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