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.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, set_cluster_target
  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 ArvadosWorkflowStep(WorkflowStep):
 122     def __init__(self,
 123                  toolpath_object,      # type: Dict[Text, Any]
 124                  pos,                  # type: int
 125                  loadingContext,       # type: LoadingContext
 126                  arvrunner,
 127                  *argc,
 128                  **argv
 129                 ):  # type: (...) -> None
 130
 131         super(ArvadosWorkflowStep, self).__init__(toolpath_object, pos, loadingContext, *argc, **argv)
 132         self.tool["class"] = "WorkflowStep"
 133         self.arvrunner = arvrunner
 134
 135     def job(self, joborder, output_callback, runtimeContext):
 136         builder = self._init_job({shortname(k): v for k,v in joborder.items()}, runtimeContext)
 137         runtimeContext = set_cluster_target(self.tool, self.arvrunner, builder, runtimeContext)
 138         return super(ArvadosWorkflowStep, self).job(joborder, output_callback, runtimeContext)
 139
 140
 141 class ArvadosWorkflow(Workflow):
 142     """Wrap cwltool Workflow to override selected methods."""
 143
 144     def __init__(self, arvrunner, toolpath_object, loadingContext):
 145         self.arvrunner = arvrunner
 146         self.wf_pdh = None
 147         self.dynamic_resource_req = []
 148         self.static_resource_req = []
 149         self.wf_reffiles = []
 150         self.loadingContext = loadingContext
 151         super(ArvadosWorkflow, self).__init__(toolpath_object, loadingContext)
 152         self.cluster_target_req, _ = self.get_requirement("http://arvados.org/cwl#ClusterTarget")
 153
 154     def job(self, joborder, output_callback, runtimeContext):
 155
 156         builder = self._init_job(joborder, runtimeContext)
 157         runtimeContext = set_cluster_target(self.tool, self.arvrunner, builder, runtimeContext)
 158
 159         req, _ = self.get_requirement("http://arvados.org/cwl#RunInSingleContainer")
 160         if not req:
 161             return super(ArvadosWorkflow, self).job(joborder, output_callback, runtimeContext)
 162
 163         # RunInSingleContainer is true
 164
 165         with SourceLine(self.tool, None, WorkflowException, logger.isEnabledFor(logging.DEBUG)):
 166             if "id" not in self.tool:
 167                 raise WorkflowException("%s object must have 'id'" % (self.tool["class"]))
 168         document_loader, workflowobj, uri = (self.doc_loader, self.doc_loader.fetch(self.tool["id"]), self.tool["id"])
 169
 170         discover_secondary_files(self.tool["inputs"], joborder)
 171
 172         with Perf(metrics, "subworkflow upload_deps"):
 173             upload_dependencies(self.arvrunner,
 174                                 os.path.basename(joborder.get("id", "#")),
 175                                 document_loader,
 176                                 joborder,
 177                                 joborder.get("id", "#"),
 178                                 False)
 179
 180             if self.wf_pdh is None:
 181                 workflowobj["requirements"] = dedup_reqs(self.requirements)
 182                 workflowobj["hints"] = dedup_reqs(self.hints)
 183
 184                 packed = pack(document_loader, workflowobj, uri, self.metadata)
 185
 186                 def visit(item):
 187                     for t in ("hints", "requirements"):
 188                         if t not in item:
 189                             continue
 190                         for req in item[t]:
 191                             if req["class"] == "ResourceRequirement":
 192                                 dyn = False
 193                                 for k in max_res_pars + sum_res_pars:
 194                                     if k in req:
 195                                         if isinstance(req[k], basestring):
 196                                             if item["id"] == "#main":
 197                                                 # only the top-level requirements/hints may contain expressions
 198                                                 self.dynamic_resource_req.append(req)
 199                                                 dyn = True
 200                                                 break
 201                                             else:
 202                                                 with SourceLine(req, k, WorkflowException):
 203                                                     raise WorkflowException("Non-top-level ResourceRequirement in single container cannot have expressions")
 204                                 if not dyn:
 205                                     self.static_resource_req.append(req)
 206
 207                 visit_class(packed["$graph"], ("Workflow", "CommandLineTool"), visit)
 208
 209                 if self.static_resource_req:
 210                     self.static_resource_req = [get_overall_res_req(self.static_resource_req)]
 211
 212                 upload_dependencies(self.arvrunner,
 213                                     runtimeContext.name,
 214                                     document_loader,
 215                                     packed,
 216                                     uri,
 217                                     False)
 218
 219                 # Discover files/directories referenced by the
 220                 # workflow (mainly "default" values)
 221                 visit_class(packed, ("File", "Directory"), self.wf_reffiles.append)
 222
 223
 224         if self.dynamic_resource_req:
 225             # Evaluate dynamic resource requirements using current builder
 226             rs = copy.copy(self.static_resource_req)
 227             for dyn_rs in self.dynamic_resource_req:
 228                 eval_req = {"class": "ResourceRequirement"}
 229                 for a in max_res_pars + sum_res_pars:
 230                     if a in dyn_rs:
 231                         eval_req[a] = builder.do_eval(dyn_rs[a])
 232                 rs.append(eval_req)
 233             job_res_reqs = [get_overall_res_req(rs)]
 234         else:
 235             job_res_reqs = self.static_resource_req
 236
 237         with Perf(metrics, "subworkflow adjust"):
 238             joborder_resolved = copy.deepcopy(joborder)
 239             joborder_keepmount = copy.deepcopy(joborder)
 240
 241             reffiles = []
 242             visit_class(joborder_keepmount, ("File", "Directory"), reffiles.append)
 243
 244             mapper = ArvPathMapper(self.arvrunner, reffiles+self.wf_reffiles, runtimeContext.basedir,
 245                                    "/keep/%s",
 246                                    "/keep/%s/%s")
 247
 248             # For containers API, we need to make sure any extra
 249             # referenced files (ie referenced by the workflow but
 250             # not in the inputs) are included in the mounts.
