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             "output_ttl", self.arvrunner.output_ttl
  47         }
  48         runtime_constraints = {}
  49
  50         resources = self.builder.resources
  51         if resources is not None:
  52             runtime_constraints["vcpus"] = resources.get("cores", 1)
  53             runtime_constraints["ram"] = resources.get("ram") * 2**20
  54
  55         mounts = {
  56             self.outdir: {
  57                 "kind": "tmp",
  58                 "capacity": resources.get("outdirSize", 0) * 2**20
  59             },
  60             self.tmpdir: {
  61                 "kind": "tmp",
  62                 "capacity": resources.get("tmpdirSize", 0) * 2**20
  63             }
  64         }
  65         scheduling_parameters = {}
  66
  67         rf = [self.pathmapper.mapper(f) for f in self.pathmapper.referenced_files]
  68         rf.sort(key=lambda k: k.resolved)
  69         prevdir = None
  70         for resolved, target, tp, stg in rf:
  71             if not stg:
  72                 continue
  73             if prevdir and target.startswith(prevdir):
  74                 continue
  75             if tp == "Directory":
  76                 targetdir = target
  77             else:
  78                 targetdir = os.path.dirname(target)
  79             sp = resolved.split("/", 1)
  80             pdh = sp[0][5:]   # remove "keep:"
  81             mounts[targetdir] = {
  82                 "kind": "collection",
  83                 "portable_data_hash": pdh
  84             }
  85             if len(sp) == 2:
  86                 if tp == "Directory":
  87                     path = sp[1]
  88                 else:
  89                     path = os.path.dirname(sp[1])
  90                 if path and path != "/":
  91                     mounts[targetdir]["path"] = path
  92             prevdir = targetdir + "/"
  93
  94         with Perf(metrics, "generatefiles %s" % self.name):
  95             if self.generatefiles["listing"]:
  96                 vwd = arvados.collection.Collection(api_client=self.arvrunner.api,
  97                                                     keep_client=self.arvrunner.keep_client,
  98                                                     num_retries=self.arvrunner.num_retries)
  99                 generatemapper = NoFollowPathMapper([self.generatefiles], "", "",
 100                                                     separateDirs=False)
 101
 102                 with Perf(metrics, "createfiles %s" % self.name):
 103                     for f, p in generatemapper.items():
 104                         if not p.target:
 105                             pass
 106                         elif p.type in ("File", "Directory"):
 107                             source, path = self.arvrunner.fs_access.get_collection(p.resolved)
 108                             vwd.copy(path, p.target, source_collection=source)
 109                         elif p.type == "CreateFile":
 110                             with vwd.open(p.target, "w") as n:
 111                                 n.write(p.resolved.encode("utf-8"))
 112
 113                 with Perf(metrics, "generatefiles.save_new %s" % self.name):
 114                     vwd.save_new()
 115
 116                 for f, p in generatemapper.items():
 117                     if not p.target:
 118                         continue
 119                     mountpoint = "%s/%s" % (self.outdir, p.target)
 120                     mounts[mountpoint] = {"kind": "collection",
 121                                           "portable_data_hash": vwd.portable_data_hash(),
 122                                           "path": p.target}
 123
 124         container_request["environment"] = {"TMPDIR": self.tmpdir, "HOME": self.outdir}
 125         if self.environment:
 126             container_request["environment"].update(self.environment)
 127
 128         if self.stdin:
 129             sp = self.stdin[6:].split("/", 1)
 130             mounts["stdin"] = {"kind": "collection",
 131                                 "portable_data_hash": sp[0],
 132                                 "path": sp[1]}
 133
 134         if self.stderr:
 135             mounts["stderr"] = {"kind": "file",
 136                                 "path": "%s/%s" % (self.outdir, self.stderr)}
 137
 138         if self.stdout:
 139             mounts["stdout"] = {"kind": "file",
 140                                 "path": "%s/%s" % (self.outdir, self.stdout)}
 141
 142         (docker_req, docker_is_req) = get_feature(self, "DockerRequirement")
 143         if not docker_req:
 144             docker_req = {"dockerImageId": "arvados/jobs"}
 145
 146         container_request["container_image"] = arv_docker_get_image(self.arvrunner.api,
 147                                                                      docker_req,
 148                                                                      pull_image,
 149                                                                      self.arvrunner.project_uuid)
 150
 151         api_req, _ = get_feature(self, "http://arvados.org/cwl#APIRequirement")
 152         if api_req:
 153             runtime_constraints["API"] = True
 154
 155         runtime_req, _ = get_feature(self, "http://arvados.org/cwl#RuntimeConstraints")
 156         if runtime_req:
 157             if "keep_cache" in runtime_req:
 158                 runtime_constraints["keep_cache_ram"] = runtime_req["keep_cache"] * 2**20
 159             if "outputDirType" in runtime_req:
 160                 if runtime_req["outputDirType"] == "local_output_dir":
 161                     # Currently the default behavior.
