sdk/cwl/arvados_cwl/arvworkflow.py

   1 # Copyright (C) The Arvados Authors. All rights reserved.
   2 #
   3 # SPDX-License-Identifier: Apache-2.0
   4
   5 from past.builtins import basestring
   6 from future.utils import viewitems
   7
   8 import os
   9 import json
  10 import copy
  11 import logging
  12
  13 from schema_salad.sourceline import SourceLine, cmap
  14 import schema_salad.ref_resolver
  15
  16 from cwltool.pack import pack
  17 from cwltool.load_tool import fetch_document, resolve_and_validate_document
  18 from cwltool.process import shortname
  19 from cwltool.workflow import Workflow, WorkflowException, WorkflowStep
  20 from cwltool.utils import adjustFileObjs, adjustDirObjs, visit_class, normalizeFilesDirs
  21 from cwltool.context import LoadingContext
  22
  23 import ruamel.yaml as yaml
  24
  25 from .runner import (upload_dependencies, packed_workflow, upload_workflow_collection,
  26                      trim_anonymous_location, remove_redundant_fields, discover_secondary_files,
  27                      make_builder, arvados_jobs_image)
  28 from .pathmapper import ArvPathMapper, trim_listing
  29 from .arvtool import ArvadosCommandTool, set_cluster_target
  30 from ._version import __version__
  31
  32 from .perf import Perf
  33
  34 logger = logging.getLogger('arvados.cwl-runner')
  35 metrics = logging.getLogger('arvados.cwl-runner.metrics')
  36
  37 max_res_pars = ("coresMin", "coresMax", "ramMin", "ramMax", "tmpdirMin", "tmpdirMax")
  38 sum_res_pars = ("outdirMin", "outdirMax")
  39
  40 def upload_workflow(arvRunner, tool, job_order, project_uuid,
  41                     runtimeContext, uuid=None,
  42                     submit_runner_ram=0, name=None, merged_map=None,
  43                     submit_runner_image=None,
  44                     git_info=None):
  45
  46     packed = packed_workflow(arvRunner, tool, merged_map, runtimeContext, git_info)
  47
  48     adjustDirObjs(job_order, trim_listing)
  49     adjustFileObjs(job_order, trim_anonymous_location)
  50     adjustDirObjs(job_order, trim_anonymous_location)
  51
  52     main = [p for p in packed["$graph"] if p["id"] == "#main"][0]
  53     for inp in main["inputs"]:
  54         sn = shortname(inp["id"])
  55         if sn in job_order:
  56             inp["default"] = job_order[sn]
  57
  58     if not name:
  59         name = tool.tool.get("label", os.path.basename(tool.tool["id"]))
  60
  61     upload_dependencies(arvRunner, name, tool.doc_loader,
  62                         packed, tool.tool["id"], False,
  63                         runtimeContext)
  64
  65     wf_runner_resources = None
  66
  67     hints = main.get("hints", [])
  68     found = False
  69     for h in hints:
  70         if h["class"] == "http://arvados.org/cwl#WorkflowRunnerResources":
  71             wf_runner_resources = h
  72             found = True
  73             break
  74     if not found:
  75         wf_runner_resources = {"class": "http://arvados.org/cwl#WorkflowRunnerResources"}
  76         hints.append(wf_runner_resources)
  77
  78     wf_runner_resources["acrContainerImage"] = arvados_jobs_image(arvRunner,
  79                                                                   submit_runner_image or "arvados/jobs:"+__version__,
  80                                                                   runtimeContext)
  81
  82     if submit_runner_ram:
  83         wf_runner_resources["ramMin"] = submit_runner_ram
  84
  85     main["hints"] = hints
  86
  87     body = {
  88         "workflow": {
  89             "name": name,
  90             "description": tool.tool.get("doc", ""),
  91             "definition":json.dumps(packed, sort_keys=True, indent=4, separators=(',',': '))
  92         }}
  93     if project_uuid:
  94         body["workflow"]["owner_uuid"] = project_uuid
  95
  96     if uuid:
  97         call = arvRunner.api.workflows().update(uuid=uuid, body=body)
  98     else:
  99         call = arvRunner.api.workflows().create(body=body)
 100     return call.execute(num_retries=arvRunner.num_retries)["uuid"]
 101
 102 def dedup_reqs(reqs):
 103     dedup = {}
 104     for r in reversed(reqs):
 105         if r["class"] not in dedup and not r["class"].startswith("http://arvados.org/cwl#"):
 106             dedup[r["class"]] = r
 107     return [dedup[r] for r in sorted(dedup.keys())]
 108
 109 def get_overall_res_req(res_reqs):
 110     """Take the overall of a list of ResourceRequirement,
 111     i.e., the max of coresMin, coresMax, ramMin, ramMax, tmpdirMin, tmpdirMax
 112     and the sum of outdirMin, outdirMax."""
