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 import os
   6 import json
   7 import copy
   8 import logging
   9
  10 from schema_salad.sourceline import SourceLine, cmap
  11
  12 from cwltool.pack import pack
  13 from cwltool.load_tool import fetch_document
  14 from cwltool.process import shortname
  15 from cwltool.workflow import Workflow, WorkflowException, WorkflowStep
  16 from cwltool.pathmapper import adjustFileObjs, adjustDirObjs, visit_class
  17 from cwltool.context import LoadingContext
  18
  19 import ruamel.yaml as yaml
  20
  21 from .runner import (upload_dependencies, packed_workflow, upload_workflow_collection,
  22                      trim_anonymous_location, remove_redundant_fields, discover_secondary_files)
  23 from .pathmapper import ArvPathMapper, trim_listing
  24 from .arvtool import ArvadosCommandTool, set_cluster_target, make_builder
  25 from .perf import Perf
  26
  27 logger = logging.getLogger('arvados.cwl-runner')
  28 metrics = logging.getLogger('arvados.cwl-runner.metrics')
  29
  30 max_res_pars = ("coresMin", "coresMax", "ramMin", "ramMax", "tmpdirMin", "tmpdirMax")
  31 sum_res_pars = ("outdirMin", "outdirMax")
  32
  33 def upload_workflow(arvRunner, tool, job_order, project_uuid, uuid=None,
  34                     submit_runner_ram=0, name=None, merged_map=None):
  35
  36     packed = packed_workflow(arvRunner, tool, merged_map)
  37
  38     adjustDirObjs(job_order, trim_listing)
  39     adjustFileObjs(job_order, trim_anonymous_location)
  40     adjustDirObjs(job_order, trim_anonymous_location)
  41
  42     main = [p for p in packed["$graph"] if p["id"] == "#main"][0]
  43     for inp in main["inputs"]:
  44         sn = shortname(inp["id"])
  45         if sn in job_order:
  46             inp["default"] = job_order[sn]
  47
  48     if not name:
  49         name = tool.tool.get("label", os.path.basename(tool.tool["id"]))
  50
  51     upload_dependencies(arvRunner, name, tool.doc_loader,
  52                         packed, tool.tool["id"], False)
  53
  54     if submit_runner_ram:
  55         hints = main.get("hints", [])
  56         found = False
  57         for h in hints:
  58             if h["class"] == "http://arvados.org/cwl#WorkflowRunnerResources":
  59                 h["ramMin"] = submit_runner_ram
  60                 found = True
  61                 break
  62         if not found:
  63             hints.append({"class": "http://arvados.org/cwl#WorkflowRunnerResources",
  64                           "ramMin": submit_runner_ram})
  65         main["hints"] = hints
  66
  67     body = {
  68         "workflow": {
  69             "name": name,
  70             "description": tool.tool.get("doc", ""),
  71             "definition":json.dumps(packed, sort_keys=True, indent=4, separators=(',',': '))
  72         }}
  73     if project_uuid:
  74         body["workflow"]["owner_uuid"] = project_uuid
  75
  76     if uuid:
  77         call = arvRunner.api.workflows().update(uuid=uuid, body=body)
  78     else:
  79         call = arvRunner.api.workflows().create(body=body)
  80     return call.execute(num_retries=arvRunner.num_retries)["uuid"]
  81
  82 def dedup_reqs(reqs):
  83     dedup = {}
  84     for r in reversed(reqs):
  85         if r["class"] not in dedup and not r["class"].startswith("http://arvados.org/cwl#"):
  86             dedup[r["class"]] = r
  87     return [dedup[r] for r in sorted(dedup.keys())]
  88
  89 def get_overall_res_req(res_reqs):
  90     """Take the overall of a list of ResourceRequirement,
  91     i.e., the max of coresMin, coresMax, ramMin, ramMax, tmpdirMin, tmpdirMax
  92     and the sum of outdirMin, outdirMax."""
