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, git_info, tool):
  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     toolname = tool.tool.get("label") or tool.metadata.get("label") or os.path.basename(tool.tool["id"])
  52     if git_info and git_info.get("http://arvados.org/cwl#gitDescribe"):
  53         toolname = "%s (%s)" % (toolname, git_info.get("http://arvados.org/cwl#gitDescribe"))
  54
  55     existing = arvRunner.api.collections().list(filters=[["portable_data_hash", "=", pdh], ["owner_uuid", "=", project_uuid]]).execute(num_retries=arvRunner.num_retries)
  56     if len(existing["items"]) == 0:
  57         col.save_new(name=toolname, owner_uuid=project_uuid, ensure_unique_name=True)
  58
  59     # now construct the wrapper
  60
  61     step = {
  62         "id": "#main/" + toolname,
  63         "in": [],
  64         "out": [],
  65         "run": "keep:%s/workflow.json#main" % pdh,
  66         "label": name
  67     }
  68
  69     newinputs = []
  70     for i in main["inputs"]:
  71         inp = {}
  72         # Make sure to only copy known fields that are meaningful at
  73         # the workflow level. In practice this ensures that if we're
  74         # wrapping a CommandLineTool we don't grab inputBinding.
  75         # Right now also excludes extension fields, which is fine,
  76         # Arvados doesn't currently look for any extension fields on
  77         # input parameters.
  78         for f in ("type", "label", "secondaryFiles", "streamable",
  79                   "doc", "id", "format", "loadContents",
  80                   "loadListing", "default"):
  81             if f in i:
  82                 inp[f] = i[f]
  83         newinputs.append(inp)
  84
  85     wrapper = {
  86         "class": "Workflow",
  87         "id": "#main",
  88         "inputs": newinputs,
  89         "outputs": [],
  90         "steps": [step]
  91     }
  92
  93     for i in main["inputs"]:
  94         step["in"].append({
  95             "id": "#main/step/%s" % shortname(i["id"]),
  96             "source": i["id"]
  97         })
  98
  99     for i in main["outputs"]:
 100         step["out"].append({"id": "#main/step/%s" % shortname(i["id"])})
 101         wrapper["outputs"].append({"outputSource": "#main/step/%s" % shortname(i["id"]),
 102                                    "type": i["type"],
 103                                    "id": i["id"]})
 104
 105     wrapper["requirements"] = [{"class": "SubworkflowFeatureRequirement"}]
 106
 107     if main.get("requirements"):
 108         wrapper["requirements"].extend(main["requirements"])
 109     if main.get("hints"):
 110         wrapper["hints"] = main["hints"]
 111
 112     doc = {"cwlVersion": "v1.2", "$graph": [wrapper]}
 113
 114     if git_info:
 115         for g in git_info:
 116             doc[g] = git_info[g]
 117
 118     return json.dumps(doc, sort_keys=True, indent=4, separators=(',',': '))
 119
 120 def upload_workflow(arvRunner, tool, job_order, project_uuid,
 121                     runtimeContext, uuid=None,
 122                     submit_runner_ram=0, name=None, merged_map=None,
 123                     submit_runner_image=None,
 124                     git_info=None):
 125
 126     packed = packed_workflow(arvRunner, tool, merged_map, runtimeContext, git_info)
 127
 128     adjustDirObjs(job_order, trim_listing)
 129     adjustFileObjs(job_order, trim_anonymous_location)
 130     adjustDirObjs(job_order, trim_anonymous_location)
 131
 132     main = [p for p in packed["$graph"] if p["id"] == "#main"][0]
 133     for inp in main["inputs"]:
 134         sn = shortname(inp["id"])
 135         if sn in job_order:
 136             inp["default"] = job_order[sn]
 137
 138     if not name:
 139         name = tool.tool.get("label", os.path.basename(tool.tool["id"]))
 140
 141     upload_dependencies(arvRunner, name, tool.doc_loader,
 142                         packed, tool.tool["id"], False,
 143                         runtimeContext)
 144
 145     wf_runner_resources = None
 146
 147     hints = main.get("hints", [])
 148     found = False
 149     for h in hints:
 150         if h["class"] == "http://arvados.org/cwl#WorkflowRunnerResources":
 151             wf_runner_resources = h
 152             found = True
 153             break
 154     if not found:
