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 import arvados.collection
  17
  18 from cwltool.pack import pack
  19 from cwltool.load_tool import fetch_document, resolve_and_validate_document
  20 from cwltool.process import shortname
  21 from cwltool.workflow import Workflow, WorkflowException, WorkflowStep
  22 from cwltool.utils import adjustFileObjs, adjustDirObjs, visit_class, normalizeFilesDirs
  23 from cwltool.context import LoadingContext
  24
  25 import ruamel.yaml as yaml
  26
  27 from .runner import (upload_dependencies, packed_workflow, upload_workflow_collection,
  28                      trim_anonymous_location, remove_redundant_fields, discover_secondary_files,
  29                      make_builder, arvados_jobs_image)
  30 from .pathmapper import ArvPathMapper, trim_listing
  31 from .arvtool import ArvadosCommandTool, set_cluster_target
  32 from ._version import __version__
  33
  34 from .perf import Perf
  35
  36 logger = logging.getLogger('arvados.cwl-runner')
  37 metrics = logging.getLogger('arvados.cwl-runner.metrics')
  38
  39 max_res_pars = ("coresMin", "coresMax", "ramMin", "ramMax", "tmpdirMin", "tmpdirMax")
  40 sum_res_pars = ("outdirMin", "outdirMax")
  41
  42 def make_wrapper_workflow(arvRunner, main, packed, project_uuid, name):
  43     col = arvados.collection.Collection(api_client=arvRunner.api,
  44                                         keep_client=arvRunner.keep_client)
  45
  46     with col.open("workflow.json", "wt") as f:
  47         json.dump(packed, f, sort_keys=True, indent=4, separators=(',',': '))
  48
  49     pdh = col.portable_data_hash()
  50
  51     existing = arvRunner.api.collections().list(filters=[["portable_data_hash", "=", pdh], ["owner_uuid", "=", project_uuid]]).execute(num_retries=arvRunner.num_retries)
  52     if len(existing["items"]) == 0:
  53         col.save_new(name=name, owner_uuid=project_uuid, ensure_unique_name=True)
  54
  55     # now construct the wrapper
  56
  57     step = {
  58         "id": "#main/step",
  59         "in": [],
  60         "out": [],
  61         "run": "keep:%s/workflow.json#main" % pdh
  62     }
  63
  64     newinputs = []
  65     for i in main["inputs"]:
  66         inp = {}
  67         # Make sure to only copy known fields that are meaningful at
  68         # the workflow level. In practice this ensures that if we're
  69         # wrapping a CommandLineTool we don't grab inputBinding.
  70         # Right now also excludes extension fields, which is fine,
  71         # Arvados doesn't currently look for any extension fields on
  72         # input parameters.
  73         for f in ("type", "label", "secondaryFiles", "streamable",
  74                   "doc", "id", "format", "loadContents",
  75                   "loadListing", "default"):
  76             if f in i:
  77                 inp[f] = i[f]
  78         newinputs.append(inp)
  79
  80     wrapper = {
  81         "class": "Workflow",
  82         "id": "#main",
  83         "inputs": newinputs,
  84         "outputs": [],
  85         "steps": [step]
  86     }
  87
  88     for i in main["inputs"]:
  89         step["in"].append({
  90             "id": "#main/step/%s" % shortname(i["id"]),
  91             "source": i["id"]
  92         })
  93
  94     for i in main["outputs"]:
  95         step["out"].append({"id": "#main/step/%s" % shortname(i["id"])})
  96         wrapper["outputs"].append({"outputSource": "#main/step/%s" % shortname(i["id"]),
  97                                    "type": i["type"],
  98                                    "id": i["id"]})
  99
 100     wrapper["requirements"] = [{"class": "SubworkflowFeatureRequirement"}]
 101
 102     if main.get("requirements"):
 103         wrapper["requirements"].extend(main["requirements"])
 104     if main.get("hints"):
 105         wrapper["hints"] = main["hints"]
 106
 107     return json.dumps({"cwlVersion": "v1.2", "$graph": [wrapper]}, sort_keys=True, indent=4, separators=(',',': '))
 108
 109 def upload_workflow(arvRunner, tool, job_order, project_uuid,
 110                     runtimeContext, uuid=None,
 111                     submit_runner_ram=0, name=None, merged_map=None,
 112                     submit_runner_image=None,
 113                     git_info=None):
 114
 115     packed = packed_workflow(arvRunner, tool, merged_map, runtimeContext, git_info)
 116
 117     adjustDirObjs(job_order, trim_listing)
 118     adjustFileObjs(job_order, trim_anonymous_location)
 119     adjustDirObjs(job_order, trim_anonymous_location)
 120
 121     main = [p for p in packed["$graph"] if p["id"] == "#main"][0]
 122     for inp in main["inputs"]:
 123         sn = shortname(inp["id"])
 124         if sn in job_order:
 125             inp["default"] = job_order[sn]
 126
 127     if not name:
 128         name = tool.tool.get("label", os.path.basename(tool.tool["id"]))
