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