         container_request = {
  34             "command": self.command_line,
  35             "owner_uuid": self.arvrunner.project_uuid,
  36             "name": self.name,
  37             "output_path": self.outdir,
  38             "cwd": self.outdir,
  39             "priority": 1,
  40             "state": "Committed",
  41             "properties": {}
  42         }
  43         runtime_constraints = {}
  44         mounts = {
  45             self.outdir: {
  46                 "kind": "tmp"
  47             }
  48         }
  49         scheduling_parameters = {}
  50
  51         dirs = set()
  52         for f in self.pathmapper.files():
  53             _, p, tp = self.pathmapper.mapper(f)
  54             if tp == "Directory" and '/' not in p[6:]:
  55                 mounts[p] = {
  56                     "kind": "collection",
  57                     "portable_data_hash": p[6:]
  58                 }
  59                 dirs.add(p[6:])
  60         for f in self.pathmapper.files():
  61             _, p, tp = self.pathmapper.mapper(f)
  62             if p[6:].split("/")[0] not in dirs:
  63                 mounts[p] = {
  64                     "kind": "collection",
  65                     "portable_data_hash": p[6:]
  66                 }
  67
  68         if self.generatefiles["listing"]:
  69             raise UnsupportedRequirement("InitialWorkDirRequirement not supported with --api=containers")
  70
  71         container_request["environment"] = {"TMPDIR": self.tmpdir, "HOME": self.outdir}
  72         if self.environment:
  73             container_request["environment"].update(self.environment)
  74
  75         if self.stdin:
  76             raise UnsupportedRequirement("Stdin redirection currently not suppported")
  77
  78         if self.stderr:
  79             raise UnsupportedRequirement("Stderr redirection currently not suppported")
  80
  81         if self.stdout:
  82             mounts["stdout"] = {"kind": "file",
  83                                 "path": "%s/%s" % (self.outdir, self.stdout)}
  84
  85         (docker_req, docker_is_req) = get_feature(self, "DockerRequirement")
  86         if not docker_req:
  87             docker_req = {"dockerImageId": "arvados/jobs"}
  88
  89         container_request["container_image"] = arv_docker_get_image(self.arvrunner.api,
  90                                                                      docker_req,
  91                                                                      pull_image,
  92                                                                      self.arvrunner.project_uuid)
  93
  94         resources = self.builder.resources
  95         if resources is not None:
  96             runtime_constraints["vcpus"] = resources.get("cores", 1)
  97             runtime_constraints["ram"] = resources.get("ram") * 2**20
  98
  99         api_req, _ = get_feature(self, "http://arvados.org/cwl#APIRequirement")
 100         if api_req:
 101             runtime_constraints["API"] = True
 102
 103         runtime_req, _ = get_feature(self, "http://arvados.org/cwl#RuntimeConstraints")
 104         if runtime_req:
 105             if "keep_cache" in runtime_req:
 106                 runtime_constraints["keep_cache_ram"] = runtime_req["keep_cache"]
 107
 108         partition_req, _ = get_feature(self, "http://arvados.org/cwl#PartitionRequirement")
 109         if partition_req:
 110             scheduling_parameters["partitions"] = aslist(partition_req["partition"])
 111
 112         container_request["mounts"] = mounts
 113         container_request["runtime_constraints"] = runtime_constraints
 114         container_request["use_existing"] = kwargs.get("enable_reuse", True)
 115         container_request["scheduling_parameters"] = scheduling_parameters
 116
 117         if kwargs.get("runnerjob", "").startswith("arvwf:"):
 118             wfuuid = kwargs["runnerjob"][6:kwargs["runnerjob"].index("#")]
 119             wfrecord = self.arvrunner.api.workflows().get(uuid=wfuuid).execute(num_retries=self.arvrunner.num_retries)
 120             if container_request["name"] == "main":
 121                 container_request["name"] = wfrecord["name"]
 122             container_request["properties"]["template_uuid"] = wfuuid
 123
 124         try:
 125             response = self.arvrunner.api.container_requests().create(
 126                 body=container_request
 127             ).execute(num_retries=self.arvrunner.num_retries)
 128
 129             self.uuid = response["uuid"]
 130             self.arvrunner.processes[self.uuid] = self
 131
 132             logger.info("%s %s state is %s", self.arvrunner.label(self), response["uuid"], response["state"])
 133
 134             if response["state"] == "Final":
 135                 self.done(response)
 136         except Exception as e:
 137             logger.error("%s got error %s" % (self.arvrunner.label(self), str(e)))
 138             self.output_callback({}, "permanentFail")
 139
 140     def done(self, record):
 141         try:
 142             container = self.arvrunner.api.containers().get(
 143                 uuid=record["container_uuid"]
 144             ).execute(num_retries=self.arvrunner.num_retries)
 145             if container["state"] == "Complete":
 146                 rcode = container["exit_code"]
 147                 if self.successCodes and rcode in self.successCodes:
 148                     processStatus = "success"
 149                 elif self.temporaryFailCodes and rcode in self.temporaryFailCodes:
 150                     processStatus = "temporaryFail"
 151                 elif self.permanentFailCodes and rcode in self.permanentFailCodes:
 152                     processStatus = "permanentFail"
 153                 elif rcode == 0:
 154                     processStatus = "success"
 155                 else:
 156                     processStatus = "permanentFail"
 157             else:
 158                 processStatus = "permanentFail"
 159
 160             if processStatus == "permanentFail":
 161                 logc = arvados.collection.CollectionReader(container["log"],
 162                                                            api_client=self.arvrunner.api,
 163                                                            keep_client=self.arvrunner.keep_client,
 164                                                            num_retries=self.arvrunner.num_retries)
 165                 done.logtail(logc, logger, "%s error log:" % self.arvrunner.label(self))
 166
 167             outputs = {}
 168             if container["output"]:
 169                 outputs = done.done_outputs(self, container, "/tmp", self.outdir, "/keep")
 170         except WorkflowException as e:
 171             logger.error("%s unable to collect output from %s:\n%s",
 172                          self.arvrunner.label(self), container["output"], e, exc_info=(e if self.arvrunner.debug else False))
 173             processStatus = "permanentFail"
 174         except Exception as e:
 175             logger.exception("%s while getting output object: %s", self.arvrunner.label(self), e)
 176             processStatus = "permanentFail"
 177         finally:
 178             self.output_callback(outputs, processStatus)
 179             if record["uuid"] in self.arvrunner.processes:
 180                 del self.arvrunner.processes[record["uuid"]]
 181
 182
 183 class RunnerContainer(Runner):
 184     """Submit and manage a container that runs arvados-cwl-runner."""
