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 from past.builtins import basestring
   6 from future.utils import viewitems
   7
   8 import os
   9 import json
  10 import copy
  11 import logging
  12
  13 from schema_salad.sourceline import SourceLine, cmap
  14
  15 from cwltool.pack import pack
  16 from cwltool.load_tool import fetch_document
  17 from cwltool.process import shortname
  18 from cwltool.workflow import Workflow, WorkflowException, WorkflowStep
  19 from cwltool.pathmapper import adjustFileObjs, adjustDirObjs, visit_class
  20 from cwltool.context import LoadingContext
  21
  22 import ruamel.yaml as yaml
  23
  24 from .runner import (upload_dependencies, packed_workflow, upload_workflow_collection,
  25                      trim_anonymous_location, remove_redundant_fields, discover_secondary_files)
  26 from .pathmapper import ArvPathMapper, trim_listing
  27 from .arvtool import ArvadosCommandTool, set_cluster_target, make_builder
  28 from .perf import Perf
  29
  30 logger = logging.getLogger('arvados.cwl-runner')
  31 metrics = logging.getLogger('arvados.cwl-runner.metrics')
  32
  33 max_res_pars = ("coresMin", "coresMax", "ramMin", "ramMax", "tmpdirMin", "tmpdirMax")
  34 sum_res_pars = ("outdirMin", "outdirMax")
  35
  36 def upload_workflow(arvRunner, tool, job_order, project_uuid, uuid=None,
  37                     submit_runner_ram=0, name=None, merged_map=None):
  38
  39     packed = packed_workflow(arvRunner, tool, merged_map)
  40
  41     adjustDirObjs(job_order, trim_listing)
  42     adjustFileObjs(job_order, trim_anonymous_location)
  43     adjustDirObjs(job_order, trim_anonymous_location)
  44
  45     main = [p for p in packed["$graph"] if p["id"] == "#main"][0]
  46     for inp in main["inputs"]:
  47         sn = shortname(inp["id"])
  48         if sn in job_order:
  49             inp["default"] = job_order[sn]
  50
  51     if not name:
  52         name = tool.tool.get("label", os.path.basename(tool.tool["id"]))
  53
  54     upload_dependencies(arvRunner, name, tool.doc_loader,
  55                         packed, tool.tool["id"], False)
  56
  57     if submit_runner_ram:
  58         hints = main.get("hints", [])
  59         found = False
  60         for h in hints:
  61             if h["class"] == "http://arvados.org/cwl#WorkflowRunnerResources":
  62                 h["ramMin"] = submit_runner_ram
  63                 found = True
  64                 break
  65         if not found:
  66             hints.append({"class": "http://arvados.org/cwl#WorkflowRunnerResources",
  67                           "ramMin": submit_runner_ram})
  68         main["hints"] = hints
  69
  70     body = {
  71         "workflow": {
  72             "name": name,
  73             "description": tool.tool.get("doc", ""),
  74             "definition":json.dumps(packed, sort_keys=True, indent=4, separators=(',',': '))
  75         }}
  76     if project_uuid:
  77         body["workflow"]["owner_uuid"] = project_uuid
  78
  79     if uuid:
  80         call = arvRunner.api.workflows().update(uuid=uuid, body=body)
  81     else:
  82         call = arvRunner.api.workflows().create(body=body)
  83     return call.execute(num_retries=arvRunner.num_retries)["uuid"]
  84
  85 def dedup_reqs(reqs):
  86     dedup = {}
  87     for r in reversed(reqs):
  88         if r["class"] not in dedup and not r["class"].startswith("http://arvados.org/cwl#"):
  89             dedup[r["class"]] = r
  90     return [dedup[r] for r in sorted(dedup.keys())]
  91
  92 def get_overall_res_req(res_reqs):
  93     """Take the overall of a list of ResourceRequirement,
  94     i.e., the max of coresMin, coresMax, ramMin, ramMax, tmpdirMin, tmpdirMax
  95     and the sum of outdirMin, outdirMax."""
