sdk/cwl/arvados_cwl/arvcontainer.py

   1 import logging
   2 import json
   3 import os
   4
   5 from cwltool.errors import WorkflowException
   6 from cwltool.process import get_feature, UnsupportedRequirement, shortname
   7 from cwltool.pathmapper import adjustFiles
   8 from cwltool.utils import aslist
   9
  10 import arvados.collection
  11
  12 from .arvdocker import arv_docker_get_image
  13 from . import done
  14 from .runner import Runner, arvados_jobs_image
  15
  16 logger = logging.getLogger('arvados.cwl-runner')
  17
  18 class ArvadosContainer(object):
  19     """Submit and manage a Crunch container request for executing a CWL CommandLineTool."""
  20
  21     def __init__(self, runner):
  22         self.arvrunner = runner
  23         self.running = False
  24         self.uuid = None
  25
  26     def update_pipeline_component(self, r):
  27         pass
  28
  29     def run(self, dry_run=False, pull_image=True, **kwargs):
  30         container_request = {
  31             "command": self.command_line,
  32             "owner_uuid": self.arvrunner.project_uuid,
  33             "name": self.name,
  34             "output_path": self.outdir,
  35             "cwd": self.outdir,
  36             "priority": 1,
  37             "state": "Committed"
  38         }
  39         runtime_constraints = {}
  40         mounts = {
  41             self.outdir: {
  42                 "kind": "tmp"
  43             }
  44         }
  45
  46         dirs = set()
  47         for f in self.pathmapper.files():
  48             _, p, tp = self.pathmapper.mapper(f)
  49             if tp == "Directory" and '/' not in p[6:]:
  50                 mounts[p] = {
  51                     "kind": "collection",
  52                     "portable_data_hash": p[6:]
  53                 }
  54                 dirs.add(p[6:])
  55         for f in self.pathmapper.files():
  56             _, p, tp = self.pathmapper.mapper(f)
  57             if p[6:].split("/")[0] not in dirs:
  58                 mounts[p] = {
  59                     "kind": "collection",
  60                     "portable_data_hash": p[6:]
  61                 }
  62
  63         if self.generatefiles["listing"]:
  64             raise UnsupportedRequirement("Generate files not supported")
  65
  66         container_request["environment"] = {"TMPDIR": self.tmpdir, "HOME": self.outdir}
  67         if self.environment:
  68             container_request["environment"].update(self.environment)
  69
  70         if self.stdin:
  71             raise UnsupportedRequirement("Stdin redirection currently not suppported")
  72
  73         if self.stderr:
  74             raise UnsupportedRequirement("Stderr redirection currently not suppported")
  75
  76         if self.stdout:
  77             mounts["stdout"] = {"kind": "file",
  78                                 "path": "%s/%s" % (self.outdir, self.stdout)}
  79
  80         (docker_req, docker_is_req) = get_feature(self, "DockerRequirement")
  81         if not docker_req:
  82             docker_req = {"dockerImageId": arvados_jobs_image(self.arvrunner)}
  83
  84         container_request["container_image"] = arv_docker_get_image(self.arvrunner.api,
  85                                                                      docker_req,
  86                                                                      pull_image,
  87                                                                      self.arvrunner.project_uuid)
  88
  89         resources = self.builder.resources
  90         if resources is not None:
  91             runtime_constraints["vcpus"] = resources.get("cores", 1)
  92             runtime_constraints["ram"] = resources.get("ram") * 2**20
  93
  94         api_req, _ = get_feature(self, "http://arvados.org/cwl#APIRequirement")
  95         if api_req:
  96             runtime_constraints["API"] = True
  97
  98         runtime_req, _ = get_feature(self, "http://arvados.org/cwl#RuntimeConstraints")
  99         if runtime_req:
 100             logger.warn("RuntimeConstraints not yet supported by container API")
 101
 102         partition_req, _ = get_feature(self, "http://arvados.org/cwl#PartitionRequirement")
 103         if partition_req:
 104             runtime_constraints["partition"] = aslist(partition_req["partition"])
 105
 106         container_request["mounts"] = mounts
 107         container_request["runtime_constraints"] = runtime_constraints
 108
 109         try:
 110             response = self.arvrunner.api.container_requests().create(
 111                 body=container_request
 112             ).execute(num_retries=self.arvrunner.num_retries)
 113
 114             self.arvrunner.processes[response["container_uuid"]] = self
 115
 116             container = self.arvrunner.api.containers().get(
 117                 uuid=response["container_uuid"]
 118             ).execute(num_retries=self.arvrunner.num_retries)
 119
 120             logger.info("Container request %s (%s) state is %s with container %s %s", self.name, response["uuid"], response["state"], container["uuid"], container["state"])
 121
 122             if container["state"] in ("Complete", "Cancelled"):
 123                 self.done(container)
 124         except Exception as e:
 125             logger.error("Got error %s" % str(e))