 162                     pass
 163                 elif runtime_req["outputDirType"] == "keep_output_dir":
 164                     mounts[self.outdir]= {
 165                         "kind": "collection",
 166                         "writable": True
 167                     }
 168
 169         partition_req, _ = get_feature(self, "http://arvados.org/cwl#PartitionRequirement")
 170         if partition_req:
 171             scheduling_parameters["partitions"] = aslist(partition_req["partition"])
 172
 173         container_request["mounts"] = mounts
 174         container_request["runtime_constraints"] = runtime_constraints
 175         container_request["use_existing"] = kwargs.get("enable_reuse", True)
 176         container_request["scheduling_parameters"] = scheduling_parameters
 177
 178         if kwargs.get("runnerjob", "").startswith("arvwf:"):
 179             wfuuid = kwargs["runnerjob"][6:kwargs["runnerjob"].index("#")]
 180             wfrecord = self.arvrunner.api.workflows().get(uuid=wfuuid).execute(num_retries=self.arvrunner.num_retries)
 181             if container_request["name"] == "main":
 182                 container_request["name"] = wfrecord["name"]
 183             container_request["properties"]["template_uuid"] = wfuuid
 184
 185         try:
 186             response = self.arvrunner.api.container_requests().create(
 187                 body=container_request
 188             ).execute(num_retries=self.arvrunner.num_retries)
 189
 190             self.uuid = response["uuid"]
 191             self.arvrunner.processes[self.uuid] = self
 192
 193             if response["state"] == "Final":
 194                 logger.info("%s reused container %s", self.arvrunner.label(self), response["container_uuid"])
 195                 self.done(response)
 196             else:
 197                 logger.info("%s %s state is %s", self.arvrunner.label(self), response["uuid"], response["state"])
 198         except Exception as e:
 199             logger.error("%s got error %s" % (self.arvrunner.label(self), str(e)))
 200             self.output_callback({}, "permanentFail")
 201
 202     def done(self, record):
 203         try:
 204             self.arvrunner.add_intermediate_output(record["output_uuid"])
 205
 206             container = self.arvrunner.api.containers().get(
 207                 uuid=record["container_uuid"]
 208             ).execute(num_retries=self.arvrunner.num_retries)
 209             if container["state"] == "Complete":
 210                 rcode = container["exit_code"]
 211                 if self.successCodes and rcode in self.successCodes:
 212                     processStatus = "success"
 213                 elif self.temporaryFailCodes and rcode in self.temporaryFailCodes:
 214                     processStatus = "temporaryFail"
 215                 elif self.permanentFailCodes and rcode in self.permanentFailCodes:
 216                     processStatus = "permanentFail"
 217                 elif rcode == 0:
 218                     processStatus = "success"
 219                 else:
 220                     processStatus = "permanentFail"
 221             else:
 222                 processStatus = "permanentFail"
 223
 224             if processStatus == "permanentFail":
 225                 logc = arvados.collection.CollectionReader(container["log"],
 226                                                            api_client=self.arvrunner.api,
 227                                                            keep_client=self.arvrunner.keep_client,
 228                                                            num_retries=self.arvrunner.num_retries)
 229                 done.logtail(logc, logger, "%s error log:" % self.arvrunner.label(self))
 230
 231             outputs = {}
 232             if container["output"]:
 233                 outputs = done.done_outputs(self, container, "/tmp", self.outdir, "/keep")
 234         except WorkflowException as e:
 235             logger.error("%s unable to collect output from %s:\n%s",
 236                          self.arvrunner.label(self), container["output"], e, exc_info=(e if self.arvrunner.debug else False))
 237             processStatus = "permanentFail"
 238         except Exception as e:
 239             logger.exception("%s while getting output object: %s", self.arvrunner.label(self), e)
 240             processStatus = "permanentFail"
 241         finally:
 242             self.output_callback(outputs, processStatus)
 243             if record["uuid"] in self.arvrunner.processes:
 244                 del self.arvrunner.processes[record["uuid"]]
 245
 246
 247 class RunnerContainer(Runner):
 248     """Submit and manage a container that runs arvados-cwl-runner."""
