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 from .pathmapper import NoFollowPathMapper
  19 from .perf import Perf
  20
  21 logger = logging.getLogger('arvados.cwl-runner')
  22 metrics = logging.getLogger('arvados.cwl-runner.metrics')
  23
  24 class ArvadosContainer(object):
  25     """Submit and manage a Crunch container request for executing a CWL CommandLineTool."""
  26
  27     def __init__(self, runner):
  28         self.arvrunner = runner
  29         self.running = False
  30         self.uuid = None
  31
  32     def update_pipeline_component(self, r):
  33         pass
  34
  35     def run(self, dry_run=False, pull_image=True, **kwargs):
  36         container_request = {
  37             "command": self.command_line,
  38             "owner_uuid": self.arvrunner.project_uuid,
  39             "name": self.name,
  40             "output_path": self.outdir,
  41             "cwd": self.outdir,
  42             "priority": 1,
  43             "state": "Committed",
  44             "properties": {}
  45         }
  46         runtime_constraints = {}
  47         mounts = {
  48             self.outdir: {
  49                 "kind": "tmp"
  50             }
  51         }
  52         scheduling_parameters = {}
  53
  54         dirs = set()
  55         for f in self.pathmapper.files():
  56             pdh, p, tp = self.pathmapper.mapper(f)
  57             if tp == "Directory" and '/' not in pdh:
  58                 mounts[p] = {
  59                     "kind": "collection",
  60                     "portable_data_hash": pdh[5:]
  61                 }
  62                 dirs.add(pdh)
  63
  64         for f in self.pathmapper.files():
  65             res, p, tp = self.pathmapper.mapper(f)
  66             if res.startswith("keep:"):
  67                 res = res[5:]
  68             elif res.startswith("/keep/"):
  69                 res = res[6:]
  70             else:
  71                 continue
  72             sp = res.split("/", 1)
  73             pdh = sp[0]
  74             if pdh not in dirs:
  75                 mounts[p] = {
  76                     "kind": "collection",
  77                     "portable_data_hash": pdh
  78                 }
  79                 if len(sp) == 2:
  80                     mounts[p]["path"] = sp[1]
  81
  82         with Perf(metrics, "generatefiles %s" % self.name):
  83             if self.generatefiles["listing"]:
  84                 vwd = arvados.collection.Collection(api_client=self.arvrunner.api,
  85                                                     keep_client=self.arvrunner.keep_client,
  86                                                     num_retries=self.arvrunner.num_retries)
  87                 generatemapper = NoFollowPathMapper([self.generatefiles], "", "",
  88                                                     separateDirs=False)
  89
  90                 with Perf(metrics, "createfiles %s" % self.name):
  91                     for f, p in generatemapper.items():
  92                         if not p.target:
  93                             pass
  94                         elif p.type in ("File", "Directory"):
  95                             source, path = self.arvrunner.fs_access.get_collection(p.resolved)
  96                             vwd.copy(path, p.target, source_collection=source)
  97                         elif p.type == "CreateFile":
  98                             with vwd.open(p.target, "w") as n:
  99                                 n.write(p.resolved.encode("utf-8"))
 100
 101                 with Perf(metrics, "generatefiles.save_new %s" % self.name):
 102                     vwd.save_new()
 103
 104                 for f, p in generatemapper.items():
 105                     if not p.target:
 106                         continue
 107                     mountpoint = "%s/%s" % (self.outdir, p.target)
 108                     mounts[mountpoint] = {"kind": "collection",
 109                                           "portable_data_hash": vwd.portable_data_hash(),
 110                                           "path": p.target}
 111
 112         container_request["environment"] = {"TMPDIR": self.tmpdir, "HOME": self.outdir}
 113         if self.environment:
 114             container_request["environment"].update(self.environment)
 115
 116         if self.stdin:
 117             raise UnsupportedRequirement("Stdin redirection currently not suppported")
 118
 119         if self.stderr:
 120             raise UnsupportedRequirement("Stderr redirection currently not suppported")
 121
 122         if self.stdout:
 123             mounts["stdout"] = {"kind": "file",
 124                                 "path": "%s/%s" % (self.outdir, self.stdout)}
 125
 126         (docker_req, docker_is_req) = get_feature(self, "DockerRequirement")
 127         if not docker_req:
 128             docker_req = {"dockerImageId": "arvados/jobs"}
 129
 130         container_request["container_image"] = arv_docker_get_image(self.arvrunner.api,
 131                                                                      docker_req,
 132                                                                      pull_image,