 185
 186     def arvados_job_spec(self, dry_run=False, pull_image=True, **kwargs):
 187         """Create an Arvados container request for this workflow.
 188
 189         The returned dict can be used to create a container passed as
 190         the +body+ argument to container_requests().create().
 191         """
 192
 193         packed = packed_workflow(self.arvrunner, self.tool)
 194
 195         adjustDirObjs(self.job_order, trim_listing)
 196
 197         container_req = {
 198             "owner_uuid": self.arvrunner.project_uuid,
 199             "name": self.name,
 200             "output_path": "/var/spool/cwl",
 201             "cwd": "/var/spool/cwl",
 202             "priority": 1,
 203             "state": "Committed",
 204             "container_image": arvados_jobs_image(self.arvrunner, self.jobs_image),
 205             "mounts": {
 206                 "/var/lib/cwl/cwl.input.json": {
 207                     "kind": "json",
 208                     "content": self.job_order
 209                 },
 210                 "stdout": {
 211                     "kind": "file",
 212                     "path": "/var/spool/cwl/cwl.output.json"
 213                 },
 214                 "/var/spool/cwl": {
 215                     "kind": "collection",
 216                     "writable": True
 217                 }
 218             },
 219             "runtime_constraints": {
 220                 "vcpus": 1,
 221                 "ram": 1024*1024 * self.submit_runner_ram,
 222                 "API": True
 223             },
 224             "properties": {}
 225         }
 226
 227         workflowpath = "/var/lib/cwl/workflow.json#main"
 228         container_req["mounts"]["/var/lib/cwl/workflow.json"] = {
 229             "kind": "json",
 230             "json": packed
 231         }
 232         if self.tool.tool.get("id", "").startswith("arvwf:"):
 233             container_req["properties"]["template_uuid"] = self.tool.tool["id"][6:33]
 234
 235         command = ["arvados-cwl-runner", "--local", "--api=containers", "--no-log-timestamps"]
 236         if self.output_name:
 237             command.append("--output-name=" + self.output_name)
 238             container_req["output_name"] = self.output_name
 239
 240         if self.output_tags:
 241             command.append("--output-tags=" + self.output_tags)
 242
 243         if kwargs.get("debug"):
 244             command.append("--debug")
 245
 246         if self.enable_reuse:
 247             command.append("--enable-reuse")
 248         else:
 249             command.append("--disable-reuse")
 250
 251         if self.on_error:
 252             command.append("--on-error=" + self.on_error)
 253
 254         command.extend([workflowpath, "/var/lib/cwl/cwl.input.json"])
 255
 256         container_req["command"] = command
 257
 258         return container_req
 259
 260
 261     def run(self, *args, **kwargs):
 262         kwargs["keepprefix"] = "keep:"
 263         job_spec = self.arvados_job_spec(*args, **kwargs)
 264         job_spec.setdefault("owner_uuid", self.arvrunner.project_uuid)
 265
 266         response = self.arvrunner.api.container_requests().create(
 267             body=job_spec
 268         ).execute(num_retries=self.arvrunner.num_retries)
 269
 270         self.uuid = response["uuid"]
 271         self.arvrunner.processes[self.uuid] = self
 272
 273         logger.info("%s submitted container %s", self.arvrunner.label(self), response["uuid"])
 274
 275         if response["state"] == "Final":
 276             self.done(response)
 277
 278     def done(self, record):
 279         try:
 280             container = self.arvrunner.api.containers().get(
 281                 uuid=record["container_uuid"]
 282             ).execute(num_retries=self.arvrunner.num_retries)
 283         except Exception as e:
 284             logger.exception("%s while getting runner container: %s", self.arvrunner.label(self), e)
 285             self.arvrunner.output_callback({}, "permanentFail")
 286         else:
 287             super(RunnerContainer, self).done(container)
 288         finally:
 289             if record["uuid"] in self.arvrunner.processes:
 290                 del self.arvrunner.processes[record["uuid"]]