  93
  94     all_res_req = {}
  95     exception_msgs = []
  96     for a in max_res_pars + sum_res_pars:
  97         all_res_req[a] = []
  98         for res_req in res_reqs:
  99             if a in res_req:
 100                 if isinstance(res_req[a], int): # integer check
 101                     all_res_req[a].append(res_req[a])
 102                 else:
 103                     msg = SourceLine(res_req, a).makeError(
 104                     "Non-top-level ResourceRequirement in single container cannot have expressions")
 105                     exception_msgs.append(msg)
 106     if exception_msgs:
 107         raise WorkflowException("\n".join(exception_msgs))
 108     else:
 109         overall_res_req = {}
 110         for a in all_res_req:
 111             if all_res_req[a]:
 112                 if a in max_res_pars:
 113                     overall_res_req[a] = max(all_res_req[a])
 114                 elif a in sum_res_pars:
 115                     overall_res_req[a] = sum(all_res_req[a])
 116         if overall_res_req:
 117             overall_res_req["class"] = "ResourceRequirement"
 118         return cmap(overall_res_req)
 119
 120 class ArvadosWorkflowStep(WorkflowStep):
 121     def __init__(self,
 122                  toolpath_object,      # type: Dict[Text, Any]
 123                  pos,                  # type: int
 124                  loadingContext,       # type: LoadingContext
 125                  arvrunner,
 126                  *argc,
 127                  **argv
 128                 ):  # type: (...) -> None
 129
 130         super(ArvadosWorkflowStep, self).__init__(toolpath_object, pos, loadingContext, *argc, **argv)
 131         self.tool["class"] = "WorkflowStep"
 132         self.arvrunner = arvrunner
 133
 134     def job(self, joborder, output_callback, runtimeContext):
 135         runtimeContext = runtimeContext.copy()
 136         runtimeContext.toplevel = True  # Preserve behavior for #13365
 137
 138         builder = make_builder({shortname(k): v for k,v in joborder.items()}, self.hints, self.requirements, runtimeContext)
 139         runtimeContext = set_cluster_target(self.tool, self.arvrunner, builder, runtimeContext)
 140         return super(ArvadosWorkflowStep, self).job(joborder, output_callback, runtimeContext)
 141
 142
 143 class ArvadosWorkflow(Workflow):
 144     """Wrap cwltool Workflow to override selected methods."""
 145
 146     def __init__(self, arvrunner, toolpath_object, loadingContext):
 147         self.arvrunner = arvrunner
 148         self.wf_pdh = None
 149         self.dynamic_resource_req = []
 150         self.static_resource_req = []
 151         self.wf_reffiles = []
 152         self.loadingContext = loadingContext
 153         super(ArvadosWorkflow, self).__init__(toolpath_object, loadingContext)
 154         self.cluster_target_req, _ = self.get_requirement("http://arvados.org/cwl#ClusterTarget")
 155
 156     def job(self, joborder, output_callback, runtimeContext):
 157
 158         builder = make_builder(joborder, self.hints, self.requirements, runtimeContext)
 159         runtimeContext = set_cluster_target(self.tool, self.arvrunner, builder, runtimeContext)
 160
 161         req, _ = self.get_requirement("http://arvados.org/cwl#RunInSingleContainer")
 162         if not req:
 163             return super(ArvadosWorkflow, self).job(joborder, output_callback, runtimeContext)
 164
 165         # RunInSingleContainer is true
 166
 167         with SourceLine(self.tool, None, WorkflowException, logger.isEnabledFor(logging.DEBUG)):
 168             if "id" not in self.tool:
 169                 raise WorkflowException("%s object must have 'id'" % (self.tool["class"]))
 170         document_loader, workflowobj, uri = (self.doc_loader, self.doc_loader.fetch(self.tool["id"]), self.tool["id"])
 171
 172         discover_secondary_files(self.tool["inputs"], joborder)
 173
 174         with Perf(metrics, "subworkflow upload_deps"):
 175             upload_dependencies(self.arvrunner,
 176                                 os.path.basename(joborder.get("id", "#")),
 177                                 document_loader,
 178                                 joborder,
 179                                 joborder.get("id", "#"),
 180                                 False)
 181
 182             if self.wf_pdh is None:
 183                 workflowobj["requirements"] = dedup_reqs(self.requirements)
 184                 workflowobj["hints"] = dedup_reqs(self.hints)
 185
 186                 packed = pack(document_loader, workflowobj, uri, self.metadata)
 187
 188                 def visit(item):
 189                     for t in ("hints", "requirements"):
 190                         if t not in item:
 191                             continue
 192                         for req in item[t]:
 193                             if req["class"] == "ResourceRequirement":
 194                                 dyn = False
 195                                 for k in max_res_pars + sum_res_pars:
 196                                     if k in req:
 197                                         if isinstance(req[k], basestring):
 198                                             if item["id"] == "#main":
 199                                                 # only the top-level requirements/hints may contain expressions
 200                                                 self.dynamic_resource_req.append(req)
 201                                                 dyn = True
 202                                                 break
 203                                             else:
 204                                                 with SourceLine(req, k, WorkflowException):
 205                                                     raise WorkflowException("Non-top-level ResourceRequirement in single container cannot have expressions")
 206                                 if not dyn:
 207                                     self.static_resource_req.append(req)
 208                             if req["class"] == "DockerRequirement":
