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.arvrunner.processes[response["uuid"]] = self
 122
 123             logger.info("Container request %s (%s) state is %s", self.name, response["uuid"], response["state"])
 124
 125             if response["state"] == "Final":
 126                 self.done(response)
 127         except Exception as e:
 128             logger.error("Got error %s" % str(e))
 129             self.output_callback({}, "permanentFail")
 130
 131     def done(self, record):
 132         try:
 133             container = self.arvrunner.api.containers().get(
 134                 uuid=record["container_uuid"]
 135             ).execute(num_retries=self.arvrunner.num_retries)
 136             if container["state"] == "Complete":
 137                 rcode = container["exit_code"]
 138                 if self.successCodes and rcode in self.successCodes:
 139                     processStatus = "success"
 140                 elif self.temporaryFailCodes and rcode in self.temporaryFailCodes:
 141                     processStatus = "temporaryFail"
 142                 elif self.permanentFailCodes and rcode in self.permanentFailCodes:
 143                     processStatus = "permanentFail"
 144                 elif rcode == 0:
 145                     processStatus = "success"
 146                 else:
 147                     processStatus = "permanentFail"
 148             else:
 149                 processStatus = "permanentFail"
 150
 151             outputs = {}
 152
 153             if container["output"]:
 154                 try:
 155                     outputs = done.done_outputs(self, container, "/tmp", self.outdir, "/keep")
 156                 except Exception as e:
 157                     logger.error("Got error %s" % str(e))
 158                     self.output_callback({}, "permanentFail")
 159             self.output_callback(outputs, processStatus)
 160         finally:
 161             del self.arvrunner.processes[record["uuid"]]
 162
 163
 164 class RunnerContainer(Runner):
 165     """Submit and manage a container that runs arvados-cwl-runner."""
 166
 167     def arvados_job_spec(self, dry_run=False, pull_image=True, **kwargs):
 168         """Create an Arvados container request for this workflow.
 169
 170         The returned dict can be used to create a container passed as
 171         the +body+ argument to container_requests().create().
 172         """
 173
 174         workflowmapper = super(RunnerContainer, self).arvados_job_spec(dry_run=dry_run, pull_image=pull_image, **kwargs)
 175
 176         container_req = {
 177             "owner_uuid": self.arvrunner.project_uuid,
 178             "name": self.name,
 179             "output_path": "/var/spool/cwl",
 180             "cwd": "/var/spool/cwl",
 181             "priority": 1,
 182             "state": "Committed",
 183             "container_image": arvados_jobs_image(self.arvrunner),
 184             "mounts": {
 185                 "/var/lib/cwl/cwl.input.json": {
 186                     "kind": "json",
 187                     "content": self.job_order
 188                 },
 189                 "stdout": {
 190                     "kind": "file",
 191                     "path": "/var/spool/cwl/cwl.output.json"
 192                 },
 193                 "/var/spool/cwl": {
 194                     "kind": "collection",
 195                     "writable": True
 196                 }
 197             },
 198             "runtime_constraints": {
 199                 "vcpus": 1,
 200                 "ram": 1024*1024 * self.submit_runner_ram,
 201                 "API": True
 202             }
 203         }
 204
 205         workflowcollection = workflowmapper.mapper(self.tool.tool["id"])[1]
 206         if workflowcollection.startswith("keep:"):
 207             workflowcollection = workflowcollection[5:workflowcollection.index('/')]
 208             workflowname = os.path.basename(self.tool.tool["id"])
 209             workflowpath = "/var/lib/cwl/workflow/%s" % workflowname
 210             container_req["mounts"]["/var/lib/cwl/workflow"] = {
 211                 "kind": "collection",
 212                 "portable_data_hash": "%s" % workflowcollection
 213                 }
 214         elif workflowcollection.startswith("arvwf:"):
 215             workflowpath = "/var/lib/cwl/workflow.json#main"
 216             fetcher = CollectionFetcher({}, None,
 217                                         api_client=self.arvrunner.api,
 218                                         keep_client=self.arvrunner.keep_client)
 219             wfobj = yaml.safe_load(fetcher.fetch_text(workflowcollection))
 220             container_req["mounts"]["/var/lib/cwl/workflow.json"] = {
 221                 "kind": "json",
 222                 "json": wfobj
 223             }
 224
 225         command = ["arvados-cwl-runner", "--local", "--api=containers"]
 226         if self.output_name:
 227             command.append("--output-name=" + self.output_name)
 228
 229         if self.output_tags:
 230             command.append("--output-tags=" + self.output_tags)
 231
 232         if self.enable_reuse:
 233             command.append("--enable-reuse")
 234         else:
 235             command.append("--disable-reuse")
 236
 237         command.extend([workflowpath, "/var/lib/cwl/cwl.input.json"])
 238
 239         container_req["command"] = command
 240
 241         return container_req
 242
 243
 244     def run(self, *args, **kwargs):
 245         kwargs["keepprefix"] = "keep:"
 246         job_spec = self.arvados_job_spec(*args, **kwargs)
 247         job_spec.setdefault("owner_uuid", self.arvrunner.project_uuid)
 248
 249         response = self.arvrunner.api.container_requests().create(
 250             body=job_spec
 251         ).execute(num_retries=self.arvrunner.num_retries)
 252
 253         self.uuid = response["uuid"]
 254         self.arvrunner.processes[response["uuid"]] = self
 255
 256         logger.info("Submitted container %s", response["uuid"])
 257
 258         if response["state"] == "Final":
 259             self.done(response)
 260
 261     def done(self, record):
 262         try:
 263             container = self.arvrunner.api.containers().get(
 264                 uuid=record["container_uuid"]
 265             ).execute(num_retries=self.arvrunner.num_retries)
 266         except Exception as e:
 267             logger.exception("While getting runner container: %s", e)
 268             self.arvrunner.output_callback({}, "permanentFail")
 269             del self.arvrunner.processes[record["uuid"]]
 270         else:
 271             super(RunnerContainer, self).done(container)
 272         finally:
 273             del self.arvrunner.processes[record["uuid"]]