sdk/cwl/arvados_cwl/runner.py

   1 import os
   2 import urlparse
   3 from functools import partial
   4 import logging
   5 import json
   6 import re
   7 from cStringIO import StringIO
   8
   9 import cwltool.draft2tool
  10 from cwltool.draft2tool import CommandLineTool
  11 import cwltool.workflow
  12 from cwltool.process import get_feature, scandeps, UnsupportedRequirement, normalizeFilesDirs, shortname
  13 from cwltool.load_tool import fetch_document
  14 from cwltool.pathmapper import adjustFileObjs, adjustDirObjs
  15 from cwltool.utils import aslist
  16 from cwltool.builder import substitute
  17
  18 import arvados.collection
  19 import ruamel.yaml as yaml
  20
  21 from .arvdocker import arv_docker_get_image
  22 from .pathmapper import ArvPathMapper
  23 from ._version import __version__
  24
  25 logger = logging.getLogger('arvados.cwl-runner')
  26
  27 cwltool.draft2tool.ACCEPTLIST_RE = re.compile(r".*")
  28
  29 def trim_listing(obj):
  30     """Remove 'listing' field from Directory objects that are keep references.
  31
  32     When Directory objects represent Keep references, it redundant and
  33     potentially very expensive to pass fully enumerated Directory objects
  34     between instances of cwl-runner (e.g. a submitting a job, or using the
  35     RunInSingleContainer feature), so delete the 'listing' field when it is
  36     safe to do so.
  37     """
  38
  39     if obj.get("location", "").startswith("keep:") and "listing" in obj:
  40         del obj["listing"]
  41     if obj.get("location", "").startswith("_:"):
  42         del obj["location"]
  43
  44 def upload_dependencies(arvrunner, name, document_loader,
  45                         workflowobj, uri, loadref_run):
  46     """Upload the dependencies of the workflowobj document to Keep.
  47
  48     Returns a pathmapper object mapping local paths to keep references.  Also
  49     does an in-place update of references in "workflowobj".
  50
  51     Use scandeps to find $import, $include, $schemas, run, File and Directory
  52     fields that represent external references.
  53
  54     If workflowobj has an "id" field, this will reload the document to ensure
  55     it is scanning the raw document prior to preprocessing.
  56     """
  57
  58     loaded = set()
  59     def loadref(b, u):
  60         joined = urlparse.urljoin(b, u)
  61         defrg, _ = urlparse.urldefrag(joined)
  62         if defrg not in loaded:
  63             loaded.add(defrg)
  64             # Use fetch_text to get raw file (before preprocessing).
  65             text = document_loader.fetch_text(defrg)
  66             if isinstance(text, bytes):
  67                 textIO = StringIO(text.decode('utf-8'))
  68             else:
  69                 textIO = StringIO(text)
  70             return yaml.safe_load(textIO)
  71         else:
  72             return {}
  73
  74     if loadref_run:
  75         loadref_fields = set(("$import", "run"))
  76     else:
  77         loadref_fields = set(("$import",))
  78
  79     scanobj = workflowobj
  80     if "id" in workflowobj:
  81         # Need raw file content (before preprocessing) to ensure
  82         # that external references in $include and $mixin are captured.
  83         scanobj = loadref("", workflowobj["id"])
  84
  85     sc = scandeps(uri, scanobj,
  86                   loadref_fields,
  87                   set(("$include", "$schemas", "location")),
  88                   loadref)
  89
  90     normalizeFilesDirs(sc)
  91
  92     if "id" in workflowobj:
  93         sc.append({"class": "File", "location": workflowobj["id"]})
  94
  95     mapper = ArvPathMapper(arvrunner, sc, "",
  96                            "keep:%s",
  97                            "keep:%s/%s",
  98                            name=name)
  99
 100     def setloc(p):
 101         if "location" in p and (not p["location"].startswith("_:")) and (not p["location"].startswith("keep:")):
 102             p["location"] = mapper.mapper(p["location"]).resolved
 103     adjustFileObjs(workflowobj, setloc)
 104     adjustDirObjs(workflowobj, setloc)
 105
 106     return mapper
 107
 108
 109 def upload_docker(arvrunner, tool):
 110     if isinstance(tool, CommandLineTool):
 111         (docker_req, docker_is_req) = get_feature(tool, "DockerRequirement")
 112         if docker_req:
 113             if docker_req.get("dockerOutputDirectory"):
 114                 # TODO: can be supported by containers API, but not jobs API.
 115                 raise UnsupportedRequirement("Option 'dockerOutputDirectory' of DockerRequirement not supported.")
