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