 113
 114     all_res_req = {}
 115     exception_msgs = []
 116     for a in max_res_pars + sum_res_pars:
 117         all_res_req[a] = []
 118         for res_req in res_reqs:
 119             if a in res_req:
 120                 if isinstance(res_req[a], int): # integer check
 121                     all_res_req[a].append(res_req[a])
 122                 else:
 123                     msg = SourceLine(res_req, a).makeError(
 124                     "Non-top-level ResourceRequirement in single container cannot have expressions")
 125                     exception_msgs.append(msg)
 126     if exception_msgs:
 127         raise WorkflowException("\n".join(exception_msgs))
 128     else:
 129         overall_res_req = {}
 130         for a in all_res_req:
 131             if all_res_req[a]:
 132                 if a in max_res_pars:
 133                     overall_res_req[a] = max(all_res_req[a])
 134                 elif a in sum_res_pars:
 135                     overall_res_req[a] = sum(all_res_req[a])
 136         if overall_res_req:
 137             overall_res_req["class"] = "ResourceRequirement"
 138         return cmap(overall_res_req)
 139
 140 class ArvadosWorkflowStep(WorkflowStep):
 141     def __init__(self,
 142                  toolpath_object,      # type: Dict[Text, Any]
 143                  pos,                  # type: int
 144                  loadingContext,       # type: LoadingContext
 145                  arvrunner,
 146                  *argc,
 147                  **argv
 148                 ):  # type: (...) -> None
 149
 150         if arvrunner.fast_submit:
 151             self.tool = toolpath_object
 152             self.tool["inputs"] = []
 153             self.tool["outputs"] = []
 154         else:
 155             super(ArvadosWorkflowStep, self).__init__(toolpath_object, pos, loadingContext, *argc, **argv)
 156             self.tool["class"] = "WorkflowStep"
 157         self.arvrunner = arvrunner
 158
 159     def job(self, joborder, output_callback, runtimeContext):
 160         runtimeContext = runtimeContext.copy()
 161         runtimeContext.toplevel = True  # Preserve behavior for #13365
 162
 163         builder = make_builder({shortname(k): v for k,v in viewitems(joborder)}, self.hints, self.requirements,
 164                                runtimeContext, self.metadata)
 165         runtimeContext = set_cluster_target(self.tool, self.arvrunner, builder, runtimeContext)
 166         return super(ArvadosWorkflowStep, self).job(joborder, output_callback, runtimeContext)
 167
 168
 169 class ArvadosWorkflow(Workflow):
 170     """Wrap cwltool Workflow to override selected methods."""