 133                                                                      self.arvrunner.project_uuid)
 134
 135         resources = self.builder.resources
 136         if resources is not None:
 137             runtime_constraints["vcpus"] = resources.get("cores", 1)
 138             runtime_constraints["ram"] = resources.get("ram") * 2**20
 139
 140         api_req, _ = get_feature(self, "http://arvados.org/cwl#APIRequirement")
 141         if api_req:
 142             runtime_constraints["API"] = True
 143
 144         runtime_req, _ = get_feature(self, "http://arvados.org/cwl#RuntimeConstraints")
 145         if runtime_req:
 146             if "keep_cache" in runtime_req:
 147                 runtime_constraints["keep_cache_ram"] = runtime_req["keep_cache"] * 2**20
 148
 149         partition_req, _ = get_feature(self, "http://arvados.org/cwl#PartitionRequirement")
 150         if partition_req:
 151             scheduling_parameters["partitions"] = aslist(partition_req["partition"])
 152
 153         container_request["mounts"] = mounts
 154         container_request["runtime_constraints"] = runtime_constraints
 155         container_request["use_existing"] = kwargs.get("enable_reuse", True)
 156         container_request["scheduling_parameters"] = scheduling_parameters
 157
 158         if kwargs.get("runnerjob", "").startswith("arvwf:"):
 159             wfuuid = kwargs["runnerjob"][6:kwargs["runnerjob"].index("#")]
 160             wfrecord = self.arvrunner.api.workflows().get(uuid=wfuuid).execute(num_retries=self.arvrunner.num_retries)
 161             if container_request["name"] == "main":
 162                 container_request["name"] = wfrecord["name"]
 163             container_request["properties"]["template_uuid"] = wfuuid
 164
 165         try:
 166             response = self.arvrunner.api.container_requests().create(
 167                 body=container_request
 168             ).execute(num_retries=self.arvrunner.num_retries)
 169
 170             self.uuid = response["uuid"]
 171             self.arvrunner.processes[self.uuid] = self
 172
 173             if response["state"] == "Final":
 174                 logger.info("%s reused container %s", self.arvrunner.label(self), response["container_uuid"])
 175                 self.done(response)
 176             else:
 177                 logger.info("%s %s state is %s", self.arvrunner.label(self), response["uuid"], response["state"])
 178         except Exception as e:
 179             logger.error("%s got error %s" % (self.arvrunner.label(self), str(e)))
 180             self.output_callback({}, "permanentFail")
 181
 182     def done(self, record):
 183         try:
 184             container = self.arvrunner.api.containers().get(
 185                 uuid=record["container_uuid"]
 186             ).execute(num_retries=self.arvrunner.num_retries)
 187             if container["state"] == "Complete":
 188                 rcode = container["exit_code"]
 189                 if self.successCodes and rcode in self.successCodes:
 190                     processStatus = "success"
 191                 elif self.temporaryFailCodes and rcode in self.temporaryFailCodes:
 192                     processStatus = "temporaryFail"
 193                 elif self.permanentFailCodes and rcode in self.permanentFailCodes:
 194                     processStatus = "permanentFail"
 195                 elif rcode == 0:
 196                     processStatus = "success"
 197                 else:
 198                     processStatus = "permanentFail"
 199             else:
 200                 processStatus = "permanentFail"
 201
 202             if processStatus == "permanentFail":
 203                 logc = arvados.collection.CollectionReader(container["log"],
 204                                                            api_client=self.arvrunner.api,
 205                                                            keep_client=self.arvrunner.keep_client,
 206                                                            num_retries=self.arvrunner.num_retries)
 207                 done.logtail(logc, logger, "%s error log:" % self.arvrunner.label(self))
 208
 209             outputs = {}
 210             if container["output"]:
 211                 outputs = done.done_outputs(self, container, "/tmp", self.outdir, "/keep")
 212         except WorkflowException as e:
 213             logger.error("%s unable to collect output from %s:\n%s",
 214                          self.arvrunner.label(self), container["output"], e, exc_info=(e if self.arvrunner.debug else False))
 215             processStatus = "permanentFail"
 216         except Exception as e:
 217             logger.exception("%s while getting output object: %s", self.arvrunner.label(self), e)
 218             processStatus = "permanentFail"
 219         finally:
 220             self.output_callback(outputs, processStatus)
 221             if record["uuid"] in self.arvrunner.processes:
 222                 del self.arvrunner.processes[record["uuid"]]
 223
 224
 225 class RunnerContainer(Runner):
 226     """Submit and manage a container that runs arvados-cwl-runner."""