 155         wf_runner_resources = {"class": "http://arvados.org/cwl#WorkflowRunnerResources"}
 156         hints.append(wf_runner_resources)
 157
 158     wf_runner_resources["acrContainerImage"] = arvados_jobs_image(arvRunner,
 159                                                                   submit_runner_image or "arvados/jobs:"+__version__,
 160                                                                   runtimeContext)
 161
 162     if submit_runner_ram:
 163         wf_runner_resources["ramMin"] = submit_runner_ram
 164
 165     main["hints"] = hints
 166
 167     wrapper = make_wrapper_workflow(arvRunner, main, packed, project_uuid, name, git_info, tool)
 168
 169     body = {
 170         "workflow": {
 171             "name": name,
 172             "description": tool.tool.get("doc", ""),
 173             "definition": wrapper
 174         }}
 175     if project_uuid:
 176         body["workflow"]["owner_uuid"] = project_uuid
 177
 178     if uuid:
 179         call = arvRunner.api.workflows().update(uuid=uuid, body=body)
 180     else:
 181         call = arvRunner.api.workflows().create(body=body)
 182     return call.execute(num_retries=arvRunner.num_retries)["uuid"]
 183
 184 def dedup_reqs(reqs):
 185     dedup = {}
 186     for r in reversed(reqs):
 187         if r["class"] not in dedup and not r["class"].startswith("http://arvados.org/cwl#"):
 188             dedup[r["class"]] = r
 189     return [dedup[r] for r in sorted(dedup.keys())]
 190
 191 def get_overall_res_req(res_reqs):
 192     """Take the overall of a list of ResourceRequirement,
 193     i.e., the max of coresMin, coresMax, ramMin, ramMax, tmpdirMin, tmpdirMax
 194     and the sum of outdirMin, outdirMax."""
 195
 196     all_res_req = {}
 197     exception_msgs = []
 198     for a in max_res_pars + sum_res_pars:
 199         all_res_req[a] = []
 200         for res_req in res_reqs:
 201             if a in res_req:
 202                 if isinstance(res_req[a], int): # integer check
 203                     all_res_req[a].append(res_req[a])
 204                 else:
 205                     msg = SourceLine(res_req, a).makeError(
 206                     "Non-top-level ResourceRequirement in single container cannot have expressions")
 207                     exception_msgs.append(msg)
 208     if exception_msgs:
 209         raise WorkflowException("\n".join(exception_msgs))
 210     else:
 211         overall_res_req = {}
 212         for a in all_res_req:
 213             if all_res_req[a]:
 214                 if a in max_res_pars:
 215                     overall_res_req[a] = max(all_res_req[a])
 216                 elif a in sum_res_pars:
 217                     overall_res_req[a] = sum(all_res_req[a])
 218         if overall_res_req:
 219             overall_res_req["class"] = "ResourceRequirement"
 220         return cmap(overall_res_req)
 221
 222 class ArvadosWorkflowStep(WorkflowStep):
 223     def __init__(self,
 224                  toolpath_object,      # type: Dict[Text, Any]
 225                  pos,                  # type: int
 226                  loadingContext,       # type: LoadingContext
 227                  arvrunner,
 228                  *argc,
 229                  **argv
 230                 ):  # type: (...) -> None
 231
 232         if arvrunner.fast_submit:
 233             self.tool = toolpath_object
 234             self.tool["inputs"] = []
 235             self.tool["outputs"] = []
 236         else:
 237             super(ArvadosWorkflowStep, self).__init__(toolpath_object, pos, loadingContext, *argc, **argv)
 238             self.tool["class"] = "WorkflowStep"
 239         self.arvrunner = arvrunner
 240
 241     def job(self, joborder, output_callback, runtimeContext):
 242         runtimeContext = runtimeContext.copy()
 243         runtimeContext.toplevel = True  # Preserve behavior for #13365
 244
 245         builder = make_builder({shortname(k): v for k,v in viewitems(joborder)}, self.hints, self.requirements,
 246                                runtimeContext, self.metadata)
 247         runtimeContext = set_cluster_target(self.tool, self.arvrunner, builder, runtimeContext)
 248         return super(ArvadosWorkflowStep, self).job(joborder, output_callback, runtimeContext)
 249
 250
 251 class ArvadosWorkflow(Workflow):
 252     """Wrap cwltool Workflow to override selected methods."""
