sdk/cwl/arvados_cwl/arvcontainer.py

   1 import logging
   2 import json
   3 import os
   4 import urllib
   5
   6 import ruamel.yaml as yaml
   7
   8 from cwltool.errors import WorkflowException
   9 from cwltool.process import get_feature, UnsupportedRequirement, shortname
  10 from cwltool.pathmapper import adjustFileObjs, adjustDirObjs
  11 from cwltool.utils import aslist
  12
  13 import arvados.collection
  14
  15 from .arvdocker import arv_docker_get_image
  16 from . import done
  17 from .runner import Runner, arvados_jobs_image, packed_workflow, trim_anonymous_location
  18 from .fsaccess import CollectionFetcher
  19 from .pathmapper import NoFollowPathMapper, trim_listing
  20 from .perf import Perf
  21
  22 logger = logging.getLogger('arvados.cwl-runner')
  23 metrics = logging.getLogger('arvados.cwl-runner.metrics')
  24
  25 class ArvadosContainer(object):
  26     """Submit and manage a Crunch container request for executing a CWL CommandLineTool."""
  27
  28     def __init__(self, runner):
  29         self.arvrunner = runner
  30         self.running = False
  31         self.uuid = None
  32
  33     def update_pipeline_component(self, r):
  34         pass
  35
  36     def run(self, dry_run=False, pull_image=True, **kwargs):
  37         container_request = {
  38             "command": self.command_line,
  39             "owner_uuid": self.arvrunner.project_uuid,
  40             "name": self.name,
  41             "output_path": self.outdir,
  42             "cwd": self.outdir,
  43             "priority": 1,
  44             "state": "Committed",
  45             "properties": {}
  46         }
  47         runtime_constraints = {}
  48         mounts = {
  49             self.outdir: {
  50                 "kind": "tmp"
  51             },
  52             self.tmpdir: {
  53                 "kind": "tmp"
  54             }
  55         }
  56         scheduling_parameters = {}
  57
  58         rf = [self.pathmapper.mapper(f) for f in self.pathmapper.referenced_files]
  59         rf.sort(key=lambda k: k.resolved)
  60         prevdir = None
  61         for resolved, target, tp, stg in rf:
  62             if not stg:
  63                 continue
  64             if prevdir and target.startswith(prevdir):
  65                 continue
  66             if tp == "Directory":
  67                 targetdir = target
  68             else:
  69                 targetdir = os.path.dirname(target)
  70             sp = resolved.split("/", 1)
  71             pdh = sp[0][5:]   # remove "keep:"
  72             mounts[targetdir] = {
  73                 "kind": "collection",
  74                 "portable_data_hash": pdh
  75             }
  76             if len(sp) == 2:
  77                 if tp == "Directory":
  78                     path = sp[1]
  79                 else:
  80                     path = os.path.dirname(sp[1])
  81                 if path and path != "/":
  82                     mounts[targetdir]["path"] = path
  83             prevdir = targetdir + "/"
  84
  85         with Perf(metrics, "generatefiles %s" % self.name):
  86             if self.generatefiles["listing"]:
  87                 vwd = arvados.collection.Collection(api_client=self.arvrunner.api,
  88                                                     keep_client=self.arvrunner.keep_client,
  89                                                     num_retries=self.arvrunner.num_retries)
  90                 generatemapper = NoFollowPathMapper([self.generatefiles], "", "",
  91                                                     separateDirs=False)
  92
  93                 with Perf(metrics, "createfiles %s" % self.name):
  94                     for f, p in generatemapper.items():
  95                         if not p.target:
  96                             pass
  97                         elif p.type in ("File", "Directory"):
  98                             source, path = self.arvrunner.fs_access.get_collection(p.resolved)
  99                             vwd.copy(path, p.target, source_collection=source)
 100                         elif p.type == "CreateFile":
 101                             with vwd.open(p.target, "w") as n:
 102                                 n.write(p.resolved.encode("utf-8"))
 103
 104                 with Perf(metrics, "generatefiles.save_new %s" % self.name):
 105                     vwd.save_new()
 106
 107                 for f, p in generatemapper.items():
 108                     if not p.target:
 109                         continue
 110                     mountpoint = "%s/%s" % (self.outdir, p.target)
 111                     mounts[mountpoint] = {"kind": "collection",
 112                                           "portable_data_hash": vwd.portable_data_hash(),
 113                                           "path": p.target}
 114
 115         container_request["environment"] = {"TMPDIR": self.tmpdir, "HOME": self.outdir}
 116         if self.environment:
 117             container_request["environment"].update(self.environment)
 118
 119         if self.stdin:
 120             sp = self.stdin[6:].split("/", 1)
 121             mounts["stdin"] = {"kind": "collection",
 122                                 "portable_data_hash": sp[0],
 123                                 "path": sp[1]}
 124
 125         if self.stderr:
 126             mounts["stderr"] = {"kind": "file",
 127                                 "path": "%s/%s" % (self.outdir, self.stderr)}
 128
 129         if self.stdout:
 130             mounts["stdout"] = {"kind": "file",
 131                                 "path": "%s/%s" % (self.outdir, self.stdout)}
