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 import schema_salad.ref_resolver
  15
  16 from cwltool.pack import pack
  17 from cwltool.load_tool import fetch_document, resolve_and_validate_document
  18 from cwltool.process import shortname
  19 from cwltool.workflow import Workflow, WorkflowException, WorkflowStep
  20 from cwltool.pathmapper import adjustFileObjs, adjustDirObjs, visit_class
  21 from cwltool.context import LoadingContext
  22
  23 import ruamel.yaml as yaml
  24
  25 from .runner import (upload_dependencies, packed_workflow, upload_workflow_collection,
  26                      trim_anonymous_location, remove_redundant_fields, discover_secondary_files,
  27                      make_builder)
  28 from .pathmapper import ArvPathMapper, trim_listing
  29 from .arvtool import ArvadosCommandTool, set_cluster_target
  30
  31 from .perf import Perf
  32
  33 logger = logging.getLogger('arvados.cwl-runner')
  34 metrics = logging.getLogger('arvados.cwl-runner.metrics')
  35
  36 max_res_pars = ("coresMin", "coresMax", "ramMin", "ramMax", "tmpdirMin", "tmpdirMax")
  37 sum_res_pars = ("outdirMin", "outdirMax")
  38
  39 def upload_workflow(arvRunner, tool, job_order, project_uuid, uuid=None,
  40                     submit_runner_ram=0, name=None, merged_map=None):
  41
  42     packed = packed_workflow(arvRunner, tool, merged_map)
  43
  44     adjustDirObjs(job_order, trim_listing)
  45     adjustFileObjs(job_order, trim_anonymous_location)
  46     adjustDirObjs(job_order, trim_anonymous_location)
  47
  48     main = [p for p in packed["$graph"] if p["id"] == "#main"][0]
  49     for inp in main["inputs"]:
  50         sn = shortname(inp["id"])
  51         if sn in job_order:
  52             inp["default"] = job_order[sn]
  53
  54     if not name:
  55         name = tool.tool.get("label", os.path.basename(tool.tool["id"]))
  56
  57     upload_dependencies(arvRunner, name, tool.doc_loader,
  58                         packed, tool.tool["id"], False)
  59
  60     if submit_runner_ram:
  61         hints = main.get("hints", [])
  62         found = False
  63         for h in hints:
  64             if h["class"] == "http://arvados.org/cwl#WorkflowRunnerResources":
  65                 h["ramMin"] = submit_runner_ram
  66                 found = True
  67                 break
  68         if not found:
  69             hints.append({"class": "http://arvados.org/cwl#WorkflowRunnerResources",
  70                           "ramMin": submit_runner_ram})
  71         main["hints"] = hints
  72
  73     body = {
  74         "workflow": {
  75             "name": name,
  76             "description": tool.tool.get("doc", ""),
  77             "definition":json.dumps(packed, sort_keys=True, indent=4, separators=(',',': '))
  78         }}
  79     if project_uuid:
  80         body["workflow"]["owner_uuid"] = project_uuid
  81
  82     if uuid:
  83         call = arvRunner.api.workflows().update(uuid=uuid, body=body)
  84     else:
  85         call = arvRunner.api.workflows().create(body=body)
  86     return call.execute(num_retries=arvRunner.num_retries)["uuid"]
  87
  88 def dedup_reqs(reqs):
  89     dedup = {}
  90     for r in reversed(reqs):
  91         if r["class"] not in dedup and not r["class"].startswith("http://arvados.org/cwl#"):
  92             dedup[r["class"]] = r
  93     return [dedup[r] for r in sorted(dedup.keys())]
  94
  95 def get_overall_res_req(res_reqs):
  96     """Take the overall of a list of ResourceRequirement,
  97     i.e., the max of coresMin, coresMax, ramMin, ramMax, tmpdirMin, tmpdirMax
  98     and the sum of outdirMin, outdirMax."""
