+# Copyright (C) The Arvados Authors. All rights reserved.
+#
+# SPDX-License-Identifier: Apache-2.0
+
import os
import urlparse
from functools import partial
import logging
import json
-import re
-from cStringIO import StringIO
+import subprocess
+
+from StringIO import StringIO
+
+from schema_salad.sourceline import SourceLine
import cwltool.draft2tool
from cwltool.draft2tool import CommandLineTool
from cwltool.pathmapper import adjustFileObjs, adjustDirObjs
from cwltool.utils import aslist
from cwltool.builder import substitute
+from cwltool.pack import pack
import arvados.collection
import ruamel.yaml as yaml
from .arvdocker import arv_docker_get_image
-from .pathmapper import ArvPathMapper
+from .pathmapper import ArvPathMapper, trim_listing
from ._version import __version__
+from . import done
logger = logging.getLogger('arvados.cwl-runner')
-cwltool.draft2tool.ACCEPTLIST_RE = re.compile(r".*")
+def trim_anonymous_location(obj):
+ """Remove 'location' field from File and Directory literals.
-def trim_listing(obj):
- """Remove 'listing' field from Directory objects that are keep references.
+ To make internal handling easier, literals are assigned a random id for
+ 'location'. However, when writing the record back out, this can break
+ reproducibility. Since it is valid for literals not have a 'location'
+ field, remove it.
- When Directory objects represent Keep references, it redundant and
- potentially very expensive to pass fully enumerated Directory objects
- between instances of cwl-runner (e.g. a submitting a job, or using the
- RunInSingleContainer feature), so delete the 'listing' field when it is
- safe to do so.
"""
- if obj.get("location", "").startswith("keep:") and "listing" in obj:
- del obj["listing"]
if obj.get("location", "").startswith("_:"):
del obj["location"]
def upload_dependencies(arvrunner, name, document_loader,
- workflowobj, uri, loadref_run):
+ workflowobj, uri, loadref_run, include_primary=True):
"""Upload the dependencies of the workflowobj document to Keep.
Returns a pathmapper object mapping local paths to keep references. Also
normalizeFilesDirs(sc)
- if "id" in workflowobj:
+ if include_primary and "id" in workflowobj:
sc.append({"class": "File", "location": workflowobj["id"]})
+ if "$schemas" in workflowobj:
+ for s in workflowobj["$schemas"]:
+ sc.append({"class": "File", "location": s})
+
mapper = ArvPathMapper(arvrunner, sc, "",
"keep:%s",
"keep:%s/%s",
- name=name)
+ name=name,
+ single_collection=True)
def setloc(p):
if "location" in p and (not p["location"].startswith("_:")) and (not p["location"].startswith("keep:")):
adjustFileObjs(workflowobj, setloc)
adjustDirObjs(workflowobj, setloc)
+ if "$schemas" in workflowobj:
+ sch = []
+ for s in workflowobj["$schemas"]:
+ sch.append(mapper.mapper(s).resolved)
+ workflowobj["$schemas"] = sch
+
return mapper
def upload_docker(arvrunner, tool):
+ """Uploads Docker images used in CommandLineTool objects."""
+
if isinstance(tool, CommandLineTool):
(docker_req, docker_is_req) = get_feature(tool, "DockerRequirement")
if docker_req:
if docker_req.get("dockerOutputDirectory"):
# TODO: can be supported by containers API, but not jobs API.
- raise UnsupportedRequirement("Option 'dockerOutputDirectory' of DockerRequirement not supported.")
+ raise SourceLine(docker_req, "dockerOutputDirectory", UnsupportedRequirement).makeError(
+ "Option 'dockerOutputDirectory' of DockerRequirement not supported.")
arv_docker_get_image(arvrunner.api, docker_req, True, arvrunner.project_uuid)
elif isinstance(tool, cwltool.workflow.Workflow):
for s in tool.steps:
upload_docker(arvrunner, s.embedded_tool)
-def upload_instance(arvrunner, name, tool, job_order):
- upload_docker(arvrunner, tool)
-
- for t in tool.tool["inputs"]:
- def setSecondary(fileobj):
- if isinstance(fileobj, dict) and fileobj.get("class") == "File":
- if "secondaryFiles" not in fileobj:
- fileobj["secondaryFiles"] = [{"location": substitute(fileobj["location"], sf), "class": "File"} for sf in t["secondaryFiles"]]
-
- if isinstance(fileobj, list):
- for e in fileobj:
- setSecondary(e)
-
- if shortname(t["id"]) in job_order and t.get("secondaryFiles"):
- setSecondary(job_order[shortname(t["id"])])
-
- workflowmapper = upload_dependencies(arvrunner,
- name,
- tool.doc_loader,
- tool.tool,
- tool.tool["id"],
- True)
- jobmapper = upload_dependencies(arvrunner,
- os.path.basename(job_order.get("id", "#")),
- tool.doc_loader,
- job_order,
- job_order.get("id", "#"),
- False)
-
- if "id" in job_order:
- del job_order["id"]
-
- return workflowmapper
-
-def arvados_jobs_image(arvrunner):
- img = "arvados/jobs:"+__version__
+def packed_workflow(arvrunner, tool):
+ """Create a packed workflow.
