X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/d9eec0b9bf52b428b5506617590d87b2d92eb246..9f35a3cf10e5d261d83556d5b33f242966101f02:/sdk/cwl/arvados_cwl/arvjob.py diff --git a/sdk/cwl/arvados_cwl/arvjob.py b/sdk/cwl/arvados_cwl/arvjob.py index e6d0cba65f..8b1a934683 100644 --- a/sdk/cwl/arvados_cwl/arvjob.py +++ b/sdk/cwl/arvados_cwl/arvjob.py @@ -1,6 +1,7 @@ import logging import re import copy +import json from cwltool.process import get_feature, shortname from cwltool.errors import WorkflowException @@ -17,6 +18,7 @@ from .perf import Perf from . import done logger = logging.getLogger('arvados.cwl-runner') +metrics = logging.getLogger('arvados.cwl-runner.metrics') tmpdirre = re.compile(r"^\S+ \S+ \d+ \d+ stderr \S+ \S+ crunchrunner: \$\(task\.tmpdir\)=(.*)") outdirre = re.compile(r"^\S+ \S+ \d+ \d+ stderr \S+ \S+ crunchrunner: \$\(task\.outdir\)=(.*)") @@ -36,21 +38,29 @@ class ArvadosJob(object): } runtime_constraints = {} - if self.generatefiles["listing"]: - vwd = arvados.collection.Collection() - script_parameters["task.vwd"] = {} - generatemapper = InitialWorkDirPathMapper([self.generatefiles], "", "", - separateDirs=False) - for f, p in generatemapper.items(): - if p.type == "CreateFile": - with vwd.open(p.target, "w") as n: - n.write(p.resolved.encode("utf-8")) - vwd.save_new() - for f, p in generatemapper.items(): - if p.type == "File": - script_parameters["task.vwd"][p.target] = p.resolved - if p.type == "CreateFile": - script_parameters["task.vwd"][p.target] = "$(task.keep)/%s/%s" % (vwd.portable_data_hash(), p.target) + with Perf(metrics, "generatefiles %s" % self.name): + if self.generatefiles["listing"]: + vwd = arvados.collection.Collection(api_client=self.arvrunner.api, + keep_client=self.arvrunner.keep_client, + num_retries=self.arvrunner.num_retries) + script_parameters["task.vwd"] = {} + generatemapper = InitialWorkDirPathMapper([self.generatefiles], "", "", + separateDirs=False) + + with Perf(metrics, "createfiles %s" % self.name): + for f, p in generatemapper.items(): + if p.type == "CreateFile": + with vwd.open(p.target, "w") as n: + n.write(p.resolved.encode("utf-8")) + + with Perf(metrics, "generatefiles.save_new %s" % self.name): + vwd.save_new() + + for f, p in generatemapper.items(): + if p.type == "File": + script_parameters["task.vwd"][p.target] = p.resolved + if p.type == "CreateFile": + script_parameters["task.vwd"][p.target] = "$(task.keep)/%s/%s" % (vwd.portable_data_hash(), p.target) script_parameters["task.env"] = {"TMPDIR": self.tmpdir, "HOME": self.outdir} if self.environment: @@ -72,11 +82,12 @@ class ArvadosJob(object): if self.permanentFailCodes: script_parameters["task.permanentFailCodes"] = self.permanentFailCodes - (docker_req, docker_is_req) = get_feature(self, "DockerRequirement") - if docker_req and kwargs.get("use_container") is not False: - runtime_constraints["docker_image"] = arv_docker_get_image(self.arvrunner.api, docker_req, pull_image, self.arvrunner.project_uuid) - else: - runtime_constraints["docker_image"] = "arvados/jobs" + with Perf(metrics, "arv_docker_get_image %s" % self.name): + (docker_req, docker_is_req) = get_feature(self, "DockerRequirement") + if docker_req and kwargs.get("use_container") is not False: + runtime_constraints["docker_image"] = arv_docker_get_image(self.arvrunner.api, docker_req, pull_image, self.arvrunner.project_uuid) + else: + runtime_constraints["docker_image"] = "arvados/jobs" resources = self.builder.resources if resources is not None: @@ -84,6 +95,16 @@ class