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, adjustFiles, UnsupportedRequirement, shortname
   7
   8 import arvados.collection
   9
  10 from .arvdocker import arv_docker_get_image
  11 from . import done
  12 from .runner import Runner
  13
  14 logger = logging.getLogger('arvados.cwl-runner')
  15
  16 class ArvadosContainer(object):
  17     """Submit and manage a Crunch job for executing a CWL CommandLineTool."""
  18
  19     def __init__(self, runner):
  20         self.arvrunner = runner
  21         self.running = False
  22
  23     def update_pipeline_component(self, r):
  24         pass
  25
  26     def run(self, dry_run=False, pull_image=True, **kwargs):
  27         container_request = {
  28             "command": self.command_line,
  29             "owner_uuid": self.arvrunner.project_uuid,
  30             "name": self.name,
  31             "output_path": "/var/spool/cwl",
  32             "cwd": "/var/spool/cwl",
  33             "priority": 1,
  34             "state": "Committed"
  35         }
  36         runtime_constraints = {}
  37         mounts = {
  38             "/var/spool/cwl": {
  39                 "kind": "tmp"
  40             }
  41         }
  42
  43         for f in self.pathmapper.files():
  44             _, p = self.pathmapper.mapper(f)
  45             mounts[p] = {
  46                 "kind": "collection",
  47                 "portable_data_hash": p[6:]
  48             }
  49
  50         if self.generatefiles:
  51             raise UnsupportedRequirement("Generate files not supported")
  52
  53             vwd = arvados.collection.Collection(api_client=self.arvrunner.api_client)
  54             container_request["task.vwd"] = {}
  55             for t in self.generatefiles:
  56                 if isinstance(self.generatefiles[t], dict):
  57                     src, rest = self.arvrunner.fs_access.get_collection(self.generatefiles[t]["path"].replace("$(task.keep)/", "keep:"))
  58                     vwd.copy(rest, t, source_collection=src)
  59                 else:
  60                     with vwd.open(t, "w") as f:
  61                         f.write(self.generatefiles[t])
  62             vwd.save_new()
  63             # TODO
  64             # for t in self.generatefiles:
  65             #     container_request["task.vwd"][t] = "$(task.keep)/%s/%s" % (vwd.portable_data_hash(), t)
  66
  67         container_request["environment"] = {"TMPDIR": "/tmp"}
  68         if self.environment:
  69             container_request["environment"].update(self.environment)
  70
  71         if self.stdin:
  72             raise UnsupportedRequirement("Stdin redirection currently not suppported")
  73
  74         if self.stdout:
  75             mounts["stdout"] = {"kind": "file",
  76                                 "path": "/var/spool/cwl/%s" % (self.stdout)}
  77
  78         (docker_req, docker_is_req) = get_feature(self, "DockerRequirement")
  79         if not docker_req:
  80             docker_req = {"dockerImageId": "arvados/jobs"}
  81
  82         container_request["container_image"] = arv_docker_get_image(self.arvrunner.api,
  83                                                                      docker_req,
  84                                                                      pull_image,
  85                                                                      self.arvrunner.project_uuid)
  86
  87         resources = self.builder.resources
  88         if resources is not None:
  89             runtime_constraints["vcpus"] = resources.get("cores", 1)
  90             runtime_constraints["ram"] = resources.get("ram") * 2**20
  91             #runtime_constraints["min_scratch_mb_per_node"] = resources.get("tmpdirSize", 0) + resources.get("outdirSize", 0)
  92
  93         container_request["mounts"] = mounts
  94         container_request["runtime_constraints"] = runtime_constraints
  95
  96         try:
  97             response = self.arvrunner.api.container_requests().create(
  98                 body=container_request
  99             ).execute(num_retries=self.arvrunner.num_retries)
 100
 101             self.arvrunner.processes[response["container_uuid"]] = self
 102
 103             logger.info("Container %s (%s) request state is %s", self.name, response["container_uuid"], response["state"])
 104
 105             if response["state"] == "Final":
 106                 self.done(response)
 107         except Exception as e:
 108             logger.error("Got error %s" % str(e))
 109             self.output_callback({}, "permanentFail")
 110
 111     def done(self, record):
 112         try:
 113             if record["state"] == "Complete":
 114                 rcode = record["exit_code"]
 115                 if self.successCodes and rcode in self.successCodes:
 116                     processStatus = "success"
 117                 elif self.temporaryFailCodes and rcode in self.temporaryFailCodes:
 118                     processStatus = "temporaryFail"
 119                 elif self.permanentFailCodes and rcode in self.permanentFailCodes:
 120                     processStatus = "permanentFail"
 121                 elif rcode == 0:
 122                     processStatus = "success"
 123                 else:
 124                     processStatus = "permanentFail"
 125             else:
 126                 processStatus = "permanentFail"
 127
 128             try:
 129                 outputs = {}
 130                 if record["output"]:
 131                     outputs = done.done(self, record, "/tmp", "/var/spool/cwl", "/keep")
 132             except WorkflowException as e:
 133                 logger.error("Error while collecting container outputs:\n%s", e, exc_info=(e if self.arvrunner.debug else False))
 134                 processStatus = "permanentFail"
 135             except Exception as e:
 136                 logger.exception("Got unknown exception while collecting job outputs:")
 137                 processStatus = "permanentFail"
 138
 139             self.output_callback(outputs, processStatus)
 140         finally:
 141             del self.arvrunner.processes[record["uuid"]]
 142
 143
 144 class RunnerContainer(Runner):
 145     """Submit and manage a container that runs arvados-cwl-runner."""
