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             if "keep_cache" in runtime_req:
 101                 runtime_constraints["keep_cache_ram"] = runtime_req["keep_cache"]
 102
 103         partition_req, _ = get_feature(self, "http://arvados.org/cwl#PartitionRequirement")
 104         if partition_req:
 105             runtime_constraints["partition"] = aslist(partition_req["partition"])
 106
 107         container_request["mounts"] = mounts
 108         container_request["runtime_constraints"] = runtime_constraints
 109         container_request["use_existing"] = kwargs.get("enable_reuse", True)
 110
 111         try:
 112             response = self.arvrunner.api.container_requests().create(
 113                 body=container_request
 114             ).execute(num_retries=self.arvrunner.num_retries)
 115
 116             self.arvrunner.processes[response["container_uuid"]] = self
 117
 118             container = self.arvrunner.api.containers().get(
 119                 uuid=response["container_uuid"]
 120             ).execute(num_retries=self.arvrunner.num_retries)
 121
 122             logger.info("Container request %s (%s) state is %s with container %s %s", self.name, response["uuid"], response["state"], container["uuid"], container["state"])
 123
 124             if container["state"] in ("Complete", "Cancelled"):
 125                 self.done(container)
 126         except Exception as e:
 127             logger.error("Got error %s" % str(e))
 128             self.output_callback({}, "permanentFail")
 129
 130     def done(self, record):
 131         try:
 132             if record["state"] == "Complete":
 133                 rcode = record["exit_code"]
 134                 if self.successCodes and rcode in self.successCodes:
 135                     processStatus = "success"
 136                 elif self.temporaryFailCodes and rcode in self.temporaryFailCodes:
 137                     processStatus = "temporaryFail"
 138                 elif self.permanentFailCodes and rcode in self.permanentFailCodes:
 139                     processStatus = "permanentFail"
 140                 elif rcode == 0:
 141                     processStatus = "success"
 142                 else:
 143                     processStatus = "permanentFail"
 144             else:
 145                 processStatus = "permanentFail"
 146
 147             try:
 148                 outputs = {}
 149                 if record["output"]:
 150                     outputs = done.done(self, record, "/tmp", self.outdir, "/keep")
 151             except WorkflowException as e:
 152                 logger.error("Error while collecting container outputs:\n%s", e, exc_info=(e if self.arvrunner.debug else False))
 153                 processStatus = "permanentFail"
 154             except Exception as e:
 155                 logger.exception("Got unknown exception while collecting job outputs:")
 156                 processStatus = "permanentFail"
 157
 158             self.output_callback(outputs, processStatus)
 159         finally:
 160             del self.arvrunner.processes[record["uuid"]]
 161
 162
 163 class RunnerContainer(Runner):
 164     """Submit and manage a container that runs arvados-cwl-runner."""
 165
 166     def arvados_job_spec(self, dry_run=False, pull_image=True, **kwargs):
 167         """Create an Arvados container request for this workflow.
 168
 169         The returned dict can be used to create a container passed as
 170         the +body+ argument to container_requests().create().
 171         """
 172
 173         workflowmapper = super(RunnerContainer, self).arvados_job_spec(dry_run=dry_run, pull_image=pull_image, **kwargs)
 174
 175         with arvados.collection.Collection(api_client=self.arvrunner.api,
 176                                            keep_client=self.arvrunner.keep_client,
 177                                            num_retries=self.arvrunner.num_retries) as jobobj:
 178             with jobobj.open("cwl.input.json", "w") as f:
 179                 json.dump(self.job_order, f, sort_keys=True, indent=4)
 180             jobobj.save_new(owner_uuid=self.arvrunner.project_uuid)
 181
 182         workflowname = os.path.basename(self.tool.tool["id"])
 183         workflowpath = "/var/lib/cwl/workflow/%s" % workflowname
 184         workflowcollection = workflowmapper.mapper(self.tool.tool["id"])[1]
 185         workflowcollection = workflowcollection[5:workflowcollection.index('/')]
 186         jobpath = "/var/lib/cwl/job/cwl.input.json"
 187
 188         command = ["arvados-cwl-runner", "--local", "--api=containers"]
 189         if self.output_name:
 190             command.append("--output-name=" + self.output_name)
 191
 192         if self.enable_reuse:
 193             command.append("--enable-reuse")
 194         else:
 195             command.append("--disable-reuse")
 196
 197         command.extend([workflowpath, jobpath])
 198
 199         return {
 200             "command": command,
 201             "owner_uuid": self.arvrunner.project_uuid,
 202             "name": self.name,
 203             "output_path": "/var/spool/cwl",
 204             "cwd": "/var/spool/cwl",
 205             "priority": 1,
 206             "state": "Committed",
 207             "container_image": arvados_jobs_image(self.arvrunner),
 208             "mounts": {
 209                 "/var/lib/cwl/workflow": {
 210                     "kind": "collection",
 211                     "portable_data_hash": "%s" % workflowcollection
 212                 },
 213                 jobpath: {
 214                     "kind": "collection",
 215                     "portable_data_hash": "%s/cwl.input.json" % jobobj.portable_data_hash()
 216                 },
 217                 "stdout": {
 218                     "kind": "file",
 219                     "path": "/var/spool/cwl/cwl.output.json"
 220                 },
 221                 "/var/spool/cwl": {
 222                     "kind": "collection",
 223                     "writable": True
 224                 }
 225             },
 226             "runtime_constraints": {
 227                 "vcpus": 1,
 228                 "ram": 1024*1024*256,
 229                 "API": True
 230             }
 231         }
 232
 233     def run(self, *args, **kwargs):
 234         kwargs["keepprefix"] = "keep:"
 235         job_spec = self.arvados_job_spec(*args, **kwargs)
 236         job_spec.setdefault("owner_uuid", self.arvrunner.project_uuid)
 237
 238         response = self.arvrunner.api.container_requests().create(
 239             body=job_spec
 240         ).execute(num_retries=self.arvrunner.num_retries)
 241
 242         self.uuid = response["uuid"]
 243         self.arvrunner.processes[response["container_uuid"]] = self
 244
 245         logger.info("Submitted container %s", response["uuid"])
 246
 247         if response["state"] in ("Complete", "Failed", "Cancelled"):
 248             self.done(response)