 249
 250     def arvados_job_spec(self, dry_run=False, pull_image=True, **kwargs):
 251         """Create an Arvados container request for this workflow.
 252
 253         The returned dict can be used to create a container passed as
 254         the +body+ argument to container_requests().create().
 255         """
 256
 257         adjustDirObjs(self.job_order, trim_listing)
 258         adjustFileObjs(self.job_order, trim_anonymous_location)
 259         adjustDirObjs(self.job_order, trim_anonymous_location)
 260
 261         container_req = {
 262             "owner_uuid": self.arvrunner.project_uuid,
 263             "name": self.name,
 264             "output_path": "/var/spool/cwl",
 265             "cwd": "/var/spool/cwl",
 266             "priority": 1,
 267             "state": "Committed",
 268             "container_image": arvados_jobs_image(self.arvrunner, self.jobs_image),
 269             "mounts": {
 270                 "/var/lib/cwl/cwl.input.json": {
 271                     "kind": "json",
 272                     "content": self.job_order
 273                 },
 274                 "stdout": {
 275                     "kind": "file",
 276                     "path": "/var/spool/cwl/cwl.output.json"
 277                 },
 278                 "/var/spool/cwl": {
 279                     "kind": "collection",
 280                     "writable": True
 281                 }
 282             },
 283             "runtime_constraints": {
 284                 "vcpus": 1,
 285                 "ram": 1024*1024 * self.submit_runner_ram,
 286                 "API": True
 287             },
 288             "properties": {}
 289         }
 290
 291         if self.tool.tool.get("id", "").startswith("keep:"):
 292             sp = self.tool.tool["id"].split('/')
 293             workflowcollection = sp[0][5:]
 294             workflowname = "/".join(sp[1:])
 295             workflowpath = "/var/lib/cwl/workflow/%s" % workflowname
 296             container_req["mounts"]["/var/lib/cwl/workflow"] = {
 297                 "kind": "collection",
 298                 "portable_data_hash": "%s" % workflowcollection
 299             }
 300         else:
 301             packed = packed_workflow(self.arvrunner, self.tool)
 302             workflowpath = "/var/lib/cwl/workflow.json#main"
 303             container_req["mounts"]["/var/lib/cwl/workflow.json"] = {
 304                 "kind": "json",
 305                 "content": packed
 306             }
 307             if self.tool.tool.get("id", "").startswith("arvwf:"):
 308                 container_req["properties"]["template_uuid"] = self.tool.tool["id"][6:33]
 309
 310         command = ["arvados-cwl-runner", "--local", "--api=containers", "--no-log-timestamps"]
 311         if self.output_name:
 312             command.append("--output-name=" + self.output_name)
 313             container_req["output_name"] = self.output_name
 314
 315         if self.output_tags:
 316             command.append("--output-tags=" + self.output_tags)
 317
 318         if kwargs.get("debug"):
 319             command.append("--debug")
 320
 321         if self.enable_reuse:
 322             command.append("--enable-reuse")
 323         else:
 324             command.append("--disable-reuse")
 325
 326         if self.on_error:
 327             command.append("--on-error=" + self.on_error)
 328
 329         command.extend([workflowpath, "/var/lib/cwl/cwl.input.json"])
 330
 331         container_req["command"] = command
 332
 333         return container_req
 334
 335
 336     def run(self, *args, **kwargs):
 337         kwargs["keepprefix"] = "keep:"
 338         job_spec = self.arvados_job_spec(*args, **kwargs)
 339         job_spec.setdefault("owner_uuid", self.arvrunner.project_uuid)
 340
 341         response = self.arvrunner.api.container_requests().create(
 342             body=job_spec
 343         ).execute(num_retries=self.arvrunner.num_retries)
 344
 345         self.uuid = response["uuid"]
 346         self.arvrunner.processes[self.uuid] = self
 347
 348         logger.info("%s submitted container %s", self.arvrunner.label(self), response["uuid"])
 349
 350         if response["state"] == "Final":
 351             self.done(response)
 352
 353     def done(self, record):
 354         try:
 355             container = self.arvrunner.api.containers().get(
 356                 uuid=record["container_uuid"]
 357             ).execute(num_retries=self.arvrunner.num_retries)
 358         except Exception as e:
 359             logger.exception("%s while getting runner container: %s", self.arvrunner.label(self), e)
 360             self.arvrunner.output_callback({}, "permanentFail")
 361         else:
 362             super(RunnerContainer, self).done(container)
 363         finally:
 364             if record["uuid"] in self.arvrunner.processes:
 365                 del self.arvrunner.processes[record["uuid"]]