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
 120     def job(self, joborder, output_callback, **kwargs):
 121         kwargs["work_api"] = self.work_api
 122         req, _ = self.get_requirement("http://arvados.org/cwl#RunInSingleContainer")
 123         if req:
 124             with SourceLine(self.tool, None, WorkflowException, logger.isEnabledFor(logging.DEBUG)):
 125                 if "id" not in self.tool:
 126                     raise WorkflowException("%s object must have 'id'" % (self.tool["class"]))
 127             document_loader, workflowobj, uri = (self.doc_loader, self.doc_loader.fetch(self.tool["id"]), self.tool["id"])
 128
 129             discover_secondary_files(self.tool["inputs"], joborder)
 130
 131             with Perf(metrics, "subworkflow upload_deps"):
 132                 upload_dependencies(self.arvrunner,
 133                                     os.path.basename(joborder.get("id", "#")),
 134                                     document_loader,
 135                                     joborder,
 136                                     joborder.get("id", "#"),
 137                                     False)
 138
 139                 if self.wf_pdh is None:
 140                     workflowobj["requirements"] = dedup_reqs(self.requirements)
 141                     workflowobj["hints"] = dedup_reqs(self.hints)
 142
 143                     packed = pack(document_loader, workflowobj, uri, self.metadata)
 144
 145                     builder = Builder()
 146                     builder.job = joborder
 147                     builder.requirements = workflowobj["requirements"]
 148                     builder.hints = workflowobj["hints"]
 149                     builder.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                                         kwargs.get("name", ""),
 179                                         document_loader,
 180                                         packed,
 181                                         uri,
 182                                         False)
 183
 184             if self.dynamic_resource_req:
 185                 builder = Builder()
 186                 builder.job = joborder
 187                 builder.requirements = self.requirements
 188                 builder.hints = self.hints
 189                 builder.resources = {}
 190
 191                 # Evaluate dynamic resource requirements using current builder
 192                 rs = copy.copy(self.static_resource_req)
 193                 for dyn_rs in self.dynamic_resource_req:
 194                     eval_req = {"class": "ResourceRequirement"}
 195                     for a in max_res_pars + sum_res_pars:
 196                         if a in dyn_rs:
 197                             eval_req[a] = builder.do_eval(dyn_rs[a])
 198                     rs.append(eval_req)
 199                 job_res_reqs = [get_overall_res_req(rs)]
 200             else:
 201                 job_res_reqs = self.static_resource_req
 202
 203             with Perf(metrics, "subworkflow adjust"):
 204                 joborder_resolved = copy.deepcopy(joborder)
 205                 joborder_keepmount = copy.deepcopy(joborder)
 206
 207                 reffiles = []
 208                 visit_class(joborder_keepmount, ("File", "Directory"), lambda x: reffiles.append(x))
 209
 210                 mapper = ArvPathMapper(self.arvrunner, reffiles, kwargs["basedir"],
 211                                  "/keep/%s",
 212                                  "/keep/%s/%s",
 213                                  **kwargs)
 214
 215                 def keepmount(obj):
 216                     remove_redundant_fields(obj)
 217                     with SourceLine(obj, None, WorkflowException, logger.isEnabledFor(logging.DEBUG)):
 218                         if "location" not in obj:
 219                             raise WorkflowException("%s object is missing required 'location' field: %s" % (obj["class"], obj))
 220                     with SourceLine(obj, "location", WorkflowException, logger.isEnabledFor(logging.DEBUG)):
 221                         if obj["location"].startswith("keep:"):
 222                             obj["location"] = mapper.mapper(obj["location"]).target
 223                             if "listing" in obj:
 224                                 del obj["listing"]
 225                         elif obj["location"].startswith("_:"):
 226                             del obj["location"]
 227                         else:
 228                             raise WorkflowException("Location is not a keep reference or a literal: '%s'" % obj["location"])
 229
 230                 visit_class(joborder_keepmount, ("File", "Directory"), keepmount)
 231
 232                 def resolved(obj):
 233                     if obj["location"].startswith("keep:"):
 234                         obj["location"] = mapper.mapper(obj["location"]).resolved
 235
 236                 visit_class(joborder_resolved, ("File", "Directory"), resolved)
 237
 238                 if self.wf_pdh is None:
 239                     adjustFileObjs(packed, keepmount)
 240                     adjustDirObjs(packed, keepmount)
 241                     self.wf_pdh = upload_workflow_collection(self.arvrunner, shortname(self.tool["id"]), packed)
 242
 243             wf_runner = cmap({
 244                 "class": "CommandLineTool",
 245                 "baseCommand": "cwltool",
 246                 "inputs": self.tool["inputs"],
 247                 "outputs": self.tool["outputs"],
 248                 "stdout": "cwl.output.json",
 249                 "requirements": self.requirements+job_res_reqs+[
 250                     {
 251                     "class": "InitialWorkDirRequirement",
 252                     "listing": [{
 253                             "entryname": "workflow.cwl",
 254                             "entry": {
 255                                 "class": "File",
 256                                 "location": "keep:%s/workflow.cwl" % self.wf_pdh
 257                             }
 258                         }, {
 259                             "entryname": "cwl.input.yml",
 260                             "entry": json.dumps(joborder_keepmount, indent=2, sort_keys=True, separators=(',',': ')).replace("\\", "\\\\").replace('$(', '\$(').replace('${', '\${')
 261                         }]
 262                 }],
 263                 "hints": self.hints,
 264                 "arguments": ["--no-container", "--move-outputs", "--preserve-entire-environment", "workflow.cwl#main", "cwl.input.yml"],
 265                 "id": "#"
 266             })
 267             kwargs["loader"] = self.doc_loader
 268             kwargs["avsc_names"] = self.doc_schema
 269             return ArvadosCommandTool(self.arvrunner, wf_runner, **kwargs).job(joborder_resolved, output_callback, **kwargs)
 270         else:
 271             return super(ArvadosWorkflow, self).job(joborder, output_callback, **kwargs)