+
+ A "packed" workflow is one where all the components have been combined into a single document."""
+
+ return pack(tool.doc_loader, tool.doc_loader.fetch(tool.tool["id"]),
+ tool.tool["id"], tool.metadata)
+
+def tag_git_version(packed):
+ if tool.tool["id"].startswith("file://"):
+ path = os.path.dirname(tool.tool["id"][7:])
+ try:
+ githash = subprocess.check_output(['git', 'log', '--first-parent', '--max-count=1', '--format=%H'], stderr=subprocess.STDOUT, cwd=path).strip()
+ except (OSError, subprocess.CalledProcessError):
+ pass
+ else:
+ packed["http://schema.org/version"] = githash
+
+
+def upload_job_order(arvrunner, name, tool, job_order):
+ """Upload local files referenced in the input object and return updated input
+ object with 'location' updated to the proper keep references.
+ """
+
+ for t in tool.tool["inputs"]:
+ def setSecondary(fileobj):
+ if isinstance(fileobj, dict) and fileobj.get("class") == "File":
+ if "secondaryFiles" not in fileobj:
+ fileobj["secondaryFiles"] = [{"location": substitute(fileobj["location"], sf), "class": "File"} for sf in t["secondaryFiles"]]
+
+ if isinstance(fileobj, list):
+ for e in fileobj:
+ setSecondary(e)
+
+ if shortname(t["id"]) in job_order and t.get("secondaryFiles"):
+ setSecondary(job_order[shortname(t["id"])])
+
+ jobmapper = upload_dependencies(arvrunner,
+ name,
+ tool.doc_loader,
+ job_order,
+ job_order.get("id", "#"),
+ False)
+
+ if "id" in job_order:
+ del job_order["id"]
+
+ # Need to filter this out, gets added by cwltool when providing
+ # parameters on the command line.
+ if "job_order" in job_order:
+ del job_order["job_order"]
+
+ return job_order
+
+def upload_workflow_deps(arvrunner, tool, override_tools):
+ # Ensure that Docker images needed by this workflow are available
+
+ upload_docker(arvrunner, tool)
+
+ document_loader = tool.doc_loader
+
+ def upload_tool_deps(deptool):
+ if "id" in deptool:
+ upload_dependencies(arvrunner,
+ "%s dependencies" % (shortname(deptool["id"])),
+ document_loader,
+ deptool,
+ deptool["id"],
+ False,
+ include_primary=False)
+ document_loader.idx[deptool["id"]] = deptool
+ override_tools[deptool["id"]] = json.dumps(deptool)
+
+ tool.visit(upload_tool_deps)
+
+def arvados_jobs_image(arvrunner, img):
+ """Determine if the right arvados/jobs image version is available. If not, try to pull and upload it."""
+
try:
arv_docker_get_image(arvrunner.api, {"dockerPull": img}, True, arvrunner.project_uuid)
except Exception as e:
raise Exception("Docker image %s is not available\n%s" % (img, e) )
return img
+def upload_workflow_collection(arvrunner, name, packed):
+ collection = arvados.collection.Collection(api_client=arvrunner.api,
+ keep_client=arvrunner.keep_client,
+ num_retries=arvrunner.num_retries)
+ with collection.open("workflow.cwl", "w") as f:
+ f.write(json.dumps(packed, indent=2, sort_keys=True, separators=(',',': ')))
+
+ filters = [["portable_data_hash", "=", collection.portable_data_hash()],
+ ["name", "like", name+"%"]]
+ if arvrunner.project_uuid:
+ filters.append(["owner_uuid", "=", arvrunner.project_uuid])
+ exists = arvrunner.api.collections().list(filters=filters).execute(num_retries=arvrunner.num_retries)
+
+ if exists["items"]:
+ logger.info("Using collection %s", exists["items"][0]["uuid"])
+ else:
+ collection.save_new(name=name,
+ owner_uuid=arvrunner.project_uuid,
+ ensure_unique_name=True,
+ num_retries=arvrunner.num_retries)
+ logger.info("Uploaded to %s", collection.manifest_locator())
+
+ return collection.portable_data_hash()
+
+
class Runner(object):
+ """Base class for runner processes, which submit an instance of
+ arvados-cwl-runner and wait for the final result."""