 171
 172     def __init__(self, arvrunner, toolpath_object, loadingContext):
 173         self.arvrunner = arvrunner
 174         self.wf_pdh = None
 175         self.dynamic_resource_req = []
 176         self.static_resource_req = []
 177         self.wf_reffiles = []
 178         self.loadingContext = loadingContext
 179         super(ArvadosWorkflow, self).__init__(toolpath_object, loadingContext)
 180         self.cluster_target_req, _ = self.get_requirement("http://arvados.org/cwl#ClusterTarget")
 181
 182     def job(self, joborder, output_callback, runtimeContext):
 183
 184         builder = make_builder(joborder, self.hints, self.requirements, runtimeContext, self.metadata)
 185         runtimeContext = set_cluster_target(self.tool, self.arvrunner, builder, runtimeContext)
 186
 187         req, _ = self.get_requirement("http://arvados.org/cwl#RunInSingleContainer")
 188         if not req:
 189             return super(ArvadosWorkflow, self).job(joborder, output_callback, runtimeContext)
 190
 191         # RunInSingleContainer is true
 192
 193         with SourceLine(self.tool, None, WorkflowException, logger.isEnabledFor(logging.DEBUG)):
 194             if "id" not in self.tool:
 195                 raise WorkflowException("%s object must have 'id'" % (self.tool["class"]))
 196
 197         discover_secondary_files(self.arvrunner.fs_access, builder,
 198                                  self.tool["inputs"], joborder)
 199         normalizeFilesDirs(joborder)
 200
 201         with Perf(metrics, "subworkflow upload_deps"):
 202             upload_dependencies(self.arvrunner,
 203                                 os.path.basename(joborder.get("id", "#")),
 204                                 self.doc_loader,
 205                                 joborder,
 206                                 joborder.get("id", "#"),
 207                                 False,
 208                                 runtimeContext)
 209
 210             if self.wf_pdh is None:
 211                 packed = pack(self.loadingContext, self.tool["id"], loader=self.doc_loader)
 212
 213                 for p in packed["$graph"]:
 214                     if p["id"] == "#main":
 215                         p["requirements"] = dedup_reqs(self.requirements)
 216                         p["hints"] = dedup_reqs(self.hints)
 217
 218                 def visit(item):
 219                     if "requirements" in item:
 220                         item["requirements"] = [i for i in item["requirements"] if i["class"] != "DockerRequirement"]
 221                     for t in ("hints", "requirements"):
 222                         if t not in item:
 223                             continue
 224                         for req in item[t]:
 225                             if req["class"] == "ResourceRequirement":
 226                                 dyn = False
 227                                 for k in max_res_pars + sum_res_pars:
 228                                     if k in req:
 229                                         if isinstance(req[k], basestring):
 230                                             if item["id"] == "#main":
 231                                                 # only the top-level requirements/hints may contain expressions
 232                                                 self.dynamic_resource_req.append(req)
 233                                                 dyn = True
 234                                                 break
 235                                             else:
 236                                                 with SourceLine(req, k, WorkflowException):
 237                                                     raise WorkflowException("Non-top-level ResourceRequirement in single container cannot have expressions")
 238                                 if not dyn:
 239                                     self.static_resource_req.append(req)
 240
 241                 visit_class(packed["$graph"], ("Workflow", "CommandLineTool"), visit)
 242
 243                 if self.static_resource_req:
 244                     self.static_resource_req = [get_overall_res_req(self.static_resource_req)]
 245
 246                 upload_dependencies(self.arvrunner,
 247                                     runtimeContext.name,
 248                                     self.doc_loader,
 249                                     packed,
 250                                     self.tool["id"],
 251                                     False,
 252                                     runtimeContext)
 253
 254                 # Discover files/directories referenced by the
 255                 # workflow (mainly "default" values)
 256                 visit_class(packed, ("File", "Directory"), self.wf_reffiles.append)
 257
 258
 259         if self.dynamic_resource_req:
 260             # Evaluate dynamic resource requirements using current builder
 261             rs = copy.copy(self.static_resource_req)
 262             for dyn_rs in self.dynamic_resource_req:
 263                 eval_req = {"class": "ResourceRequirement"}
 264                 for a in max_res_pars + sum_res_pars:
 265                     if a in dyn_rs:
 266                         eval_req[a] = builder.do_eval(dyn_rs[a])
 267                 rs.append(eval_req)
 268             job_res_reqs = [get_overall_res_req(rs)]
 269         else:
 270             job_res_reqs = self.static_resource_req
 271
 272         with Perf(metrics, "subworkflow adjust"):
 273             joborder_resolved = copy.deepcopy(joborder)
 274             joborder_keepmount = copy.deepcopy(joborder)
 275
 276             reffiles = []
 277             visit_class(joborder_keepmount, ("File", "Directory"), reffiles.append)
 278
 279             mapper = ArvPathMapper(self.arvrunner, reffiles+self.wf_reffiles, runtimeContext.basedir,
 280                                    "/keep/%s",
 281                                    "/keep/%s/%s")
 282
 283             # For containers API, we need to make sure any extra
 284             # referenced files (ie referenced by the workflow but
 285             # not in the inputs) are included in the mounts.