 253
 254     def __init__(self, arvrunner, toolpath_object, loadingContext):
 255         self.arvrunner = arvrunner
 256         self.wf_pdh = None
 257         self.dynamic_resource_req = []
 258         self.static_resource_req = []
 259         self.wf_reffiles = []
 260         self.loadingContext = loadingContext
 261         super(ArvadosWorkflow, self).__init__(toolpath_object, loadingContext)
 262         self.cluster_target_req, _ = self.get_requirement("http://arvados.org/cwl#ClusterTarget")
 263
 264     def job(self, joborder, output_callback, runtimeContext):
 265
 266         builder = make_builder(joborder, self.hints, self.requirements, runtimeContext, self.metadata)
 267         runtimeContext = set_cluster_target(self.tool, self.arvrunner, builder, runtimeContext)
 268
 269         req, _ = self.get_requirement("http://arvados.org/cwl#RunInSingleContainer")
 270         if not req:
 271             return super(ArvadosWorkflow, self).job(joborder, output_callback, runtimeContext)
 272
 273         # RunInSingleContainer is true
 274
 275         with SourceLine(self.tool, None, WorkflowException, logger.isEnabledFor(logging.DEBUG)):
 276             if "id" not in self.tool:
 277                 raise WorkflowException("%s object must have 'id'" % (self.tool["class"]))
 278
 279         discover_secondary_files(self.arvrunner.fs_access, builder,
 280                                  self.tool["inputs"], joborder)
 281         normalizeFilesDirs(joborder)
 282
 283         with Perf(metrics, "subworkflow upload_deps"):
 284             upload_dependencies(self.arvrunner,
 285                                 os.path.basename(joborder.get("id", "#")),
 286                                 self.doc_loader,
 287                                 joborder,
 288                                 joborder.get("id", "#"),
 289                                 False,
 290                                 runtimeContext)
 291
 292             if self.wf_pdh is None:
 293                 packed = pack(self.loadingContext, self.tool["id"], loader=self.doc_loader)
 294
 295                 for p in packed["$graph"]:
 296                     if p["id"] == "#main":
 297                         p["requirements"] = dedup_reqs(self.requirements)
 298                         p["hints"] = dedup_reqs(self.hints)
 299
 300                 def visit(item):
 301                     if "requirements" in item:
 302                         item["requirements"] = [i for i in item["requirements"] if i["class"] != "DockerRequirement"]
 303                     for t in ("hints", "requirements"):
 304                         if t not in item:
 305                             continue
 306                         for req in item[t]:
 307                             if req["class"] == "ResourceRequirement":
 308                                 dyn = False
 309                                 for k in max_res_pars + sum_res_pars:
 310                                     if k in req:
 311                                         if isinstance(req[k], basestring):
 312                                             if item["id"] == "#main":
 313                                                 # only the top-level requirements/hints may contain expressions
 314                                                 self.dynamic_resource_req.append(req)
 315                                                 dyn = True
 316                                                 break
 317                                             else:
 318                                                 with SourceLine(req, k, WorkflowException):
 319                                                     raise WorkflowException("Non-top-level ResourceRequirement in single container cannot have expressions")
 320                                 if not dyn:
 321                                     self.static_resource_req.append(req)
 322
 323                 visit_class(packed["$graph"], ("Workflow", "CommandLineTool"), visit)
 324
 325                 if self.static_resource_req:
 326                     self.static_resource_req = [get_overall_res_req(self.static_resource_req)]
 327
 328                 upload_dependencies(self.arvrunner,
 329                                     runtimeContext.name,
 330                                     self.doc_loader,
 331                                     packed,
 332                                     self.tool["id"],
 333                                     False,
 334                                     runtimeContext)
 335
 336                 # Discover files/directories referenced by the
 337                 # workflow (mainly "default" values)
 338                 visit_class(packed, ("File", "Directory"), self.wf_reffiles.append)
 339
 340
 341         if self.dynamic_resource_req:
 342             # Evaluate dynamic resource requirements using current builder
 343             rs = copy.copy(self.static_resource_req)
 344             for dyn_rs in self.dynamic_resource_req:
 345                 eval_req = {"class": "ResourceRequirement"}
 346                 for a in max_res_pars + sum_res_pars:
 347                     if a in dyn_rs:
 348                         eval_req[a] = builder.do_eval(dyn_rs[a])
 349                 rs.append(eval_req)
 350             job_res_reqs = [get_overall_res_req(rs)]
 351         else:
 352             job_res_reqs = self.static_resource_req
 353
 354         with Perf(metrics, "subworkflow adjust"):
 355             joborder_resolved = copy.deepcopy(joborder)
 356             joborder_keepmount = copy.deepcopy(joborder)
 357
 358             reffiles = []
 359             visit_class(joborder_keepmount, ("File", "Directory"), reffiles.append)
 360
 361             mapper = ArvPathMapper(self.arvrunner, reffiles+self.wf_reffiles, runtimeContext.basedir,
 362                                    "/keep/%s",
 363                                    "/keep/%s/%s")
 364
 365             # For containers API, we need to make sure any extra
 366             # referenced files (ie referenced by the workflow but
 367             # not in the inputs) are included in the mounts.
