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