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     if submit_runner_ram:
  56         hints = main.get("hints", [])
  57         found = False
  58         for h in hints:
  59             if h["class"] == "http://arvados.org/cwl#WorkflowRunnerResources":
  60                 h["ramMin"] = submit_runner_ram
  61                 found = True
  62                 break
  63         if not found:
  64             hints.append({"class": "http://arvados.org/cwl#WorkflowRunnerResources",
  65                           "ramMin": submit_runner_ram})
  66         main["hints"] = hints
  67
  68     body = {
  69         "workflow": {
  70             "name": name,
  71             "description": tool.tool.get("doc", ""),
  72             "definition":json.dumps(packed, sort_keys=True, indent=4, separators=(',',': '))
  73         }}
  74     if project_uuid:
  75         body["workflow"]["owner_uuid"] = project_uuid
  76
  77     if uuid:
  78         call = arvRunner.api.workflows().update(uuid=uuid, body=body)
  79     else:
  80         call = arvRunner.api.workflows().create(body=body)
  81     return call.execute(num_retries=arvRunner.num_retries)["uuid"]
  82
  83 def dedup_reqs(reqs):
  84     dedup = {}
  85     for r in reversed(reqs):
  86         if r["class"] not in dedup and not r["class"].startswith("http://arvados.org/cwl#"):
  87             dedup[r["class"]] = r
  88     return [dedup[r] for r in sorted(dedup.keys())]
  89
  90 def get_overall_res_req(res_reqs):
  91     """Take the overall of a list of ResourceRequirement,
  92     i.e., the max of coresMin, coresMax, ramMin, ramMax, tmpdirMin, tmpdirMax
  93     and the sum of outdirMin, outdirMax."""
  94
  95     all_res_req = {}
  96     exception_msgs = []
  97     for a in max_res_pars + sum_res_pars:
  98         all_res_req[a] = []
  99         for res_req in res_reqs:
 100             if a in res_req:
 101                 if isinstance(res_req[a], int): # integer check
 102                     all_res_req[a].append(res_req[a])
 103                 else:
 104                     msg = SourceLine(res_req, a).makeError(
 105                     "Non-top-level ResourceRequirement in single container cannot have expressions")
 106                     exception_msgs.append(msg)
 107     if exception_msgs:
 108         raise WorkflowException("\n".join(exception_msgs))
 109     else:
 110         overall_res_req = {}
 111         for a in all_res_req:
 112             if all_res_req[a]:
 113                 if a in max_res_pars:
 114                     overall_res_req[a] = max(all_res_req[a])
 115                 elif a in sum_res_pars:
 116                     overall_res_req[a] = sum(all_res_req[a])
 117         if overall_res_req:
 118             overall_res_req["class"] = "ResourceRequirement"
 119         return cmap(overall_res_req)
 120
 121 class ArvadosWorkflow(Workflow):
 122     """Wrap cwltool Workflow to override selected methods."""
 123
 124     def __init__(self, arvrunner, toolpath_object, loadingContext):
 125         super(ArvadosWorkflow, self).__init__(toolpath_object, loadingContext)
 126         self.arvrunner = arvrunner
 127         self.wf_pdh = None
 128         self.dynamic_resource_req = []
 129         self.static_resource_req = []
 130         self.wf_reffiles = []
 131         self.loadingContext = loadingContext
 132
 133     def job(self, joborder, output_callback, runtimeContext):
 134         req, _ = self.get_requirement("http://arvados.org/cwl#RunInSingleContainer")
 135         if req:
 136             with SourceLine(self.tool, None, WorkflowException, logger.isEnabledFor(logging.DEBUG)):
 137                 if "id" not in self.tool:
 138                     raise WorkflowException("%s object must have 'id'" % (self.tool["class"]))
 139             document_loader, workflowobj, uri = (self.doc_loader, self.doc_loader.fetch(self.tool["id"]), self.tool["id"])
 140
 141             discover_secondary_files(self.tool["inputs"], joborder)
 142
 143             with Perf(metrics, "subworkflow upload_deps"):
 144                 upload_dependencies(self.arvrunner,
 145                                     os.path.basename(joborder.get("id", "#")),
 146                                     document_loader,
 147                                     joborder,
 148                                     joborder.get("id", "#"),
 149                                     False)
 150
 151                 if self.wf_pdh is None:
 152                     workflowobj["requirements"] = dedup_reqs(self.requirements)
 153                     workflowobj["hints"] = dedup_reqs(self.hints)
 154
 155                     packed = pack(document_loader, workflowobj, uri, self.metadata)
 156
 157                     builder = Builder(joborder,
 158                                       requirements=workflowobj["requirements"],
 159                                       hints=workflowobj["hints"],
 160                                       resources={})
 161
 162                     def visit(item):
 163                         for t in ("hints", "requirements"):
 164                             if t not in item:
 165                                 continue
 166                             for req in item[t]:
 167                                 if req["class"] == "ResourceRequirement":
 168                                     dyn = False
 169                                     for k in max_res_pars + sum_res_pars:
 170                                         if k in req:
 171                                             if isinstance(req[k], basestring):
 172                                                 if item["id"] == "#main":
 173                                                     # only the top-level requirements/hints may contain expressions
 174                                                     self.dynamic_resource_req.append(req)
 175                                                     dyn = True
 176                                                     break
 177                                                 else:
 178                                                     with SourceLine(req, k, WorkflowException):