  96
  97     all_res_req = {}
  98     exception_msgs = []
  99     for a in max_res_pars + sum_res_pars:
 100         all_res_req[a] = []
 101         for res_req in res_reqs:
 102             if a in res_req:
 103                 if isinstance(res_req[a], int): # integer check
 104                     all_res_req[a].append(res_req[a])
 105                 else:
 106                     msg = SourceLine(res_req, a).makeError(
 107                     "Non-top-level ResourceRequirement in single container cannot have expressions")
 108                     exception_msgs.append(msg)
 109     if exception_msgs:
 110         raise WorkflowException("\n".join(exception_msgs))
 111     else:
 112         overall_res_req = {}
 113         for a in all_res_req:
 114             if all_res_req[a]:
 115                 if a in max_res_pars:
 116                     overall_res_req[a] = max(all_res_req[a])
 117                 elif a in sum_res_pars:
 118                     overall_res_req[a] = sum(all_res_req[a])
 119         if overall_res_req:
 120             overall_res_req["class"] = "ResourceRequirement"
 121         return cmap(overall_res_req)
 122
 123 class ArvadosWorkflowStep(WorkflowStep):
 124     def __init__(self,
 125                  toolpath_object,      # type: Dict[Text, Any]
 126                  pos,                  # type: int
 127                  loadingContext,       # type: LoadingContext
 128                  arvrunner,
 129                  *argc,
 130                  **argv
 131                 ):  # type: (...) -> None
 132
 133         super(ArvadosWorkflowStep, self).__init__(toolpath_object, pos, loadingContext, *argc, **argv)
 134         self.tool["class"] = "WorkflowStep"
 135         self.arvrunner = arvrunner
 136
 137     def job(self, joborder, output_callback, runtimeContext):
 138         runtimeContext = runtimeContext.copy()
 139         runtimeContext.toplevel = True  # Preserve behavior for #13365
 140
 141         builder = make_builder({shortname(k): v for k,v in viewitems(joborder)}, self.hints, self.requirements, runtimeContext)
 142         runtimeContext = set_cluster_target(self.tool, self.arvrunner, builder, runtimeContext)
 143         return super(ArvadosWorkflowStep, self).job(joborder, output_callback, runtimeContext)
 144
 145
 146 class ArvadosWorkflow(Workflow):
 147     """Wrap cwltool Workflow to override selected methods."""
 148
 149     def __init__(self, arvrunner, toolpath_object, loadingContext):
 150         self.arvrunner = arvrunner
 151         self.wf_pdh = None
 152         self.dynamic_resource_req = []
 153         self.static_resource_req = []
 154         self.wf_reffiles = []
 155         self.loadingContext = loadingContext
 156         super(ArvadosWorkflow, self).__init__(toolpath_object, loadingContext)
 157         self.cluster_target_req, _ = self.get_requirement("http://arvados.org/cwl#ClusterTarget")
 158
 159     def job(self, joborder, output_callback, runtimeContext):
 160
 161         builder = make_builder(joborder, self.hints, self.requirements, runtimeContext)
 162         runtimeContext = set_cluster_target(self.tool, self.arvrunner, builder, runtimeContext)
 163
 164         req, _ = self.get_requirement("http://arvados.org/cwl#RunInSingleContainer")
 165         if not req:
 166             return super(ArvadosWorkflow, self).job(joborder, output_callback, runtimeContext)
 167
 168         # RunInSingleContainer is true
 169
 170         with SourceLine(self.tool, None, WorkflowException, logger.isEnabledFor(logging.DEBUG)):
 171             if "id" not in self.tool:
 172                 raise WorkflowException("%s object must have 'id'" % (self.tool["class"]))
 173         document_loader, workflowobj, uri = (self.doc_loader, self.doc_loader.fetch(self.tool["id"]), self.tool["id"])
 174
 175         discover_secondary_files(self.tool["inputs"], joborder)
 176
 177         with Perf(metrics, "subworkflow upload_deps"):
 178             upload_dependencies(self.arvrunner,
 179                                 os.path.basename(joborder.get("id", "#")),
 180                                 document_loader,
 181                                 joborder,
 182                                 joborder.get("id", "#"),
 183                                 False)
 184
 185             if self.wf_pdh is None:
 186                 workflowobj["requirements"] = dedup_reqs(self.requirements)
 187                 workflowobj["hints"] = dedup_reqs(self.hints)
 188
 189                 packed = pack(document_loader, workflowobj, uri, self.metadata)
 190
 191                 def visit(item):
 192                     for t in ("hints", "requirements"):
 193                         if t not in item:
 194                             continue
 195                         for req in item[t]:
 196                             if req["class"] == "ResourceRequirement":
 197                                 dyn = False
 198                                 for k in max_res_pars + sum_res_pars:
 199                                     if k in req:
 200                                         if isinstance(req[k], basestring):
 201                                             if item["id"] == "#main":
 202                                                 # only the top-level requirements/hints may contain expressions
 203                                                 self.dynamic_resource_req.append(req)
 204                                                 dyn = True
 205                                                 break
 206                                             else:
 207                                                 with SourceLine(req, k, WorkflowException):
 208                                                     raise WorkflowException("Non-top-level ResourceRequirement in single container cannot have expressions")
 209                                 if not dyn:
 210                                     self.static_resource_req.append(req)
 211                             if req["class"] == "DockerRequirement":