 132
 133         (docker_req, docker_is_req) = get_feature(self, "DockerRequirement")
 134         if not docker_req:
 135             docker_req = {"dockerImageId": "arvados/jobs"}
 136
 137         container_request["container_image"] = arv_docker_get_image(self.arvrunner.api,
 138                                                                      docker_req,
 139                                                                      pull_image,
 140                                                                      self.arvrunner.project_uuid)
 141
 142         resources = self.builder.resources
 143         if resources is not None:
 144             runtime_constraints["vcpus"] = resources.get("cores", 1)
 145             runtime_constraints["ram"] = resources.get("ram") * 2**20
 146
 147         api_req, _ = get_feature(self, "http://arvados.org/cwl#APIRequirement")
 148         if api_req:
 149             runtime_constraints["API"] = True
 150
 151         runtime_req, _ = get_feature(self, "http://arvados.org/cwl#RuntimeConstraints")
 152         if runtime_req:
 153             if "keep_cache" in runtime_req:
 154                 runtime_constraints["keep_cache_ram"] = runtime_req["keep_cache"] * 2**20
 155
 156         partition_req, _ = get_feature(self, "http://arvados.org/cwl#PartitionRequirement")
 157         if partition_req:
 158             scheduling_parameters["partitions"] = aslist(partition_req["partition"])
 159
 160         container_request["mounts"] = mounts
 161         container_request["runtime_constraints"] = runtime_constraints
 162         container_request["use_existing"] = kwargs.get("enable_reuse", True)
 163         container_request["scheduling_parameters"] = scheduling_parameters
 164
 165         if kwargs.get("runnerjob", "").startswith("arvwf:"):
 166             wfuuid = kwargs["runnerjob"][6:kwargs["runnerjob"].index("#")]
 167             wfrecord = self.arvrunner.api.workflows().get(uuid=wfuuid).execute(num_retries=self.arvrunner.num_retries)
 168             if container_request["name"] == "main":
 169                 container_request["name"] = wfrecord["name"]
 170             container_request["properties"]["template_uuid"] = wfuuid
 171
 172         try:
 173             response = self.arvrunner.api.container_requests().create(
 174                 body=container_request
 175             ).execute(num_retries=self.arvrunner.num_retries)
 176
 177             self.uuid = response["uuid"]
 178             self.arvrunner.processes[self.uuid] = self
 179
 180             if response["state"] == "Final":
 181                 logger.info("%s reused container %s", self.arvrunner.label(self), response["container_uuid"])
 182                 self.done(response)
 183             else:
 184                 logger.info("%s %s state is %s", self.arvrunner.label(self), response["uuid"], response["state"])
 185         except Exception as e:
 186             logger.error("%s got error %s" % (self.arvrunner.label(self), str(e)))
 187             self.output_callback({}, "permanentFail")
 188
 189     def done(self, record):
 190         try:
 191             container = self.arvrunner.api.containers().get(
 192                 uuid=record["container_uuid"]
 193             ).execute(num_retries=self.arvrunner.num_retries)
 194             if container["state"] == "Complete":
 195                 rcode = container["exit_code"]
 196                 if self.successCodes and rcode in self.successCodes:
 197                     processStatus = "success"
 198                 elif self.temporaryFailCodes and rcode in self.temporaryFailCodes:
 199                     processStatus = "temporaryFail"
 200                 elif self.permanentFailCodes and rcode in self.permanentFailCodes:
 201                     processStatus = "permanentFail"
 202                 elif rcode == 0:
 203                     processStatus = "success"
 204                 else:
 205                     processStatus = "permanentFail"
 206             else:
 207                 processStatus = "permanentFail"
 208
 209             if processStatus == "permanentFail":
 210                 logc = arvados.collection.CollectionReader(container["log"],
 211                                                            api_client=self.arvrunner.api,
 212                                                            keep_client=self.arvrunner.keep_client,
 213                                                            num_retries=self.arvrunner.num_retries)
 214                 done.logtail(logc, logger, "%s error log:" % self.arvrunner.label(self))
 215
 216             outputs = {}
 217             if container["output"]:
 218                 outputs = done.done_outputs(self, container, "/tmp", self.outdir, "/keep")
 219         except WorkflowException as e:
 220             logger.error("%s unable to collect output from %s:\n%s",
 221                          self.arvrunner.label(self), container["output"], e, exc_info=(e if self.arvrunner.debug else False))
 222             processStatus = "permanentFail"
 223         except Exception as e:
 224             logger.exception("%s while getting output object: %s", self.arvrunner.label(self), e)
 225             processStatus = "permanentFail"
 226         finally:
 227             self.output_callback(outputs, processStatus)
 228             if record["uuid"] in self.arvrunner.processes:
 229                 del self.arvrunner.processes[record["uuid"]]
 230
 231
 232 class RunnerContainer(Runner):
 233     """Submit and manage a container that runs arvados-cwl-runner."""