 179                                                         raise WorkflowException("Non-top-level ResourceRequirement in single container cannot have expressions")
 180                                     if not dyn:
 181                                         self.static_resource_req.append(req)
 182
 183                     visit_class(packed["$graph"], ("Workflow", "CommandLineTool"), visit)
 184
 185                     if self.static_resource_req:
 186                         self.static_resource_req = [get_overall_res_req(self.static_resource_req)]
 187
 188                     upload_dependencies(self.arvrunner,
 189                                         runtimeContext.name,
 190                                         document_loader,
 191                                         packed,
 192                                         uri,
 193                                         False)
 194
 195                     # Discover files/directories referenced by the
 196                     # workflow (mainly "default" values)
 197                     visit_class(packed, ("File", "Directory"), self.wf_reffiles.append)
 198
 199
 200             if self.dynamic_resource_req:
 201                 builder = Builder(joborder,
 202                                   requirements=self.requirements,
 203                                   hints=self.hints,
 204                                   resources={})
 205
 206                 # Evaluate dynamic resource requirements using current builder
 207                 rs = copy.copy(self.static_resource_req)
 208                 for dyn_rs in self.dynamic_resource_req:
 209                     eval_req = {"class": "ResourceRequirement"}
 210                     for a in max_res_pars + sum_res_pars:
 211                         if a in dyn_rs:
 212                             eval_req[a] = builder.do_eval(dyn_rs[a])
 213                     rs.append(eval_req)
 214                 job_res_reqs = [get_overall_res_req(rs)]
 215             else:
 216                 job_res_reqs = self.static_resource_req
 217
 218             with Perf(metrics, "subworkflow adjust"):
 219                 joborder_resolved = copy.deepcopy(joborder)
 220                 joborder_keepmount = copy.deepcopy(joborder)
 221
 222                 reffiles = []
 223                 visit_class(joborder_keepmount, ("File", "Directory"), reffiles.append)
 224
 225                 mapper = ArvPathMapper(self.arvrunner, reffiles+self.wf_reffiles, runtimeContext.basedir,
 226                                        "/keep/%s",
 227                                        "/keep/%s/%s")
 228
 229                 # For containers API, we need to make sure any extra
 230                 # referenced files (ie referenced by the workflow but
 231                 # not in the inputs) are included in the mounts.
 232                 if self.wf_reffiles:
 233                     runtimeContext = runtimeContext.copy()
 234                     runtimeContext.extra_reffiles = copy.deepcopy(self.wf_reffiles)
 235
 236                 def keepmount(obj):
 237                     remove_redundant_fields(obj)
 238                     with SourceLine(obj, None, WorkflowException, logger.isEnabledFor(logging.DEBUG)):
 239                         if "location" not in obj:
 240                             raise WorkflowException("%s object is missing required 'location' field: %s" % (obj["class"], obj))
 241                     with SourceLine(obj, "location", WorkflowException, logger.isEnabledFor(logging.DEBUG)):
 242                         if obj["location"].startswith("keep:"):
 243                             obj["location"] = mapper.mapper(obj["location"]).target
 244                             if "listing" in obj:
 245                                 del obj["listing"]
 246                         elif obj["location"].startswith("_:"):
 247                             del obj["location"]
 248                         else:
 249                             raise WorkflowException("Location is not a keep reference or a literal: '%s'" % obj["location"])
 250
 251                 visit_class(joborder_keepmount, ("File", "Directory"), keepmount)
 252
 253                 def resolved(obj):
 254                     if obj["location"].startswith("keep:"):
 255                         obj["location"] = mapper.mapper(obj["location"]).resolved
 256
 257                 visit_class(joborder_resolved, ("File", "Directory"), resolved)
 258
 259                 if self.wf_pdh is None:
 260                     adjustFileObjs(packed, keepmount)
 261                     adjustDirObjs(packed, keepmount)
 262                     self.wf_pdh = upload_workflow_collection(self.arvrunner, shortname(self.tool["id"]), packed)
 263
 264             wf_runner = cmap({
 265                 "class": "CommandLineTool",
 266                 "baseCommand": "cwltool",
 267                 "inputs": self.tool["inputs"],
 268                 "outputs": self.tool["outputs"],
 269                 "stdout": "cwl.output.json",
 270                 "requirements": self.requirements+job_res_reqs+[
 271                     {"class": "InlineJavascriptRequirement"},
 272                     {
 273                     "class": "InitialWorkDirRequirement",
 274                     "listing": [{
 275                             "entryname": "workflow.cwl",
 276                             "entry": '$({"class": "File", "location": "keep:%s/workflow.cwl"})' % self.wf_pdh
 277                         }, {
 278                             "entryname": "cwl.input.yml",
 279                             "entry": json.dumps(joborder_keepmount, indent=2, sort_keys=True, separators=(',',': ')).replace("\\", "\\\\").replace('$(', '\$(').replace('${', '\${')
 280                         }]
 281                 }],
 282                 "hints": self.hints,
 283                 "arguments": ["--no-container", "--move-outputs", "--preserve-entire-environment", "workflow.cwl#main", "cwl.input.yml"],
 284                 "id": "#"
 285             })
 286             return ArvadosCommandTool(self.arvrunner, wf_runner, self.loadingContext).job(joborder_resolved, output_callback, runtimeContext)
 287         else:
 288             return super(ArvadosWorkflow, self).job(joborder, output_callback, runtimeContext)