+import logging
+import json
+import os
+
+import ruamel.yaml as yaml
+
+from cwltool.errors import WorkflowException
+from cwltool.process import get_feature, UnsupportedRequirement, shortname
+from cwltool.pathmapper import adjustFiles
+from cwltool.utils import aslist
+
+import arvados.collection
+
+from .arvdocker import arv_docker_get_image
+from . import done
+from .runner import Runner, arvados_jobs_image
+from .fsaccess import CollectionFetcher
+
+logger = logging.getLogger('arvados.cwl-runner')
+
class ArvadosContainer(object):
- """Submit and manage a Crunch job for executing a CWL CommandLineTool."""
+ """Submit and manage a Crunch container request for executing a CWL CommandLineTool."""
def __init__(self, runner):
self.arvrunner = runner
self.running = False
+ self.uuid = None
+
+ def update_pipeline_component(self, r):
+ pass
def run(self, dry_run=False, pull_image=True, **kwargs):
container_request = {
- "command": self.command_line
+ "command": self.command_line,
"owner_uuid": self.arvrunner.project_uuid,
"name": self.name,
- "output_path", "/var/spool/cwl",
- "cwd", "/var/spool/cwl",
- "priority": 1
+ "output_path": self.outdir,
+ "cwd": self.outdir,
+ "priority": 1,
+ "state": "Committed",
+ "properties": {}
}
runtime_constraints = {}
- mounts = {}
-
- if self.generatefiles:
- vwd = arvados.collection.Collection()
- container_request["task.vwd"] = {}
- for t in self.generatefiles:
- if isinstance(self.generatefiles[t], dict):
- src, rest = self.arvrunner.fs_access.get_collection(self.generatefiles[t]["path"].replace("$(task.keep)/", "keep:"))
- vwd.copy(rest, t, source_collection=src)
- else:
- with vwd.open(t, "w") as f:
- f.write(self.generatefiles[t])
- vwd.save_new()
- # TODO
- # for t in self.generatefiles:
- # container_request["task.vwd"][t] = "$(task.keep)/%s/%s" % (vwd.portable_data_hash(), t)
-
- container_request["environment"] = {"TMPDIR": "/tmp"}
+ mounts = {
+ self.outdir: {
+ "kind": "tmp"
+ }
+ }
+ scheduling_parameters = {}
+
+ dirs = set()
+ for f in self.pathmapper.files():
+ _, p, tp = self.pathmapper.mapper(f)
+ if tp == "Directory" and '/' not in p[6:]:
+ mounts[p] = {
+ "kind": "collection",
+ "portable_data_hash": p[6:]
+ }
+ dirs.add(p[6:])
+ for f in self.pathmapper.files():
+ _, p, tp = self.pathmapper.mapper(f)
+ if p[6:].split("/")[0] not in dirs:
+ mounts[p] = {
+ "kind": "collection",
+ "portable_data_hash": p[6:]
+ }
+
+ if self.generatefiles["listing"]:
+ raise UnsupportedRequirement("InitialWorkDirRequirement not supported with --api=containers")
+
+ container_request["environment"] = {"TMPDIR": self.tmpdir, "HOME": self.outdir}
if self.environment:
container_request["environment"].update(self.environment)
- # TODO, not supported
- #if self.stdin:
- # container_request["task.stdin"] = self.pathmapper.mapper(self.stdin)[1]
+ if self.stdin:
+ raise UnsupportedRequirement("Stdin redirection currently not suppported")
+
+ if self.stderr:
+ raise UnsupportedRequirement("Stderr redirection currently not suppported")
if self.stdout:
mounts["stdout"] = {"kind": "file",
- "path": self.stdout}
+ "path": "%s/%s" % (self.outdir, self.stdout)}
(docker_req, docker_is_req) = get_feature(self, "DockerRequirement")
if not docker_req:
- docker_req = "arvados/jobs"
+ docker_req = {"dockerImageId": arvados_jobs_image(self.arvrunner)}
container_request["container_image"] = arv_docker_get_image(self.arvrunner.api,
docker_req,
if resources is not None:
runtime_constraints["vcpus"] = resources.get("cores", 1)
runtime_constraints["ram"] = resources.get("ram") * 2**20
- #runtime_constraints["min_scratch_mb_per_node"] = resources.get("tmpdirSize", 0) + resources.get("outdirSize", 0)
+
+ api_req, _ = get_feature(self, "http://arvados.org/cwl#APIRequirement")
+ if api_req:
+ runtime_constraints["API"] = True
+
+ runtime_req, _ = get_feature(self, "http://arvados.org/cwl#RuntimeConstraints")
+ if runtime_req:
+ if "keep_cache" in runtime_req:
+ runtime_constraints["keep_cache_ram"] = runtime_req["keep_cache"]
+
+ partition_req, _ = get_feature(self, "http://arvados.org/cwl#PartitionRequirement")
+ if partition_req:
+ scheduling_parameters["partitions"] = aslist(partition_req["partition"])
container_request["mounts"] = mounts
container_request["runtime_constraints"] = runtime_constraints
+ container_request["use_existing"] = kwargs.get("enable_reuse", True)
+ container_request["scheduling_parameters"] = scheduling_parameters
+
+ if kwargs.get("runnerjob", "").startswith("arvwf:"):
+ wfuuid = kwargs["runnerjob"][6:kwargs["runnerjob"].index("#")]