 234
 235     def arvados_job_spec(self, dry_run=False, pull_image=True, **kwargs):
 236         """Create an Arvados container request for this workflow.
 237
 238         The returned dict can be used to create a container passed as
 239         the +body+ argument to container_requests().create().
 240         """
 241
 242         adjustDirObjs(self.job_order, trim_listing)
 243         adjustFileObjs(self.job_order, trim_anonymous_location)
 244         adjustDirObjs(self.job_order, trim_anonymous_location)
 245
 246         container_req = {
 247             "owner_uuid": self.arvrunner.project_uuid,
 248             "name": self.name,
 249             "output_path": "/var/spool/cwl",
 250             "cwd": "/var/spool/cwl",
 251             "priority": 1,
 252             "state": "Committed",
 253             "container_image": arvados_jobs_image(self.arvrunner, self.jobs_image),
 254             "mounts": {
 255                 "/var/lib/cwl/cwl.input.json": {
 256                     "kind": "json",
 257                     "content": self.job_order
 258                 },
 259                 "stdout": {
 260                     "kind": "file",
 261                     "path": "/var/spool/cwl/cwl.output.json"
 262                 },
 263                 "/var/spool/cwl": {
 264                     "kind": "collection",
 265                     "writable": True
 266                 }
 267             },
 268             "runtime_constraints": {
 269                 "vcpus": 1,
 270                 "ram": 1024*1024 * self.submit_runner_ram,
 271                 "API": True
 272             },
 273             "properties": {}
 274         }
 275
 276         if self.tool.tool.get("id", "").startswith("keep:"):
 277             sp = self.tool.tool["id"].split('/')
 278             workflowcollection = sp[0][5:]
 279             workflowname = "/".join(sp[1:])
 280             workflowpath = "/var/lib/cwl/workflow/%s" % workflowname
 281             container_req["mounts"]["/var/lib/cwl/workflow"] = {
 282                 "kind": "collection",
 283                 "portable_data_hash": "%s" % workflowcollection
 284             }
 285         else:
 286             packed = packed_workflow(self.arvrunner, self.tool)
 287             workflowpath = "/var/lib/cwl/workflow.json#main"
 288             container_req["mounts"]["/var/lib/cwl/workflow.json"] = {
 289                 "kind": "json",
 290                 "content": packed
 291             }
 292             if self.tool.tool.get("id", "").startswith("arvwf:"):
 293                 container_req["properties"]["template_uuid"] = self.tool.tool["id"][6:33]
 294
 295         command = ["arvados-cwl-runner", "--local", "--api=containers", "--no-log-timestamps"]
 296         if self.output_name:
 297             command.append("--output-name=" + self.output_name)
 298             container_req["output_name"] = self.output_name
 299
 300         if self.output_tags:
 301             command.append("--output-tags=" + self.output_tags)
 302
 303         if kwargs.get("debug"):
 304             command.append("--debug")
 305
 306         if self.enable_reuse:
 307             command.append("--enable-reuse")
 308         else:
 309             command.append("--disable-reuse")
 310
 311         if self.on_error:
 312             command.append("--on-error=" + self.on_error)
 313
 314         command.extend([workflowpath, "/var/lib/cwl/cwl.input.json"])
 315
 316         container_req["command"] = command
 317
 318         return container_req
 319
 320
 321     def run(self, *args, **kwargs):
 322         kwargs["keepprefix"] = "keep:"
 323         job_spec = self.arvados_job_spec(*args, **kwargs)
 324         job_spec.setdefault("owner_uuid", self.arvrunner.project_uuid)
 325
 326         response = self.arvrunner.api.container_requests().create(
 327             body=job_spec
 328         ).execute(num_retries=self.arvrunner.num_retries)
 329
 330         self.uuid = response["uuid"]
 331         self.arvrunner.processes[self.uuid] = self
 332
 333         logger.info("%s submitted container %s", self.arvrunner.label(self), response["uuid"])
 334
 335         if response["state"] == "Final":
 336             self.done(response)
 337
 338     def done(self, record):
 339         try:
 340             container = self.arvrunner.api.containers().get(
 341                 uuid=record["container_uuid"]
 342             ).execute(num_retries=self.arvrunner.num_retries)
 343         except Exception as e:
 344             logger.exception("%s while getting runner container: %s", self.arvrunner.label(self), e)
 345             self.arvrunner.output_callback({}, "permanentFail")
 346         else:
 347             super(RunnerContainer, self).done(container)
 348         finally:
 349             if record["uuid"] in self.arvrunner.processes:
 350                 del self.arvrunner.processes[record["uuid"]]