 126             self.output_callback({}, "permanentFail")
 127
 128     def done(self, record):
 129         try:
 130             if record["state"] == "Complete":
 131                 rcode = record["exit_code"]
 132                 if self.successCodes and rcode in self.successCodes:
 133                     processStatus = "success"
 134                 elif self.temporaryFailCodes and rcode in self.temporaryFailCodes:
 135                     processStatus = "temporaryFail"
 136                 elif self.permanentFailCodes and rcode in self.permanentFailCodes:
 137                     processStatus = "permanentFail"
 138                 elif rcode == 0:
 139                     processStatus = "success"
 140                 else:
 141                     processStatus = "permanentFail"
 142             else:
 143                 processStatus = "permanentFail"
 144
 145             try:
 146                 outputs = {}
 147                 if record["output"]:
 148                     outputs = done.done(self, record, "/tmp", self.outdir, "/keep")
 149             except WorkflowException as e:
 150                 logger.error("Error while collecting container outputs:\n%s", e, exc_info=(e if self.arvrunner.debug else False))
 151                 processStatus = "permanentFail"
 152             except Exception as e:
 153                 logger.exception("Got unknown exception while collecting job outputs:")
 154                 processStatus = "permanentFail"
 155
 156             self.output_callback(outputs, processStatus)
 157         finally:
 158             del self.arvrunner.processes[record["uuid"]]
 159
 160
 161 class RunnerContainer(Runner):
 162     """Submit and manage a container that runs arvados-cwl-runner."""
 163
 164     def arvados_job_spec(self, dry_run=False, pull_image=True, **kwargs):
 165         """Create an Arvados container request for this workflow.
 166
 167         The returned dict can be used to create a container passed as
 168         the +body+ argument to container_requests().create().
 169         """
 170
 171         workflowmapper = super(RunnerContainer, self).arvados_job_spec(dry_run=dry_run, pull_image=pull_image, **kwargs)
 172
 173         with arvados.collection.Collection(api_client=self.arvrunner.api,
 174                                            keep_client=self.arvrunner.keep_client,
 175                                            num_retries=self.arvrunner.num_retries) as jobobj:
 176             with jobobj.open("cwl.input.json", "w") as f:
 177                 json.dump(self.job_order, f, sort_keys=True, indent=4)
 178             jobobj.save_new(owner_uuid=self.arvrunner.project_uuid)
 179
 180         workflowname = os.path.basename(self.tool.tool["id"])
 181         workflowpath = "/var/lib/cwl/workflow/%s" % workflowname
 182         workflowcollection = workflowmapper.mapper(self.tool.tool["id"])[1]
 183         workflowcollection = workflowcollection[5:workflowcollection.index('/')]
 184         jobpath = "/var/lib/cwl/job/cwl.input.json"
 185
 186         command = ["arvados-cwl-runner", "--local", "--api=containers"]
 187         if self.output_name:
 188             command.append("--output-name=" + self.output_name)
 189         command.extend([workflowpath, jobpath])
 190
 191         return {
 192             "command": command,
 193             "owner_uuid": self.arvrunner.project_uuid,
 194             "name": self.name,
 195             "output_path": "/var/spool/cwl",
 196             "cwd": "/var/spool/cwl",
 197             "priority": 1,
 198             "state": "Committed",
 199             "container_image": arvados_jobs_image(self.arvrunner),
 200             "mounts": {
 201                 "/var/lib/cwl/workflow": {
 202                     "kind": "collection",
 203                     "portable_data_hash": "%s" % workflowcollection
 204                 },
 205                 jobpath: {
 206                     "kind": "collection",
 207                     "portable_data_hash": "%s/cwl.input.json" % jobobj.portable_data_hash()
 208                 },
 209                 "stdout": {
 210                     "kind": "file",
 211                     "path": "/var/spool/cwl/cwl.output.json"
 212                 },
 213                 "/var/spool/cwl": {
 214                     "kind": "collection",
 215                     "writable": True
 216                 }
 217             },
 218             "runtime_constraints": {
 219                 "vcpus": 1,
 220                 "ram": 1024*1024*256,
 221                 "API": True
 222             }
 223         }
 224
 225     def run(self, *args, **kwargs):
 226         kwargs["keepprefix"] = "keep:"
 227         job_spec = self.arvados_job_spec(*args, **kwargs)
 228         job_spec.setdefault("owner_uuid", self.arvrunner.project_uuid)
 229
 230         response = self.arvrunner.api.container_requests().create(
 231             body=job_spec
 232         ).execute(num_retries=self.arvrunner.num_retries)
 233
 234         self.uuid = response["uuid"]
 235         self.arvrunner.processes[response["container_uuid"]] = self
 236
 237         logger.info("Submitted container %s", response["uuid"])
 238
 239         if response["state"] in ("Complete", "Failed", "Cancelled"):
 240             self.done(response)