 129
 130     upload_dependencies(arvRunner, name, tool.doc_loader,
 131                         packed, tool.tool["id"], False,
 132                         runtimeContext)
 133
 134     wf_runner_resources = None
 135
 136     hints = main.get("hints", [])
 137     found = False
 138     for h in hints:
 139         if h["class"] == "http://arvados.org/cwl#WorkflowRunnerResources":
 140             wf_runner_resources = h
 141             found = True
 142             break
 143     if not found:
 144         wf_runner_resources = {"class": "http://arvados.org/cwl#WorkflowRunnerResources"}
 145         hints.append(wf_runner_resources)
 146
 147     wf_runner_resources["acrContainerImage"] = arvados_jobs_image(arvRunner,
 148                                                                   submit_runner_image or "arvados/jobs:"+__version__,
 149                                                                   runtimeContext)
 150
 151     if submit_runner_ram:
 152         wf_runner_resources["ramMin"] = submit_runner_ram
 153
 154     main["hints"] = hints
 155
 156     wrapper = make_wrapper_workflow(arvRunner, main, packed, project_uuid, name)
 157
 158     body = {
 159         "workflow": {
 160             "name": name,
 161             "description": tool.tool.get("doc", ""),
 162             "definition": wrapper
 163         }}
 164     if project_uuid:
 165         body["workflow"]["owner_uuid"] = project_uuid
 166
 167     if uuid:
 168         call = arvRunner.api.workflows().update(uuid=uuid, body=body)
 169     else:
 170         call = arvRunner.api.workflows().create(body=body)
 171     return call.execute(num_retries=arvRunner.num_retries)["uuid"]
 172
 173 def dedup_reqs(reqs):
 174     dedup = {}
 175     for r in reversed(reqs):
 176         if r["class"] not in dedup and not r["class"].startswith("http://arvados.org/cwl#"):
 177             dedup[r["class"]] = r
 178     return [dedup[r] for r in sorted(dedup.keys())]
 179
 180 def get_overall_res_req(res_reqs):
 181     """Take the overall of a list of ResourceRequirement,
 182     i.e., the max of coresMin, coresMax, ramMin, ramMax, tmpdirMin, tmpdirMax
 183     and the sum of outdirMin, outdirMax."""
 184
 185     all_res_req = {}
 186     exception_msgs = []
 187     for a in max_res_pars + sum_res_pars:
 188         all_res_req[a] = []
 189         for res_req in res_reqs:
 190             if a in res_req:
 191                 if isinstance(res_req[a], int): # integer check
 192                     all_res_req[a].append(res_req[a])
 193                 else:
 194                     msg = SourceLine(res_req, a).makeError(
 195                     "Non-top-level ResourceRequirement in single container cannot have expressions")
 196                     exception_msgs.append(msg)
 197     if exception_msgs:
 198         raise WorkflowException("\n".join(exception_msgs))
 199     else:
 200         overall_res_req = {}
 201         for a in all_res_req:
 202             if all_res_req[a]:
 203                 if a in max_res_pars:
 204                     overall_res_req[a] = max(all_res_req[a])
 205                 elif a in sum_res_pars:
 206                     overall_res_req[a] = sum(all_res_req[a])
 207         if overall_res_req:
 208             overall_res_req["class"] = "ResourceRequirement"
 209         return cmap(overall_res_req)
 210
 211 class ArvadosWorkflowStep(WorkflowStep):
 212     def __init__(self,
 213                  toolpath_object,      # type: Dict[Text, Any]
 214                  pos,                  # type: int
 215                  loadingContext,       # type: LoadingContext
 216                  arvrunner,
 217                  *argc,
 218                  **argv
 219                 ):  # type: (...) -> None
 220
 221         if arvrunner.fast_submit:
 222             self.tool = toolpath_object
 223             self.tool["inputs"] = []
 224             self.tool["outputs"] = []
 225         else:
 226             super(ArvadosWorkflowStep, self).__init__(toolpath_object, pos, loadingContext, *argc, **argv)
 227             self.tool["class"] = "WorkflowStep"
 228         self.arvrunner = arvrunner
 229
 230     def job(self, joborder, output_callback, runtimeContext):
 231         runtimeContext = runtimeContext.copy()
 232         runtimeContext.toplevel = True  # Preserve behavior for #13365
 233
 234         builder = make_builder({shortname(k): v for k,v in viewitems(joborder)}, self.hints, self.requirements,
 235                                runtimeContext, self.metadata)
 236         runtimeContext = set_cluster_target(self.tool, self.arvrunner, builder, runtimeContext)
 237         return super(ArvadosWorkflowStep, self).job(joborder, output_callback, runtimeContext)
 238
 239
 240 class ArvadosWorkflow(Workflow):
 241     """Wrap cwltool Workflow to override selected methods."""