 368             if self.wf_reffiles:
 369                 runtimeContext = runtimeContext.copy()
 370                 runtimeContext.extra_reffiles = copy.deepcopy(self.wf_reffiles)
 371
 372             def keepmount(obj):
 373                 remove_redundant_fields(obj)
 374                 with SourceLine(obj, None, WorkflowException, logger.isEnabledFor(logging.DEBUG)):
 375                     if "location" not in obj:
 376                         raise WorkflowException("%s object is missing required 'location' field: %s" % (obj["class"], obj))
 377                 with SourceLine(obj, "location", WorkflowException, logger.isEnabledFor(logging.DEBUG)):
 378                     if obj["location"].startswith("keep:"):
 379                         obj["location"] = mapper.mapper(obj["location"]).target
 380                         if "listing" in obj:
 381                             del obj["listing"]
 382                     elif obj["location"].startswith("_:"):
 383                         del obj["location"]
 384                     else:
 385                         raise WorkflowException("Location is not a keep reference or a literal: '%s'" % obj["location"])
 386
 387             visit_class(joborder_keepmount, ("File", "Directory"), keepmount)
 388
 389             def resolved(obj):
 390                 if obj["location"].startswith("keep:"):
 391                     obj["location"] = mapper.mapper(obj["location"]).resolved
 392
 393             visit_class(joborder_resolved, ("File", "Directory"), resolved)
 394
 395             if self.wf_pdh is None:
 396                 adjustFileObjs(packed, keepmount)
 397                 adjustDirObjs(packed, keepmount)
 398                 self.wf_pdh = upload_workflow_collection(self.arvrunner, shortname(self.tool["id"]), packed, runtimeContext)
 399
 400         self.loadingContext = self.loadingContext.copy()
 401         self.loadingContext.metadata = self.loadingContext.metadata.copy()
 402         self.loadingContext.metadata["http://commonwl.org/cwltool#original_cwlVersion"] = "v1.0"
 403
 404         if len(job_res_reqs) == 1:
 405             # RAM request needs to be at least 128 MiB or the workflow
 406             # runner itself won't run reliably.
 407             if job_res_reqs[0].get("ramMin", 1024) < 128:
 408                 job_res_reqs[0]["ramMin"] = 128
 409
 410         arguments = ["--no-container", "--move-outputs", "--preserve-entire-environment", "workflow.cwl", "cwl.input.yml"]
 411         if runtimeContext.debug:
 412             arguments.insert(0, '--debug')
 413
 414         wf_runner = cmap({
 415             "class": "CommandLineTool",
 416             "baseCommand": "cwltool",
 417             "inputs": self.tool["inputs"],
 418             "outputs": self.tool["outputs"],
 419             "stdout": "cwl.output.json",
 420             "requirements": self.requirements+job_res_reqs+[
 421                 {"class": "InlineJavascriptRequirement"},
 422                 {
 423                 "class": "InitialWorkDirRequirement",
 424                 "listing": [{
 425                         "entryname": "workflow.cwl",
 426                         "entry": '$({"class": "File", "location": "keep:%s/workflow.cwl"})' % self.wf_pdh
 427                     }, {
 428                         "entryname": "cwl.input.yml",
 429                         "entry": json.dumps(joborder_keepmount, indent=2, sort_keys=True, separators=(',',': ')).replace("\\", "\\\\").replace('$(', '\$(').replace('${', '\${')
 430                     }]
 431             }],
 432             "hints": self.hints,
 433             "arguments": arguments,
 434             "id": "#"
 435         })
 436         return ArvadosCommandTool(self.arvrunner, wf_runner, self.loadingContext).job(joborder_resolved, output_callback, runtimeContext)
 437
 438     def make_workflow_step(self,
 439                            toolpath_object,      # type: Dict[Text, Any]
 440                            pos,                  # type: int
 441                            loadingContext,       # type: LoadingContext
 442                            *argc,
 443                            **argv
 444     ):
 445         # (...) -> WorkflowStep
 446         return ArvadosWorkflowStep(toolpath_object, pos, loadingContext, self.arvrunner, *argc, **argv)