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