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