+ wfrecord = self.arvrunner.api.workflows().get(uuid=wfuuid).execute(num_retries=self.arvrunner.num_retries)
+ if container_request["name"] == "main":
+ container_request["name"] = wfrecord["name"]
+ container_request["properties"]["template_uuid"] = wfuuid
try:
response = self.arvrunner.api.container_requests().create(
body=container_request
).execute(num_retries=self.arvrunner.num_retries)
- self.arvrunner.jobs[response["uuid"]] = self
+ self.uuid = response["uuid"]
+ self.arvrunner.processes[self.uuid] = self
- logger.info("Container %s (%s) is %s", self.name, response["uuid"], response["state"])
+ logger.info("Container request %s (%s) state is %s", self.name, response["uuid"], response["state"])
- if response["state"] in ("Complete", "Cancelled"):
+ if response["state"] == "Final":
self.done(response)
except Exception as e:
logger.error("Got error %s" % str(e))
def done(self, record):
try:
- if record["state"] == "Complete":
- processStatus = "success"
+ container = self.arvrunner.api.containers().get(
+ uuid=record["container_uuid"]
+ ).execute(num_retries=self.arvrunner.num_retries)
+ if container["state"] == "Complete":
+ rcode = container["exit_code"]
+ if self.successCodes and rcode in self.successCodes:
+ processStatus = "success"
+ elif self.temporaryFailCodes and rcode in self.temporaryFailCodes:
+ processStatus = "temporaryFail"
+ elif self.permanentFailCodes and rcode in self.permanentFailCodes:
+ processStatus = "permanentFail"
+ elif rcode == 0:
+ processStatus = "success"
+ else:
+ processStatus = "permanentFail"
else:
processStatus = "permanentFail"
- try:
- outputs = {}
- if record["output"]:
- logc = arvados.collection.Collection(record["log"])
- log = logc.open(logc.keys()[0])
- tmpdir = None
- outdir = None
- keepdir = None
- for l in log:
- # Determine the tmpdir, outdir and keepdir paths from
- # the job run. Unfortunately, we can't take the first
- # values we find (which are expected to be near the
- # top) and stop scanning because if the node fails and
- # the job restarts on a different node these values
- # will different runs, and we need to know about the
- # final run that actually produced output.
-
- g = tmpdirre.match(l)
- if g:
- tmpdir = g.group(1)
- g = outdirre.match(l)
- if g:
- outdir = g.group(1)
- g = keepre.match(l)
- if g:
- keepdir = g.group(1)
-
- colname = "Output %s of %s" % (record["output"][0:7], self.name)
-
- # check if collection already exists with same owner, name and content
- collection_exists = self.arvrunner.api.collections().list(
- filters=[["owner_uuid", "=", self.arvrunner.project_uuid],
- ['portable_data_hash', '=', record["output"]],
- ["name", "=", colname]]
- ).execute(num_retries=self.arvrunner.num_retries)
-
- if not collection_exists["items"]:
- # Create a collection located in the same project as the
- # pipeline with the contents of the output.
- # First, get output record.
- collections = self.arvrunner.api.collections().list(
- limit=1,
- filters=[['portable_data_hash', '=', record["output"]]],
- select=["manifest_text"]
- ).execute(num_retries=self.arvrunner.num_retries)
-
- if not collections["items"]:
- raise WorkflowException(
- "Job output '%s' cannot be found on API server" % (
- record["output"]))
-
- # Create new collection in the parent project
- # with the output contents.
- self.arvrunner.api.collections().create(body={
- "owner_uuid": self.arvrunner.project_uuid,
- "name": colname,
- "portable_data_hash": record["output"],
- "manifest_text": collections["items"][0]["manifest_text"]
- }, ensure_unique_name=True).execute(
- num_retries=self.arvrunner.num_retries)
-
- self.builder.outdir = outdir
- self.builder.pathmapper.keepdir = keepdir
- outputs = self.collect_outputs("keep:" + record["output"])
- except WorkflowException as e:
- logger.error("Error while collecting job outputs:\n%s", e, exc_info=(e if self.arvrunner.debug else False))
- processStatus = "permanentFail"
- except Exception as e:
- logger.exception("Got unknown exception while collecting job outputs:")
- processStatus = "permanentFail"
+ outputs = {}
+ if container["output"]:
+ try:
+ outputs = done.done_outputs(self, container, "/tmp", self.outdir, "/keep")
+ except Exception as e:
+ logger.error("Got error %s" % str(e))
+ self.output_callback({}, "permanentFail")
self.output_callback(outputs, processStatus)
finally:
- del self.arvrunner.jobs[record["uuid"]]
+ del self.arvrunner.processes[record["uuid"]]
+
+
+class RunnerContainer(Runner):
+ """Submit and manage a container that runs arvados-cwl-runner."""