 227
 228     def arvados_job_spec(self, dry_run=False, pull_image=True, **kwargs):
 229         """Create an Arvados container request for this workflow.
 230
 231         The returned dict can be used to create a container passed as
 232         the +body+ argument to container_requests().create().
 233         """
 234
 235         adjustDirObjs(self.job_order, trim_listing)
 236
 237         container_req = {
 238             "owner_uuid": self.arvrunner.project_uuid,
 239             "name": self.name,
 240             "output_path": "/var/spool/cwl",
 241             "cwd": "/var/spool/cwl",
 242             "priority": 1,
 243             "state": "Committed",
 244             "container_image": arvados_jobs_image(self.arvrunner, self.jobs_image),
 245             "mounts": {
 246                 "/var/lib/cwl/cwl.input.json": {
 247                     "kind": "json",
 248                     "content": self.job_order
 249                 },
 250                 "stdout": {
 251                     "kind": "file",
 252                     "path": "/var/spool/cwl/cwl.output.json"
 253                 },
 254                 "/var/spool/cwl": {
 255                     "kind": "collection",
 256                     "writable": True
 257                 }
 258             },
 259             "runtime_constraints": {
 260                 "vcpus": 1,
 261                 "ram": 1024*1024 * self.submit_runner_ram,
 262                 "API": True
 263             },
 264             "properties": {}
 265         }
 266
 267         if self.tool.tool.get("id", "").startswith("keep:"):
 268             sp = self.tool.tool["id"].split('/')
 269             workflowcollection = sp[0][5:]
 270             workflowname = "/".join(sp[1:])
 271             workflowpath = "/var/lib/cwl/workflow/%s" % workflowname
 272             container_req["mounts"]["/var/lib/cwl/workflow"] = {
 273                 "kind": "collection",
 274                 "portable_data_hash": "%s" % workflowcollection
 275             }
 276         else:
 277             packed = packed_workflow(self.arvrunner, self.tool)
 278             workflowpath = "/var/lib/cwl/workflow.json#main"
 279             container_req["mounts"]["/var/lib/cwl/workflow.json"] = {
 280                 "kind": "json",
 281                 "content": packed
 282             }
 283             if self.tool.tool.get("id", "").startswith("arvwf:"):
 284                 container_req["properties"]["template_uuid"] = self.tool.tool["id"][6:33]
 285
 286         command = ["arvados-cwl-runner", "--local", "--api=containers", "--no-log-timestamps"]
 287         if self.output_name:
 288             command.append("--output-name=" + self.output_name)
 289             container_req["output_name"] = self.output_name
 290
 291         if self.output_tags:
 292             command.append("--output-tags=" + self.output_tags)
 293
 294         if kwargs.get("debug"):
 295             command.append("--debug")
 296
 297         if self.enable_reuse:
 298             command.append("--enable-reuse")
 299         else:
 300             command.append("--disable-reuse")
 301
 302         if self.on_error:
 303             command.append("--on-error=" + self.on_error)
 304
 305         command.extend([workflowpath, "/var/lib/cwl/cwl.input.json"])
 306
 307         container_req["command"] = command
 308
 309         return container_req
 310
 311
 312     def run(self, *args, **kwargs):
 313         kwargs["keepprefix"] = "keep:"
 314         job_spec = self.arvados_job_spec(*args, **kwargs)
 315         job_spec.setdefault("owner_uuid", self.arvrunner.project_uuid)
 316
 317         response = self.arvrunner.api.container_requests().create(
 318             body=job_spec
 319         ).execute(num_retries=self.arvrunner.num_retries)
 320
 321         self.uuid = response["uuid"]
 322         self.arvrunner.processes[self.uuid] = self
 323
 324         logger.info("%s submitted container %s", self.arvrunner.label(self), response["uuid"])
 325
 326         if response["state"] == "Final":
 327             self.done(response)
 328
 329     def done(self, record):
 330         try:
 331             container = self.arvrunner.api.containers().get(
 332                 uuid=record["container_uuid"]
 333             ).execute(num_retries=self.arvrunner.num_retries)
 334         except Exception as e:
 335             logger.exception("%s while getting runner container: %s", self.arvrunner.label(self), e)
 336             self.arvrunner.output_callback({}, "permanentFail")
 337         else:
 338             super(RunnerContainer, self).done(container)
 339         finally:
 340             if record["uuid"] in self.arvrunner.processes:
 341                 del self.arvrunner.processes[record["uuid"]]