  99
 100     all_res_req = {}
 101     exception_msgs = []
 102     for a in max_res_pars + sum_res_pars:
 103         all_res_req[a] = []
 104         for res_req in res_reqs:
 105             if a in res_req:
 106                 if isinstance(res_req[a], int): # integer check
 107                     all_res_req[a].append(res_req[a])
 108                 else:
 109                     msg = SourceLine(res_req, a).makeError(
 110                     "Non-top-level ResourceRequirement in single container cannot have expressions")
 111                     exception_msgs.append(msg)
 112     if exception_msgs:
 113         raise WorkflowException("\n".join(exception_msgs))
 114     else:
 115         overall_res_req = {}
 116         for a in all_res_req:
 117             if all_res_req[a]:
 118                 if a in max_res_pars:
 119                     overall_res_req[a] = max(all_res_req[a])
 120                 elif a in sum_res_pars:
 121                     overall_res_req[a] = sum(all_res_req[a])
 122         if overall_res_req:
 123             overall_res_req["class"] = "ResourceRequirement"
 124         return cmap(overall_res_req)
 125
 126 class ArvadosWorkflowStep(WorkflowStep):
 127     def __init__(self,
 128                  toolpath_object,      # type: Dict[Text, Any]
 129                  pos,                  # type: int
 130                  loadingContext,       # type: LoadingContext
 131                  arvrunner,
 132                  *argc,
 133                  **argv
 134                 ):  # type: (...) -> None
 135
 136         super(ArvadosWorkflowStep, self).__init__(toolpath_object, pos, loadingContext, *argc, **argv)
 137         self.tool["class"] = "WorkflowStep"
 138         self.arvrunner = arvrunner
 139
 140     def job(self, joborder, output_callback, runtimeContext):
 141         runtimeContext = runtimeContext.copy()
 142         runtimeContext.toplevel = True  # Preserve behavior for #13365
 143
 144         builder = make_builder({shortname(k): v for k,v in viewitems(joborder)}, self.hints, self.requirements, runtimeContext)
 145         runtimeContext = set_cluster_target(self.tool, self.arvrunner, builder, runtimeContext)
 146         return super(ArvadosWorkflowStep, self).job(joborder, output_callback, runtimeContext)
 147
 148
 149 class ArvadosWorkflow(Workflow):
 150     """Wrap cwltool Workflow to override selected methods."""
 151
 152     def __init__(self, arvrunner, toolpath_object, loadingContext):
 153         self.arvrunner = arvrunner
 154         self.wf_pdh = None
 155         self.dynamic_resource_req = []
 156         self.static_resource_req = []
 157         self.wf_reffiles = []
 158         self.loadingContext = loadingContext
 159         super(ArvadosWorkflow, self).__init__(toolpath_object, loadingContext)
 160         self.cluster_target_req, _ = self.get_requirement("http://arvados.org/cwl#ClusterTarget")
 161
 162     def job(self, joborder, output_callback, runtimeContext):
 163
 164         builder = make_builder(joborder, self.hints, self.requirements, runtimeContext)
 165         runtimeContext = set_cluster_target(self.tool, self.arvrunner, builder, runtimeContext)
 166
 167         req, _ = self.get_requirement("http://arvados.org/cwl#RunInSingleContainer")
 168         if not req:
 169             return super(ArvadosWorkflow, self).job(joborder, output_callback, runtimeContext)
 170
 171         # RunInSingleContainer is true
 172
 173         with SourceLine(self.tool, None, WorkflowException, logger.isEnabledFor(logging.DEBUG)):
 174             if "id" not in self.tool:
 175                 raise WorkflowException("%s object must have 'id'" % (self.tool["class"]))
 176
 177         discover_secondary_files(self.arvrunner.fs_access, builder,
 178                                  self.tool["inputs"], joborder)
 179
 180         with Perf(metrics, "subworkflow upload_deps"):
 181             upload_dependencies(self.arvrunner,
 182                                 os.path.basename(joborder.get("id", "#")),
 183                                 self.doc_loader,
 184                                 joborder,
 185                                 joborder.get("id", "#"),
 186                                 False)
 187
 188             if self.wf_pdh is None:
 189                 ### We have to reload the document to get the original version.
 190                 #
 191                 # The workflow document we have in memory right now was
 192                 # updated to the internal CWL version.  We need to reload the
 193                 # document to go back to its original version, because the
 194                 # version of cwltool installed in the user's container is
 195                 # likely to reject our internal version, and we don't want to
 196                 # break existing workflows.
