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