 251             if self.wf_reffiles:
 252                 runtimeContext = runtimeContext.copy()
 253                 runtimeContext.extra_reffiles = copy.deepcopy(self.wf_reffiles)
 254
 255             def keepmount(obj):
 256                 remove_redundant_fields(obj)
 257                 with SourceLine(obj, None, WorkflowException, logger.isEnabledFor(logging.DEBUG)):
 258                     if "location" not in obj:
 259                         raise WorkflowException("%s object is missing required 'location' field: %s" % (obj["class"], obj))
 260                 with SourceLine(obj, "location", WorkflowException, logger.isEnabledFor(logging.DEBUG)):
 261                     if obj["location"].startswith("keep:"):
 262                         obj["location"] = mapper.mapper(obj["location"]).target
 263                         if "listing" in obj:
 264                             del obj["listing"]
 265                     elif obj["location"].startswith("_:"):
 266                         del obj["location"]
 267                     else:
 268                         raise WorkflowException("Location is not a keep reference or a literal: '%s'" % obj["location"])
 269
 270             visit_class(joborder_keepmount, ("File", "Directory"), keepmount)
 271
 272             def resolved(obj):
 273                 if obj["location"].startswith("keep:"):
 274                     obj["location"] = mapper.mapper(obj["location"]).resolved
 275
 276             visit_class(joborder_resolved, ("File", "Directory"), resolved)
 277
 278             if self.wf_pdh is None:
 279                 adjustFileObjs(packed, keepmount)
 280                 adjustDirObjs(packed, keepmount)
 281                 self.wf_pdh = upload_workflow_collection(self.arvrunner, shortname(self.tool["id"]), packed)
 282
 283         wf_runner = cmap({
 284             "class": "CommandLineTool",
 285             "baseCommand": "cwltool",
 286             "inputs": self.tool["inputs"],
 287             "outputs": self.tool["outputs"],
 288             "stdout": "cwl.output.json",
 289             "requirements": self.requirements+job_res_reqs+[
 290                 {"class": "InlineJavascriptRequirement"},
 291                 {
 292                 "class": "InitialWorkDirRequirement",
 293                 "listing": [{
 294                         "entryname": "workflow.cwl",
 295                         "entry": '$({"class": "File", "location": "keep:%s/workflow.cwl"})' % self.wf_pdh
 296                     }, {
 297                         "entryname": "cwl.input.yml",
 298                         "entry": json.dumps(joborder_keepmount, indent=2, sort_keys=True, separators=(',',': ')).replace("\\", "\\\\").replace('$(', '\$(').replace('${', '\${')
 299                     }]
 300             }],
 301             "hints": self.hints,
 302             "arguments": ["--no-container", "--move-outputs", "--preserve-entire-environment", "workflow.cwl#main", "cwl.input.yml"],
 303             "id": "#"
 304         })
 305         return ArvadosCommandTool(self.arvrunner, wf_runner, self.loadingContext).job(joborder_resolved, output_callback, runtimeContext)
 306
 307     def make_workflow_step(self,
 308                            toolpath_object,      # type: Dict[Text, Any]
 309                            pos,                  # type: int
 310                            loadingContext,       # type: LoadingContext
 311                            *argc,
 312                            **argv
 313     ):
 314         # (...) -> WorkflowStep
 315         return ArvadosWorkflowStep(toolpath_object, pos, loadingContext, self.arvrunner, *argc, **argv)