 209                                 if "http://arvados.org/cwl#dockerCollectionPDH" in req:
 210                                     del req["http://arvados.org/cwl#dockerCollectionPDH"]
 211
 212                 visit_class(packed["$graph"], ("Workflow", "CommandLineTool"), visit)
 213
 214                 if self.static_resource_req:
 215                     self.static_resource_req = [get_overall_res_req(self.static_resource_req)]
 216
 217                 upload_dependencies(self.arvrunner,
 218                                     runtimeContext.name,
 219                                     document_loader,
 220                                     packed,
 221                                     uri,
 222                                     False)
 223
 224                 # Discover files/directories referenced by the
 225                 # workflow (mainly "default" values)
 226                 visit_class(packed, ("File", "Directory"), self.wf_reffiles.append)
 227
 228
 229         if self.dynamic_resource_req:
 230             # Evaluate dynamic resource requirements using current builder
 231             rs = copy.copy(self.static_resource_req)
 232             for dyn_rs in self.dynamic_resource_req:
 233                 eval_req = {"class": "ResourceRequirement"}
 234                 for a in max_res_pars + sum_res_pars:
 235                     if a in dyn_rs:
 236                         eval_req[a] = builder.do_eval(dyn_rs[a])
 237                 rs.append(eval_req)
 238             job_res_reqs = [get_overall_res_req(rs)]
 239         else:
 240             job_res_reqs = self.static_resource_req
 241
 242         with Perf(metrics, "subworkflow adjust"):
 243             joborder_resolved = copy.deepcopy(joborder)
 244             joborder_keepmount = copy.deepcopy(joborder)
 245
 246             reffiles = []
 247             visit_class(joborder_keepmount, ("File", "Directory"), reffiles.append)
 248
 249             mapper = ArvPathMapper(self.arvrunner, reffiles+self.wf_reffiles, runtimeContext.basedir,
 250                                    "/keep/%s",
 251                                    "/keep/%s/%s")
 252
 253             # For containers API, we need to make sure any extra
 254             # referenced files (ie referenced by the workflow but
 255             # not in the inputs) are included in the mounts.
 256             if self.wf_reffiles:
 257                 runtimeContext = runtimeContext.copy()
 258                 runtimeContext.extra_reffiles = copy.deepcopy(self.wf_reffiles)
 259
 260             def keepmount(obj):
 261                 remove_redundant_fields(obj)
 262                 with SourceLine(obj, None, WorkflowException, logger.isEnabledFor(logging.DEBUG)):
 263                     if "location" not in obj:
 264                         raise WorkflowException("%s object is missing required 'location' field: %s" % (obj["class"], obj))
 265                 with SourceLine(obj, "location", WorkflowException, logger.isEnabledFor(logging.DEBUG)):
 266                     if obj["location"].startswith("keep:"):
 267                         obj["location"] = mapper.mapper(obj["location"]).target
 268                         if "listing" in obj:
 269                             del obj["listing"]
 270                     elif obj["location"].startswith("_:"):
 271                         del obj["location"]
 272                     else:
 273                         raise WorkflowException("Location is not a keep reference or a literal: '%s'" % obj["location"])
 274
 275             visit_class(joborder_keepmount, ("File", "Directory"), keepmount)
 276
 277             def resolved(obj):
 278                 if obj["location"].startswith("keep:"):
 279                     obj["location"] = mapper.mapper(obj["location"]).resolved
 280
 281             visit_class(joborder_resolved, ("File", "Directory"), resolved)
 282
 283             if self.wf_pdh is None:
 284                 adjustFileObjs(packed, keepmount)
 285                 adjustDirObjs(packed, keepmount)
 286                 self.wf_pdh = upload_workflow_collection(self.arvrunner, shortname(self.tool["id"]), packed)
 287
 288         wf_runner = cmap({
 289             "class": "CommandLineTool",
 290             "baseCommand": "cwltool",
 291             "inputs": self.tool["inputs"],
 292             "outputs": self.tool["outputs"],
 293             "stdout": "cwl.output.json",
 294             "requirements": self.requirements+job_res_reqs+[
 295                 {"class": "InlineJavascriptRequirement"},
 296                 {
 297                 "class": "InitialWorkDirRequirement",
 298                 "listing": [{
 299                         "entryname": "workflow.cwl",
 300                         "entry": '$({"class": "File", "location": "keep:%s/workflow.cwl"})' % self.wf_pdh
 301                     }, {
 302                         "entryname": "cwl.input.yml",
 303                         "entry": json.dumps(joborder_keepmount, indent=2, sort_keys=True, separators=(',',': ')).replace("\\", "\\\\").replace('$(', '\$(').replace('${', '\${')
 304                     }]
 305             }],
 306             "hints": self.hints,
 307             "arguments": ["--no-container", "--move-outputs", "--preserve-entire-environment", "workflow.cwl#main", "cwl.input.yml"],
 308             "id": "#"
 309         })
 310         return ArvadosCommandTool(self.arvrunner, wf_runner, self.loadingContext).job(joborder_resolved, output_callback, runtimeContext)
 311
 312     def make_workflow_step(self,
 313                            toolpath_object,      # type: Dict[Text, Any]
 314                            pos,                  # type: int
 315                            loadingContext,       # type: LoadingContext
 316                            *argc,
 317                            **argv
 318     ):
 319         # (...) -> WorkflowStep
 320         return ArvadosWorkflowStep(toolpath_object, pos, loadingContext, self.arvrunner, *argc, **argv)