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