 197                 #
 198                 # What's going on here is that the updater replaces the
 199                 # documents/fragments in the index with updated ones, the
 200                 # index is also used as a cache, so we need to go through the
 201                 # loading process with an empty index and updating turned off
 202                 # so we have the original un-updated documents.
 203                 #
 204                 loadingContext = self.loadingContext.copy()
 205                 loadingContext.do_update = False
 206                 document_loader = schema_salad.ref_resolver.SubLoader(loadingContext.loader)
 207                 loadingContext.loader = document_loader
 208                 loadingContext.loader.idx = {}
 209                 uri = self.tool["id"]
 210                 loadingContext, workflowobj, uri = fetch_document(uri, loadingContext)
 211                 loadingContext, uri = resolve_and_validate_document(
 212                     loadingContext, workflowobj, uri
 213                 )
 214                 workflowobj, metadata = loadingContext.loader.resolve_ref(uri)
 215                 ###
 216
 217                 workflowobj["requirements"] = dedup_reqs(self.requirements)
 218                 workflowobj["hints"] = dedup_reqs(self.hints)
 219
 220                 packed = pack(document_loader, workflowobj, uri, metadata)
 221
 222                 def visit(item):
 223                     for t in ("hints", "requirements"):
 224                         if t not in item:
 225                             continue
 226                         for req in item[t]:
 227                             if req["class"] == "ResourceRequirement":
 228                                 dyn = False
 229                                 for k in max_res_pars + sum_res_pars:
 230                                     if k in req:
 231                                         if isinstance(req[k], basestring):
 232                                             if item["id"] == "#main":
 233                                                 # only the top-level requirements/hints may contain expressions
 234                                                 self.dynamic_resource_req.append(req)
 235                                                 dyn = True
 236                                                 break
 237                                             else:
 238                                                 with SourceLine(req, k, WorkflowException):
 239                                                     raise WorkflowException("Non-top-level ResourceRequirement in single container cannot have expressions")
 240                                 if not dyn:
 241                                     self.static_resource_req.append(req)
 242                             if req["class"] == "DockerRequirement":
 243                                 if "http://arvados.org/cwl#dockerCollectionPDH" in req:
 244                                     del req["http://arvados.org/cwl#dockerCollectionPDH"]
 245
 246                 visit_class(packed["$graph"], ("Workflow", "CommandLineTool"), visit)
 247
 248                 if self.static_resource_req:
 249                     self.static_resource_req = [get_overall_res_req(self.static_resource_req)]
 250
 251                 upload_dependencies(self.arvrunner,
 252                                     runtimeContext.name,
 253                                     document_loader,
 254                                     packed,
 255                                     uri,
 256                                     False)
 257
 258                 # Discover files/directories referenced by the
 259                 # workflow (mainly "default" values)
 260                 visit_class(packed, ("File", "Directory"), self.wf_reffiles.append)
 261
 262
 263         if self.dynamic_resource_req:
 264             # Evaluate dynamic resource requirements using current builder
 265             rs = copy.copy(self.static_resource_req)
 266             for dyn_rs in self.dynamic_resource_req:
 267                 eval_req = {"class": "ResourceRequirement"}
 268                 for a in max_res_pars + sum_res_pars:
 269                     if a in dyn_rs:
 270                         eval_req[a] = builder.do_eval(dyn_rs[a])
 271                 rs.append(eval_req)
 272             job_res_reqs = [get_overall_res_req(rs)]
 273         else:
 274             job_res_reqs = self.static_resource_req
 275
 276         with Perf(metrics, "subworkflow adjust"):
 277             joborder_resolved = copy.deepcopy(joborder)
 278             joborder_keepmount = copy.deepcopy(joborder)
 279
 280             reffiles = []
 281             visit_class(joborder_keepmount, ("File", "Directory"), reffiles.append)
 282
 283             mapper = ArvPathMapper(self.arvrunner, reffiles+self.wf_reffiles, runtimeContext.basedir,
 284                                    "/keep/%s",
 285                                    "/keep/%s/%s")
 286
 287             # For containers API, we need to make sure any extra
 288             # referenced files (ie referenced by the workflow but
 289             # not in the inputs) are included in the mounts.
