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