 286             if self.wf_reffiles:
 287                 runtimeContext = runtimeContext.copy()
 288                 runtimeContext.extra_reffiles = copy.deepcopy(self.wf_reffiles)
 289
 290             def keepmount(obj):
 291                 remove_redundant_fields(obj)
 292                 with SourceLine(obj, None, WorkflowException, logger.isEnabledFor(logging.DEBUG)):
 293                     if "location" not in obj:
 294                         raise WorkflowException("%s object is missing required 'location' field: %s" % (obj["class"], obj))
 295                 with SourceLine(obj, "location", WorkflowException, logger.isEnabledFor(logging.DEBUG)):
 296                     if obj["location"].startswith("keep:"):
 297                         obj["location"] = mapper.mapper(obj["location"]).target
 298                         if "listing" in obj:
 299                             del obj["listing"]
 300                     elif obj["location"].startswith("_:"):
 301                         del obj["location"]
 302                     else:
 303                         raise WorkflowException("Location is not a keep reference or a literal: '%s'" % obj["location"])
 304
 305             visit_class(joborder_keepmount, ("File", "Directory"), keepmount)
 306
 307             def resolved(obj):
 308                 if obj["location"].startswith("keep:"):
 309                     obj["location"] = mapper.mapper(obj["location"]).resolved
 310
 311             visit_class(joborder_resolved, ("File", "Directory"), resolved)
 312
 313             if self.wf_pdh is None:
 314                 adjustFileObjs(packed, keepmount)
 315                 adjustDirObjs(packed, keepmount)
 316                 self.wf_pdh = upload_workflow_collection(self.arvrunner, shortname(self.tool["id"]), packed, runtimeContext)
 317
 318         self.loadingContext = self.loadingContext.copy()
 319         self.loadingContext.metadata = self.loadingContext.metadata.copy()
 320         self.loadingContext.metadata["http://commonwl.org/cwltool#original_cwlVersion"] = "v1.0"
 321
 322         if len(job_res_reqs) == 1:
 323             # RAM request needs to be at least 128 MiB or the workflow
 324             # runner itself won't run reliably.
 325             if job_res_reqs[0].get("ramMin", 1024) < 128:
 326                 job_res_reqs[0]["ramMin"] = 128
 327
 328         arguments = ["--no-container", "--move-outputs", "--preserve-entire-environment", "workflow.cwl", "cwl.input.yml"]
 329         if runtimeContext.debug:
 330             arguments.insert(0, '--debug')
 331
 332         wf_runner = cmap({
 333             "class": "CommandLineTool",
 334             "baseCommand": "cwltool",
 335             "inputs": self.tool["inputs"],
 336             "outputs": self.tool["outputs"],
 337             "stdout": "cwl.output.json",
 338             "requirements": self.requirements+job_res_reqs+[
 339                 {"class": "InlineJavascriptRequirement"},
 340                 {
 341                 "class": "InitialWorkDirRequirement",
 342                 "listing": [{
 343                         "entryname": "workflow.cwl",
 344                         "entry": '$({"class": "File", "location": "keep:%s/workflow.cwl"})' % self.wf_pdh
 345                     }, {
 346                         "entryname": "cwl.input.yml",
 347                         "entry": json.dumps(joborder_keepmount, indent=2, sort_keys=True, separators=(',',': ')).replace("\\", "\\\\").replace('$(', '\$(').replace('${', '\${')
 348                     }]
 349             }],
 350             "hints": self.hints,
 351             "arguments": arguments,
 352             "id": "#"
 353         })
 354         return ArvadosCommandTool(self.arvrunner, wf_runner, self.loadingContext).job(joborder_resolved, output_callback, runtimeContext)
 355
 356     def make_workflow_step(self,
 357                            toolpath_object,      # type: Dict[Text, Any]
 358                            pos,                  # type: int
 359                            loadingContext,       # type: LoadingContext
 360                            *argc,
 361                            **argv
 362     ):
 363         # (...) -> WorkflowStep
 364         return ArvadosWorkflowStep(toolpath_object, pos, loadingContext, self.arvrunner, *argc, **argv)