 212                                 if "http://arvados.org/cwl#dockerCollectionPDH" in req:
 213                                     del req["http://arvados.org/cwl#dockerCollectionPDH"]
 214
 215                 visit_class(packed["$graph"], ("Workflow", "CommandLineTool"), visit)
 216
 217                 if self.static_resource_req:
 218                     self.static_resource_req = [get_overall_res_req(self.static_resource_req)]
 219
 220                 upload_dependencies(self.arvrunner,
 221                                     runtimeContext.name,
 222                                     document_loader,
 223                                     packed,
 224                                     uri,
 225                                     False)
 226
 227                 # Discover files/directories referenced by the
 228                 # workflow (mainly "default" values)
 229                 visit_class(packed, ("File", "Directory"), self.wf_reffiles.append)
 230
 231
 232         if self.dynamic_resource_req:
 233             # Evaluate dynamic resource requirements using current builder
 234             rs = copy.copy(self.static_resource_req)
 235             for dyn_rs in self.dynamic_resource_req:
 236                 eval_req = {"class": "ResourceRequirement"}
 237                 for a in max_res_pars + sum_res_pars:
 238                     if a in dyn_rs:
 239                         eval_req[a] = builder.do_eval(dyn_rs[a])
 240                 rs.append(eval_req)
 241             job_res_reqs = [get_overall_res_req(rs)]
 242         else:
 243             job_res_reqs = self.static_resource_req
 244
 245         with Perf(metrics, "subworkflow adjust"):
 246             joborder_resolved = copy.deepcopy(joborder)
 247             joborder_keepmount = copy.deepcopy(joborder)
 248
 249             reffiles = []
 250             visit_class(joborder_keepmount, ("File", "Directory"), reffiles.append)
 251
 252             mapper = ArvPathMapper(self.arvrunner, reffiles+self.wf_reffiles, runtimeContext.basedir,
 253                                    "/keep/%s",
 254                                    "/keep/%s/%s")
 255
 256             # For containers API, we need to make sure any extra
 257             # referenced files (ie referenced by the workflow but
 258             # not in the inputs) are included in the mounts.
 259             if self.wf_reffiles:
 260                 runtimeContext = runtimeContext.copy()
 261                 runtimeContext.extra_reffiles = copy.deepcopy(self.wf_reffiles)
 262
 263             def keepmount(obj):
 264                 remove_redundant_fields(obj)
 265                 with SourceLine(obj, None, WorkflowException, logger.isEnabledFor(logging.DEBUG)):
 266                     if "location" not in obj:
 267                         raise WorkflowException("%s object is missing required 'location' field: %s" % (obj["class"], obj))
 268                 with SourceLine(obj, "location", WorkflowException, logger.isEnabledFor(logging.DEBUG)):
 269                     if obj["location"].startswith("keep:"):
 270                         obj["location"] = mapper.mapper(obj["location"]).target
 271                         if "listing" in obj:
 272                             del obj["listing"]
 273                     elif obj["location"].startswith("_:"):
 274                         del obj["location"]
 275                     else:
 276                         raise WorkflowException("Location is not a keep reference or a literal: '%s'" % obj["location"])
 277
 278             visit_class(joborder_keepmount, ("File", "Directory"), keepmount)
 279
 280             def resolved(obj):
 281                 if obj["location"].startswith("keep:"):
 282                     obj["location"] = mapper.mapper(obj["location"]).resolved
 283
 284             visit_class(joborder_resolved, ("File", "Directory"), resolved)
 285
 286             if self.wf_pdh is None:
 287                 adjustFileObjs(packed, keepmount)
 288                 adjustDirObjs(packed, keepmount)
 289                 self.wf_pdh = upload_workflow_collection(self.arvrunner, shortname(self.tool["id"]), packed)
 290
 291         wf_runner = cmap({
 292             "class": "CommandLineTool",
 293             "baseCommand": "cwltool",
 294             "inputs": self.tool["inputs"],
 295             "outputs": self.tool["outputs"],
 296             "stdout": "cwl.output.json",
 297             "requirements": self.requirements+job_res_reqs+[
 298                 {"class": "InlineJavascriptRequirement"},
 299                 {
 300                 "class": "InitialWorkDirRequirement",
 301                 "listing": [{
 302                         "entryname": "workflow.cwl",
 303                         "entry": '$({"class": "File", "location": "keep:%s/workflow.cwl"})' % self.wf_pdh
 304                     }, {
 305                         "entryname": "cwl.input.yml",
 306                         "entry": json.dumps(joborder_keepmount, indent=2, sort_keys=True, separators=(',',': ')).replace("\\", "\\\\").replace('$(', '\$(').replace('${', '\${')
 307                     }]
 308             }],
 309             "hints": self.hints,
 310             "arguments": ["--no-container", "--move-outputs", "--preserve-entire-environment", "workflow.cwl#main", "cwl.input.yml"],
 311             "id": "#"
 312         })
 313         return ArvadosCommandTool(self.arvrunner, wf_runner, self.loadingContext).job(joborder_resolved, output_callback, runtimeContext)
 314
 315     def make_workflow_step(self,
 316                            toolpath_object,      # type: Dict[Text, Any]
 317                            pos,                  # type: int
 318                            loadingContext,       # type: LoadingContext
 319                            *argc,
 320                            **argv
 321     ):
 322         # (...) -> WorkflowStep
 323         return ArvadosWorkflowStep(toolpath_object, pos, loadingContext, self.arvrunner, *argc, **argv)