 146
 147     def arvados_job_spec(self, dry_run=False, pull_image=True, **kwargs):
 148         """Create an Arvados job specification for this workflow.
 149
 150         The returned dict can be used to create a job (i.e., passed as
 151         the +body+ argument to jobs().create()), or as a component in
 152         a pipeline template or pipeline instance.
 153         """
 154
 155         workflowmapper = super(RunnerContainer, self).arvados_job_spec(dry_run=dry_run, pull_image=pull_image, **kwargs)
 156
 157         with arvados.collection.Collection(api_client=self.arvrunner.api) as jobobj:
 158             with jobobj.open("cwl.input.json", "w") as f:
 159                 json.dump(self.job_order, f, sort_keys=True, indent=4)
 160             jobobj.save_new(owner_uuid=self.arvrunner.project_uuid)
 161
 162         workflowname = os.path.basename(self.tool.tool["id"])
 163         workflowpath = "/var/lib/cwl/workflow/%s" % workflowname
 164         workflowcollection = workflowmapper.mapper(self.tool.tool["id"])[1]
 165         workflowcollection = workflowcollection[5:workflowcollection.index('/')]
 166         jobpath = "/var/lib/cwl/job/cwl.input.json"
 167
 168         container_image = arv_docker_get_image(self.arvrunner.api,
 169                                                {"dockerImageId": "arvados/jobs"},
 170                                                pull_image,
 171                                                self.arvrunner.project_uuid)
 172
 173         return {
 174             "command": ["arvados-cwl-runner", "--local", "--api=containers", workflowpath, jobpath],
 175             "owner_uuid": self.arvrunner.project_uuid,
 176             "name": self.name,
 177             "output_path": "/var/spool/cwl",
 178             "cwd": "/var/spool/cwl",
 179             "priority": 1,
 180             "state": "Committed",
 181             "container_image": container_image,
 182             "mounts": {
 183                 "/var/lib/cwl/workflow": {
 184                     "kind": "collection",
 185                     "portable_data_hash": "%s" % workflowcollection
 186                 },
 187                 jobpath: {
 188                     "kind": "collection",
 189                     "portable_data_hash": "%s/cwl.input.json" % jobobj.portable_data_hash()
 190                 },
 191                 "stdout": {
 192                     "kind": "file",
 193                     "path": "/var/spool/cwl/cwl.output.json"
 194                 },
 195                 "/var/spool/cwl": {
 196                     "kind": "collection",
 197                     "writable": True
 198                 }
 199             },
 200             "runtime_constraints": {
 201                 "vcpus": 1,
 202                 "ram": 1024*1024*256,
 203                 "API": True
 204             }
 205         }
 206
 207     def run(self, *args, **kwargs):
 208         kwargs["keepprefix"] = "keep:"
 209         job_spec = self.arvados_job_spec(*args, **kwargs)
 210         job_spec.setdefault("owner_uuid", self.arvrunner.project_uuid)
 211
 212         response = self.arvrunner.api.container_requests().create(
 213             body=job_spec
 214         ).execute(num_retries=self.arvrunner.num_retries)
 215
 216         self.uuid = response["uuid"]
 217         self.arvrunner.processes[response["container_uuid"]] = self
 218
 219         logger.info("Submitted container %s", response["uuid"])
 220
 221         if response["state"] in ("Complete", "Failed", "Cancelled"):
 222             self.done(response)