 290             if self.wf_reffiles:
 291                 runtimeContext = runtimeContext.copy()
 292                 runtimeContext.extra_reffiles = copy.deepcopy(self.wf_reffiles)
 293
 294             def keepmount(obj):
 295                 remove_redundant_fields(obj)
 296                 with SourceLine(obj, None, WorkflowException, logger.isEnabledFor(logging.DEBUG)):
 297                     if "location" not in obj:
 298                         raise WorkflowException("%s object is missing required 'location' field: %s" % (obj["class"], obj))
 299                 with SourceLine(obj, "location", WorkflowException, logger.isEnabledFor(logging.DEBUG)):
 300                     if obj["location"].startswith("keep:"):
 301                         obj["location"] = mapper.mapper(obj["location"]).target
 302                         if "listing" in obj:
 303                             del obj["listing"]
 304                     elif obj["location"].startswith("_:"):
 305                         del obj["location"]
 306                     else:
 307                         raise WorkflowException("Location is not a keep reference or a literal: '%s'" % obj["location"])
 308
 309             visit_class(joborder_keepmount, ("File", "Directory"), keepmount)
 310
 311             def resolved(obj):
 312                 if obj["location"].startswith("keep:"):
 313                     obj["location"] = mapper.mapper(obj["location"]).resolved
 314
 315             visit_class(joborder_resolved, ("File", "Directory"), resolved)
 316
 317             if self.wf_pdh is None:
 318                 adjustFileObjs(packed, keepmount)
 319                 adjustDirObjs(packed, keepmount)
 320                 self.wf_pdh = upload_workflow_collection(self.arvrunner, shortname(self.tool["id"]), packed)
 321
 322         self.loadingContext = self.loadingContext.copy()
 323         self.loadingContext.metadata = self.loadingContext.metadata.copy()
 324         self.loadingContext.metadata["http://commonwl.org/cwltool#original_cwlVersion"] = "v1.0"
 325
 326         if len(job_res_reqs) == 1:
 327             # RAM request needs to be at least 128 MiB or the workflow
 328             # runner itself won't run reliably.
 329             if job_res_reqs[0].get("ramMin", 1024) < 128:
 330                 job_res_reqs[0]["ramMin"] = 128
 331
 332         wf_runner = cmap({
 333             "class": "CommandLineTool",
 334             "baseCommand": "cwltool",
 335             "inputs": self.tool["inputs"],
 336             "outputs": self.tool["outputs"],
 337             "stdout": "cwl.output.json",
 338             "requirements": self.requirements+job_res_reqs+[
 339                 {"class": "InlineJavascriptRequirement"},
 340                 {
 341                 "class": "InitialWorkDirRequirement",
 342                 "listing": [{
 343                         "entryname": "workflow.cwl",
 344                         "entry": '$({"class": "File", "location": "keep:%s/workflow.cwl"})' % self.wf_pdh
 345                     }, {
 346                         "entryname": "cwl.input.yml",
 347                         "entry": json.dumps(joborder_keepmount, indent=2, sort_keys=True, separators=(',',': ')).replace("\\", "\\\\").replace('$(', '\$(').replace('${', '\${')
 348                     }]
 349             }],
 350             "hints": self.hints,
 351             "arguments": ["--no-container", "--move-outputs", "--preserve-entire-environment", "workflow.cwl#main", "cwl.input.yml"],
 352             "id": "#"
 353         })
 354         return ArvadosCommandTool(self.arvrunner, wf_runner, self.loadingContext).job(joborder_resolved, output_callback, runtimeContext)
 355
 356     def make_workflow_step(self,
 357                            toolpath_object,      # type: Dict[Text, Any]
 358                            pos,                  # type: int
 359                            loadingContext,       # type: LoadingContext
 360                            *argc,
 361                            **argv
 362     ):
 363         # (...) -> WorkflowStep
 364         return ArvadosWorkflowStep(toolpath_object, pos, loadingContext, self.arvrunner, *argc, **argv)