 116             arv_docker_get_image(arvrunner.api, docker_req, True, arvrunner.project_uuid)
 117     elif isinstance(tool, cwltool.workflow.Workflow):
 118         for s in tool.steps:
 119             upload_docker(arvrunner, s.embedded_tool)
 120
 121 def upload_instance(arvrunner, name, tool, job_order):
 122         upload_docker(arvrunner, tool)
 123
 124         for t in tool.tool["inputs"]:
 125             def setSecondary(fileobj):
 126                 if isinstance(fileobj, dict) and fileobj.get("class") == "File":
 127                     if "secondaryFiles" not in fileobj:
 128                         fileobj["secondaryFiles"] = [{"location": substitute(fileobj["location"], sf), "class": "File"} for sf in t["secondaryFiles"]]
 129
 130                 if isinstance(fileobj, list):
 131                     for e in fileobj:
 132                         setSecondary(e)
 133
 134             if shortname(t["id"]) in job_order and t.get("secondaryFiles"):
 135                 setSecondary(job_order[shortname(t["id"])])
 136
 137         workflowmapper = upload_dependencies(arvrunner,
 138                                              name,
 139                                              tool.doc_loader,
 140                                              tool.tool,
 141                                              tool.tool["id"],
 142                                              True)
 143         jobmapper = upload_dependencies(arvrunner,
 144                                         os.path.basename(job_order.get("id", "#")),
 145                                         tool.doc_loader,
 146                                         job_order,
 147                                         job_order.get("id", "#"),
 148                                         False)
 149
 150         if "id" in job_order:
 151             del job_order["id"]
 152
 153         return workflowmapper
 154
 155 def arvados_jobs_image(arvrunner):
 156     img = "arvados/jobs:"+__version__
 157     try:
 158         arv_docker_get_image(arvrunner.api, {"dockerPull": img}, True, arvrunner.project_uuid)
 159     except Exception as e:
 160         raise Exception("Docker image %s is not available\n%s" % (img, e) )
 161     return img
 162
 163 class Runner(object):
 164     def __init__(self, runner, tool, job_order, enable_reuse,
 165                  output_name, output_tags, submit_runner_ram=0,
 166                  name=None):
 167         self.arvrunner = runner
 168         self.tool = tool
 169         self.job_order = job_order
 170         self.running = False
 171         self.enable_reuse = enable_reuse
 172         self.uuid = None
 173         self.final_output = None
 174         self.output_name = output_name
 175         self.output_tags = output_tags
 176         self.name = name
 177
 178         if submit_runner_ram:
 179             self.submit_runner_ram = submit_runner_ram
 180         else:
 181             self.submit_runner_ram = 1024
 182
 183         if self.submit_runner_ram <= 0:
 184             raise Exception("Value of --submit-runner-ram must be greater than zero")
 185
 186     def update_pipeline_component(self, record):
 187         pass
 188
 189     def arvados_job_spec(self, *args, **kwargs):
 190         if self.name is None:
 191             self.name = os.path.basename(self.tool.tool["id"])
 192         workflowmapper = upload_instance(self.arvrunner, self.name, self.tool, self.job_order)
 193         adjustDirObjs(self.job_order, trim_listing)
 194         return workflowmapper
 195
 196     def done(self, record):
 197         if record["state"] == "Complete":
 198             if record.get("exit_code") is not None:
 199                 if record["exit_code"] == 33:
 200                     processStatus = "UnsupportedRequirement"
 201                 elif record["exit_code"] == 0:
 202                     processStatus = "success"
 203                 else:
 204                     processStatus = "permanentFail"
 205             else:
 206                 processStatus = "success"
 207         else:
 208             processStatus = "permanentFail"
 209
 210         outputs = {}
 211         try:
 212             try:
 213                 self.final_output = record["output"]
 214                 outc = arvados.collection.CollectionReader(self.final_output,
 215                                                            api_client=self.arvrunner.api,
 216                                                            keep_client=self.arvrunner.keep_client,
 217                                                            num_retries=self.arvrunner.num_retries)
 218                 if "cwl.output.json" in outc:
 219                     with outc.open("cwl.output.json") as f:
 220                         if f.size() > 0:
 221                             outputs = json.load(f)
 222                 def keepify(fileobj):
 223                     path = fileobj["location"]
 224                     if not path.startswith("keep:"):
 225                         fileobj["location"] = "keep:%s/%s" % (record["output"], path)
 226                 adjustFileObjs(outputs, keepify)
 227                 adjustDirObjs(outputs, keepify)
 228             except Exception as e:
 229                 logger.exception("While getting final output object: %s", e)
 230             self.arvrunner.output_callback(outputs, processStatus)
 231         finally:
 232             if record["uuid"] in self.arvrunner.processes:
 233                 del self.arvrunner.processes[record["uuid"]]