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