+
def __init__(self, runner, tool, job_order, enable_reuse,
- output_name, output_tags, submit_runner_ram=0):
+ output_name, output_tags, submit_runner_ram=0,
+ name=None, on_error=None, submit_runner_image=None,
+ intermediate_output_ttl=0):
self.arvrunner = runner
self.tool = tool
self.job_order = job_order
self.final_output = None
self.output_name = output_name
self.output_tags = output_tags
+ self.name = name
+ self.on_error = on_error
+ self.jobs_image = submit_runner_image or "arvados/jobs:"+__version__
+ self.intermediate_output_ttl = intermediate_output_ttl
+
if submit_runner_ram:
self.submit_runner_ram = submit_runner_ram
else:
- self.submit_runner_ram = 1024
+ self.submit_runner_ram = 3000
if self.submit_runner_ram <= 0:
raise Exception("Value of --submit-runner-ram must be greater than zero")
def update_pipeline_component(self, record):
pass
- def arvados_job_spec(self, *args, **kwargs):
- self.name = os.path.basename(self.tool.tool["id"])
- workflowmapper = upload_instance(self.arvrunner, self.name, self.tool, self.job_order)
- adjustDirObjs(self.job_order, trim_listing)
- return workflowmapper
-
def done(self, record):
- if record["state"] == "Complete":
- if record.get("exit_code") is not None:
- if record["exit_code"] == 33:
- processStatus = "UnsupportedRequirement"
- elif record["exit_code"] == 0:
- processStatus = "success"
+ """Base method for handling a completed runner."""
+
+ try:
+ if record["state"] == "Complete":
+ if record.get("exit_code") is not None:
+ if record["exit_code"] == 33:
+ processStatus = "UnsupportedRequirement"
+ elif record["exit_code"] == 0:
+ processStatus = "success"
+ else:
+ processStatus = "permanentFail"
else:
- processStatus = "permanentFail"
+ processStatus = "success"
else:
- processStatus = "success"
- else:
- processStatus = "permanentFail"
+ processStatus = "permanentFail"
- outputs = {}
- try:
- try:
- self.final_output = record["output"]
- outc = arvados.collection.CollectionReader(self.final_output,
+ outputs = {}
+
+ if processStatus == "permanentFail":
+ logc = arvados.collection.CollectionReader(record["log"],
api_client=self.arvrunner.api,
keep_client=self.arvrunner.keep_client,
num_retries=self.arvrunner.num_retries)
- if "cwl.output.json" in outc:
- with outc.open("cwl.output.json") as f:
- if f.size() > 0:
- outputs = json.load(f)
- def keepify(fileobj):
- path = fileobj["location"]
- if not path.startswith("keep:"):
- fileobj["location"] = "keep:%s/%s" % (record["output"], path)
- adjustFileObjs(outputs, keepify)
- adjustDirObjs(outputs, keepify)
- except Exception as e:
- logger.exception("While getting final output object: %s", e)
+ done.logtail(logc, logger, "%s error log:" % self.arvrunner.label(self), maxlen=40)
+
+ self.final_output = record["output"]
+ outc = arvados.collection.CollectionReader(self.final_output,
+ api_client=self.arvrunner.api,
+ keep_client=self.arvrunner.keep_client,
+ num_retries=self.arvrunner.num_retries)
+ if "cwl.output.json" in outc:
+ with outc.open("cwl.output.json") as f:
+ if f.size() > 0:
+ outputs = json.load(f)
+ def keepify(fileobj):
+ path = fileobj["location"]
+ if not path.startswith("keep:"):
+ fileobj["location"] = "keep:%s/%s" % (record["output"], path)
+ adjustFileObjs(outputs, keepify)
+ adjustDirObjs(outputs, keepify)
+ except Exception as e:
+ logger.exception("[%s] While getting final output object: %s", self.name, e)
+ self.arvrunner.output_callback({}, "permanentFail")
+ else:
self.arvrunner.output_callback(outputs, processStatus)
finally:
if record["uuid"] in self.arvrunner.processes:
projects / arvados.git / blobdiff