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