sdk/cwl/arvados_cwl/arvjob.py

   1 import logging
   2 import re
   3 from . import done
   4 from .arvdocker import arv_docker_get_image
   5 from cwltool.process import get_feature
   6 from cwltool.errors import WorkflowException
   7 import arvados.collection
   8
   9 logger = logging.getLogger('arvados.cwl-runner')
  10
  11 tmpdirre = re.compile(r"^\S+ \S+ \d+ \d+ stderr \S+ \S+ crunchrunner: \$\(task\.tmpdir\)=(.*)")
  12 outdirre = re.compile(r"^\S+ \S+ \d+ \d+ stderr \S+ \S+ crunchrunner: \$\(task\.outdir\)=(.*)")
  13 keepre = re.compile(r"^\S+ \S+ \d+ \d+ stderr \S+ \S+ crunchrunner: \$\(task\.keep\)=(.*)")
  14
  15 class ArvadosJob(object):
  16     """Submit and manage a Crunch job for executing a CWL CommandLineTool."""
  17
  18     def __init__(self, runner):
  19         self.arvrunner = runner
  20         self.running = False
  21
  22     def run(self, dry_run=False, pull_image=True, **kwargs):
  23         script_parameters = {
  24             "command": self.command_line
  25         }
  26         runtime_constraints = {}
  27
  28         if self.generatefiles:
  29             vwd = arvados.collection.Collection()
  30             script_parameters["task.vwd"] = {}
  31             for t in self.generatefiles:
  32                 if isinstance(self.generatefiles[t], dict):
  33                     src, rest = self.arvrunner.fs_access.get_collection(self.generatefiles[t]["path"].replace("$(task.keep)/", "keep:"))
  34                     vwd.copy(rest, t, source_collection=src)
  35                 else:
  36                     with vwd.open(t, "w") as f:
  37                         f.write(self.generatefiles[t])
  38             vwd.save_new()
  39             for t in self.generatefiles:
  40                 script_parameters["task.vwd"][t] = "$(task.keep)/%s/%s" % (vwd.portable_data_hash(), t)
  41
  42         script_parameters["task.env"] = {"TMPDIR": "$(task.tmpdir)"}
  43         if self.environment:
  44             script_parameters["task.env"].update(self.environment)
  45
  46         if self.stdin:
  47             script_parameters["task.stdin"] = self.pathmapper.mapper(self.stdin)[1]
  48
  49         if self.stdout:
  50             script_parameters["task.stdout"] = self.stdout
  51
  52         (docker_req, docker_is_req) = get_feature(self, "DockerRequirement")
  53         if docker_req and kwargs.get("use_container") is not False:
  54             runtime_constraints["docker_image"] = arv_docker_get_image(self.arvrunner.api, docker_req, pull_image, self.arvrunner.project_uuid)
  55         else:
  56             runtime_constraints["docker_image"] = "arvados/jobs"
  57
  58         resources = self.builder.resources
  59         if resources is not None:
  60             runtime_constraints["min_cores_per_node"] = resources.get("cores", 1)
  61             runtime_constraints["min_ram_mb_per_node"] = resources.get("ram")
  62             runtime_constraints["min_scratch_mb_per_node"] = resources.get("tmpdirSize", 0) + resources.get("outdirSize", 0)
  63
  64         filters = [["repository", "=", "arvados"],
  65                    ["script", "=", "crunchrunner"],
  66                    ["script_version", "in git", "9e5b98e8f5f4727856b53447191f9c06e3da2ba6"]]
  67         if not self.arvrunner.ignore_docker_for_reuse:
  68             filters.append(["docker_image_locator", "in docker", runtime_constraints["docker_image"]])
  69
  70         try:
  71             response = self.arvrunner.api.jobs().create(
  72                 body={
  73                     "owner_uuid": self.arvrunner.project_uuid,
  74                     "script": "crunchrunner",
  75                     "repository": "arvados",
  76                     "script_version": "master",
  77                     "minimum_script_version": "9e5b98e8f5f4727856b53447191f9c06e3da2ba6",
  78                     "script_parameters": {"tasks": [script_parameters]},
  79                     "runtime_constraints": runtime_constraints
  80                 },
  81                 filters=filters,
  82                 find_or_create=kwargs.get("enable_reuse", True)
  83             ).execute(num_retries=self.arvrunner.num_retries)
  84
  85             self.arvrunner.jobs[response["uuid"]] = self
  86
  87             self.update_pipeline_component(response)
  88
  89             logger.info("Job %s (%s) is %s", self.name, response["uuid"], response["state"])
  90
  91             if response["state"] in ("Complete", "Failed", "Cancelled"):
  92                 self.done(response)
  93         except Exception as e:
  94             logger.error("Got error %s" % str(e))
  95             self.output_callback({}, "permanentFail")
  96
  97     def update_pipeline_component(self, record):
  98         if self.arvrunner.pipeline:
  99             self.arvrunner.pipeline["components"][self.name] = {"job": record}
 100             self.arvrunner.pipeline = self.arvrunner.api.pipeline_instances().update(uuid=self.arvrunner.pipeline["uuid"],
 101                                                                                  body={
 102                                                                                     "components": self.arvrunner.pipeline["components"]
 103                                                                                  }).execute(num_retries=self.arvrunner.num_retries)
 104         if self.arvrunner.uuid:
 105             try:
 106                 job = self.arvrunner.api.jobs().get(uuid=self.arvrunner.uuid).execute()
 107                 if job:
 108                     components = job["components"]
 109                     components[self.name] = record["uuid"]
 110                     self.arvrunner.api.jobs().update(uuid=self.arvrunner.uuid,
 111                         body={
 112                             "components": components
 113                         }).execute(num_retries=self.arvrunner.num_retries)
 114             except Exception as e:
 115                 logger.info("Error adding to components: %s", e)
 116
 117     def done(self, record):
 118         try:
 119             self.update_pipeline_component(record)
 120         except:
 121             pass
 122
 123         try:
 124             if record["state"] == "Complete":
 125                 processStatus = "success"
 126             else:
 127                 processStatus = "permanentFail"
 128
 129             outputs = {}
 130             try:
 131                 if record["output"]:
 132                     logc = arvados.collection.Collection(record["log"])
 133                     log = logc.open(logc.keys()[0])
 134                     tmpdir = None
 135                     outdir = None
 136                     keepdir = None
 137                     for l in log:
 138                         # Determine the tmpdir, outdir and keepdir paths from
 139                         # the job run.  Unfortunately, we can't take the first
 140                         # values we find (which are expected to be near the
 141                         # top) and stop scanning because if the node fails and
 142                         # the job restarts on a different node these values
 143                         # will different runs, and we need to know about the
 144                         # final run that actually produced output.
 145
 146                         g = tmpdirre.match(l)
 147                         if g:
 148                             tmpdir = g.group(1)
 149                         g = outdirre.match(l)
 150                         if g:
 151                             outdir = g.group(1)
 152                         g = keepre.match(l)
 153                         if g:
 154                             keepdir = g.group(1)
 155
 156                     outputs = done.done(self, record, tmpdir, outdir, keepdir)
 157             except WorkflowException as e:
 158                 logger.error("Error while collecting job outputs:\n%s", e, exc_info=(e if self.arvrunner.debug else False))
 159                 processStatus = "permanentFail"
 160             except Exception as e:
 161                 logger.exception("Got unknown exception while collecting job outputs:")
 162                 processStatus = "permanentFail"
 163
 164             self.output_callback(outputs, processStatus)
 165         finally:
 166             del self.arvrunner.jobs[record["uuid"]]