 242
 243     def __init__(self, arvrunner, toolpath_object, loadingContext):
 244         self.arvrunner = arvrunner
 245         self.wf_pdh = None
 246         self.dynamic_resource_req = []
 247         self.static_resource_req = []
 248         self.wf_reffiles = []
 249         self.loadingContext = loadingContext
 250         super(ArvadosWorkflow, self).__init__(toolpath_object, loadingContext)
 251         self.cluster_target_req, _ = self.get_requirement("http://arvados.org/cwl#ClusterTarget")
 252
 253     def job(self, joborder, output_callback, runtimeContext):
 254
 255         builder = make_builder(joborder, self.hints, self.requirements, runtimeContext, self.metadata)
 256         runtimeContext = set_cluster_target(self.tool, self.arvrunner, builder, runtimeContext)
 257
 258         req, _ = self.get_requirement("http://arvados.org/cwl#RunInSingleContainer")
 259         if not req:
 260             return super(ArvadosWorkflow, self).job(joborder, output_callback, runtimeContext)
 261
 262         # RunInSingleContainer is true
 263
 264         with SourceLine(self.tool, None, WorkflowException, logger.isEnabledFor(logging.DEBUG)):
 265             if "id" not in self.tool:
 266                 raise WorkflowException("%s object must have 'id'" % (self.tool["class"]))
 267
 268         discover_secondary_files(self.arvrunner.fs_access, builder,
 269                                  self.tool["inputs"], joborder)
 270         normalizeFilesDirs(joborder)
 271
 272         with Perf(metrics, "subworkflow upload_deps"):
 273             upload_dependencies(self.arvrunner,
 274                                 os.path.basename(joborder.get("id", "#")),
 275                                 self.doc_loader,
 276                                 joborder,
 277                                 joborder.get("id", "#"),
 278                                 False,
 279                                 runtimeContext)
 280
 281             if self.wf_pdh is None:
 282                 packed = pack(self.loadingContext, self.tool["id"], loader=self.doc_loader)
 283
 284                 for p in packed["$graph"]:
 285                     if p["id"] == "#main":
 286                         p["requirements"] = dedup_reqs(self.requirements)
 287                         p["hints"] = dedup_reqs(self.hints)
 288
 289                 def visit(item):
 290                     if "requirements" in item:
 291                         item["requirements"] = [i for i in item["requirements"] if i["class"] != "DockerRequirement"]
 292                     for t in ("hints", "requirements"):
 293                         if t not in item:
 294                             continue
 295                         for req in item[t]:
 296                             if req["class"] == "ResourceRequirement":
 297                                 dyn = False
 298                                 for k in max_res_pars + sum_res_pars:
 299                                     if k in req:
 300                                         if isinstance(req[k], basestring):
 301                                             if item["id"] == "#main":
 302                                                 # only the top-level requirements/hints may contain expressions
 303                                                 self.dynamic_resource_req.append(req)
 304                                                 dyn = True
 305                                                 break
 306                                             else:
 307                                                 with SourceLine(req, k, WorkflowException):
 308                                                     raise WorkflowException("Non-top-level ResourceRequirement in single container cannot have expressions")
 309                                 if not dyn:
 310                                     self.static_resource_req.append(req)
 311
 312                 visit_class(packed["$graph"], ("Workflow", "CommandLineTool"), visit)
 313
 314                 if self.static_resource_req:
 315                     self.static_resource_req = [get_overall_res_req(self.static_resource_req)]
 316
 317                 upload_dependencies(self.arvrunner,
 318                                     runtimeContext.name,
 319                                     self.doc_loader,
 320                                     packed,
 321                                     self.tool["id"],
 322                                     False,
 323                                     runtimeContext)
 324
 325                 # Discover files/directories referenced by the
 326                 # workflow (mainly "default" values)
 327                 visit_class(packed, ("File", "Directory"), self.wf_reffiles.append)
 328
 329
 330         if self.dynamic_resource_req:
 331             # Evaluate dynamic resource requirements using current builder
 332             rs = copy.copy(self.static_resource_req)
 333             for dyn_rs in self.dynamic_resource_req:
 334                 eval_req = {"class": "ResourceRequirement"}
 335                 for a in max_res_pars + sum_res_pars:
 336                     if a in dyn_rs:
 337                         eval_req[a] = builder.do_eval(dyn_rs[a])
 338                 rs.append(eval_req)
 339             job_res_reqs = [get_overall_res_req(rs)]
 340         else:
 341             job_res_reqs = self.static_resource_req
 342
 343         with Perf(metrics, "subworkflow adjust"):
 344             joborder_resolved = copy.deepcopy(joborder)
 345             joborder_keepmount = copy.deepcopy(joborder)
 346
 347             reffiles = []
 348             visit_class(joborder_keepmount, ("File", "Directory"), reffiles.append)
 349
 350             mapper = ArvPathMapper(self.arvrunner, reffiles+self.wf_reffiles, runtimeContext.basedir,
 351                                    "/keep/%s",
 352                                    "/keep/%s/%s")
 353
 354             # For containers API, we need to make sure any extra
 355             # referenced files (ie referenced by the workflow but
 356             # not in the inputs) are included in the mounts.