ArvadosJob(object): runtime_constraints["min_ram_mb_per_node"] = resources.get("ram") runtime_constraints["min_scratch_mb_per_node"] = resources.get("tmpdirSize", 0) + resources.get("outdirSize", 0) + runtime_req, _ = get_feature(self, "http://arvados.org/cwl#RuntimeConstraints") + if runtime_req: + if "keep_cache" in runtime_req: + runtime_constraints["keep_cache_mb_per_task"] = runtime_req["keep_cache"] + if "outputDirType" in runtime_req: + if runtime_req["outputDirType"] == "local_output_dir": + script_parameters["task.keepTmpOutput"] = False + elif runtime_req["outputDirType"] == "keep_output_dir": + script_parameters["task.keepTmpOutput"] = True + filters = [["repository", "=", "arvados"], ["script", "=", "crunchrunner"], ["script_version", "in git", "9e5b98e8f5f4727856b53447191f9c06e3da2ba6"]] @@ -91,7 +112,7 @@ class ArvadosJob(object): filters.append(["docker_image_locator", "in docker", runtime_constraints["docker_image"]]) try: - with Perf(logger, "create %s" % self.name): + with Perf(metrics, "create %s" % self.name): response = self.arvrunner.api.jobs().create( body={ "owner_uuid": self.arvrunner.project_uuid, @@ -113,7 +134,7 @@ class ArvadosJob(object): logger.info("Job %s (%s) is %s", self.name, response["uuid"], response["state"]) if response["state"] in ("Complete", "Failed", "Cancelled"): - with Perf(logger, "done %s" % self.name): + with Perf(metrics, "done %s" % self.name): self.done(response) except Exception as e: logger.error("Got error %s" % str(e)) @@ -122,7 +143,7 @@ class ArvadosJob(object): def update_pipeline_component(self, record): if self.arvrunner.pipeline: self.arvrunner.pipeline["components"][self.name] = {"job": record} - with Perf(logger, "update_pipeline_component %s" % self.name): + with Perf(metrics, "update_pipeline_component %s" % self.name): self.arvrunner.pipeline = self.arvrunner.api.pipeline_instances().update(uuid=self.arvrunner.pipeline["uuid"], body={ "components": self.arvrunner.pipeline["components"] @@ -155,8 +176,11 @@ class ArvadosJob(object): outputs = {} try: if record["output"]: - with Perf(logger, "inspect log %s" % self.name): - logc = arvados.collection.Collection(record["log"]) + with Perf(metrics, "inspect log %s" % self.name): + logc = arvados.collection.CollectionReader(record["log"], + api_client=self.arvrunner.api, + keep_client=self.arvrunner.keep_client, + num_retries=self.arvrunner.num_retries) log = logc.open(logc.keys()[0]) tmpdir = None outdir = None @@ -180,14 +204,16 @@ class ArvadosJob(object): if g: keepdir = g.group(1) - with Perf(logger, "output collection %s" % self.name): + with Perf(metrics, "output collection %s" % self.name): outputs = done.done(self, record, tmpdir, outdir, keepdir) 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" + outputs = None except Exception as e: logger.exception("Got unknown exception while collecting job outputs:") processStatus = "permanentFail" + outputs = None self.output_callback(outputs, processStatus) finally: @@ -208,6 +234,8 @@ class RunnerJob(Runner): workflowmapper = super(RunnerJob, self).arvados_job_spec(dry_run=dry_run, pull_image=pull_image, **kwargs) self.job_order["cwl:tool"] = workflowmapper.mapper(self.tool.tool["id"]).target[5:] + if self.output_name: + self.job_order["arv:output_name"] = self.output_name return { "script": "cwl-runner", "script_version": "master", @@ -263,7 +291,8 @@ class RunnerTemplate(object): runner=runner, tool=tool, job_order=job_order, - enable_reuse=enable_reuse) + enable_reuse=enable_reuse, + output_name=None) def pipeline_component_spec(self): """Return a component that Workbench and a-r-p-i will understand.