+
+ def arvados_job_spec(self, dry_run=False, pull_image=True, **kwargs):
+ """Create an Arvados container request for this workflow.
+
+ The returned dict can be used to create a container passed as
+ the +body+ argument to container_requests().create().
+ """
+
+ workflowmapper = super(RunnerContainer, self).arvados_job_spec(dry_run=dry_run, pull_image=pull_image, **kwargs)
+
+ container_req = {
+ "owner_uuid": self.arvrunner.project_uuid,
+ "name": self.name,
+ "output_path": "/var/spool/cwl",
+ "cwd": "/var/spool/cwl",
+ "priority": 1,
+ "state": "Committed",
+ "container_image": arvados_jobs_image(self.arvrunner),
+ "mounts": {
+ "/var/lib/cwl/cwl.input.json": {
+ "kind": "json",
+ "content": self.job_order
+ },
+ "stdout": {
+ "kind": "file",
+ "path": "/var/spool/cwl/cwl.output.json"
+ },
+ "/var/spool/cwl": {
+ "kind": "collection",
+ "writable": True
+ }
+ },
+ "runtime_constraints": {
+ "vcpus": 1,
+ "ram": 1024*1024 * self.submit_runner_ram,
+ "API": True
+ },
+ "properties": {}
+ }
+
+ workflowcollection = workflowmapper.mapper(self.tool.tool["id"])[1]
+ if workflowcollection.startswith("keep:"):
+ workflowcollection = workflowcollection[5:workflowcollection.index('/')]
+ workflowname = os.path.basename(self.tool.tool["id"])
+ workflowpath = "/var/lib/cwl/workflow/%s" % workflowname
+ container_req["mounts"]["/var/lib/cwl/workflow"] = {
+ "kind": "collection",
+ "portable_data_hash": "%s" % workflowcollection
+ }
+ elif workflowcollection.startswith("arvwf:"):
+ workflowpath = "/var/lib/cwl/workflow.json#main"
+ wfuuid = workflowcollection[6:workflowcollection.index("#")]
+ wfrecord = self.arvrunner.api.workflows().get(uuid=wfuuid).execute(num_retries=self.arvrunner.num_retries)
+ wfobj = yaml.safe_load(wfrecord["definition"])
+ if container_req["name"].startswith("arvwf:"):
+ container_req["name"] = wfrecord["name"]
+ container_req["mounts"]["/var/lib/cwl/workflow.json"] = {
+ "kind": "json",
+ "json": wfobj
+ }
+ container_req["properties"]["template_uuid"] = wfuuid
+
+ command = ["arvados-cwl-runner", "--local", "--api=containers"]
+ if self.output_name:
+ command.append("--output-name=" + self.output_name)
+
+ if self.output_tags:
+ command.append("--output-tags=" + self.output_tags)
+
+ if self.enable_reuse:
+ command.append("--enable-reuse")
+ else:
+ command.append("--disable-reuse")
+
+ command.extend([workflowpath, "/var/lib/cwl/cwl.input.json"])
+
+ container_req["command"] = command
+
+ return container_req
+
+
+ def run(self, *args, **kwargs):
+ kwargs["keepprefix"] = "keep:"
+ job_spec = self.arvados_job_spec(*args, **kwargs)
+ job_spec.setdefault("owner_uuid", self.arvrunner.project_uuid)
+
+ response = self.arvrunner.api.container_requests().create(
+ body=job_spec
+ ).execute(num_retries=self.arvrunner.num_retries)
+
+ self.uuid = response["uuid"]
+ self.arvrunner.processes[self.uuid] = self
+
+ logger.info("Submitted container %s", response["uuid"])
+
+ if response["state"] == "Final":
+ self.done(response)
+
+ def done(self, record):
+ try:
+ container = self.arvrunner.api.containers().get(
+ uuid=record["container_uuid"]
+ ).execute(num_retries=self.arvrunner.num_retries)
+ except Exception as e:
+ logger.exception("While getting runner container: %s", e)
+ self.arvrunner.output_callback({}, "permanentFail")
+ del self.arvrunner.processes[record["uuid"]]
+ else:
+ super(RunnerContainer, self).done(container)
+ finally:
+ del self.arvrunner.processes[record["uuid"]]