 357             if self.wf_reffiles:
 358                 runtimeContext = runtimeContext.copy()
 359                 runtimeContext.extra_reffiles = copy.deepcopy(self.wf_reffiles)
 360
 361             def keepmount(obj):
 362                 remove_redundant_fields(obj)
 363                 with SourceLine(obj, None, WorkflowException, logger.isEnabledFor(logging.DEBUG)):
 364                     if "location" not in obj:
 365                         raise WorkflowException("%s object is missing required 'location' field: %s" % (obj["class"], obj))
 366                 with SourceLine(obj, "location", WorkflowException, logger.isEnabledFor(logging.DEBUG)):
 367                     if obj["location"].startswith("keep:"):
 368                         obj["location"] = mapper.mapper(obj["location"]).target
 369                         if "listing" in obj:
 370                             del obj["listing"]
 371                     elif obj["location"].startswith("_:"):
 372                         del obj["location"]
 373                     else:
 374                         raise WorkflowException("Location is not a keep reference or a literal: '%s'" % obj["location"])
 375
 376             visit_class(joborder_keepmount, ("File", "Directory"), keepmount)
 377
 378             def resolved(obj):
 379                 if obj["location"].startswith("keep:"):
 380                     obj["location"] = mapper.mapper(obj["location"]).resolved
 381
 382             visit_class(joborder_resolved, ("File", "Directory"), resolved)
 383
 384             if self.wf_pdh is None:
 385                 adjustFileObjs(packed, keepmount)
 386                 adjustDirObjs(packed, keepmount)
 387                 self.wf_pdh = upload_workflow_collection(self.arvrunner, shortname(self.tool["id"]), packed, runtimeContext)
 388
 389         self.loadingContext = self.loadingContext.copy()
 390         self.loadingContext.metadata = self.loadingContext.metadata.copy()
 391         self.loadingContext.metadata["http://commonwl.org/cwltool#original_cwlVersion"] = "v1.0"
 392
 393         if len(job_res_reqs) == 1:
 394             # RAM request needs to be at least 128 MiB or the workflow
 395             # runner itself won't run reliably.
 396             if job_res_reqs[0].get("ramMin", 1024) < 128:
 397                 job_res_reqs[0]["ramMin"] = 128
 398
 399         arguments = ["--no-container", "--move-outputs", "--preserve-entire-environment", "workflow.cwl", "cwl.input.yml"]
 400         if runtimeContext.debug:
 401             arguments.insert(0, '--debug')
 402
 403         wf_runner = cmap({
 404             "class": "CommandLineTool",
 405             "baseCommand": "cwltool",
 406             "inputs": self.tool["inputs"],
 407             "outputs": self.tool["outputs"],
 408             "stdout": "cwl.output.json",
 409             "requirements": self.requirements+job_res_reqs+[
 410                 {"class": "InlineJavascriptRequirement"},
 411                 {
 412                 "class": "InitialWorkDirRequirement",
 413                 "listing": [{
 414                         "entryname": "workflow.cwl",
 415                         "entry": '$({"class": "File", "location": "keep:%s/workflow.cwl"})' % self.wf_pdh
 416                     }, {
 417                         "entryname": "cwl.input.yml",
 418                         "entry": json.dumps(joborder_keepmount, indent=2, sort_keys=True, separators=(',',': ')).replace("\\", "\\\\").replace('$(', '\$(').replace('${', '\${')
 419                     }]
 420             }],
 421             "hints": self.hints,
 422             "arguments": arguments,
 423             "id": "#"
 424         })
 425         return ArvadosCommandTool(self.arvrunner, wf_runner, self.loadingContext).job(joborder_resolved, output_callback, runtimeContext)
 426
 427     def make_workflow_step(self,
 428                            toolpath_object,      # type: Dict[Text, Any]
 429                            pos,                  # type: int
 430                            loadingContext,       # type: LoadingContext
 431                            *argc,
 432                            **argv
 433     ):
 434         # (...) -> WorkflowStep
 435         return ArvadosWorkflowStep(toolpath_object, pos, loadingContext, self.arvrunner, *argc, **argv)