X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/275637f3ce7fcaacd8f41cd038b15332785298ef..ecf3cf1ec8544cffacc620919e35404e086b931a:/sdk/cwl/arvados_cwl/arvjob.py diff --git a/sdk/cwl/arvados_cwl/arvjob.py b/sdk/cwl/arvados_cwl/arvjob.py index 4ba19a6bab..d104d56e91 100644 --- a/sdk/cwl/arvados_cwl/arvjob.py +++ b/sdk/cwl/arvados_cwl/arvjob.py @@ -1,26 +1,40 @@ +# Copyright (C) The Arvados Authors. All rights reserved. +# +# SPDX-License-Identifier: Apache-2.0 + import logging import re import copy +import json +import time -from cwltool.process import get_feature, shortname +from cwltool.process import get_feature, shortname, UnsupportedRequirement from cwltool.errors import WorkflowException from cwltool.draft2tool import revmap_file, CommandLineTool from cwltool.load_tool import fetch_document from cwltool.builder import Builder +from cwltool.pathmapper import adjustFileObjs, adjustDirObjs, visit_class + +from schema_salad.sourceline import SourceLine + +import ruamel.yaml as yaml import arvados.collection +from arvados.errors import ApiError from .arvdocker import arv_docker_get_image -from .runner import Runner -from .pathmapper import InitialWorkDirPathMapper +from .runner import Runner, arvados_jobs_image, packed_workflow, upload_workflow_collection, trim_anonymous_location, remove_redundant_fields +from .pathmapper import VwdPathMapper, trim_listing from .perf import Perf from . import done +from ._version import __version__ logger = logging.getLogger('arvados.cwl-runner') +metrics = logging.getLogger('arvados.cwl-runner.metrics') + +crunchrunner_re = re.compile(r"^\S+ \S+ \d+ \d+ stderr \S+ \S+ crunchrunner: \$\(task\.(tmpdir|outdir|keep)\)=(.*)") -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: \$\(task\.outdir\)=(.*)") -keepre = re.compile(r"^\S+ \S+ \d+ \d+ stderr \S+ \S+ crunchrunner: \$\(task\.keep\)=(.*)") +crunchrunner_git_commit = 'a3f2cb186e437bfce0031b024b2157b73ed2717d' class ArvadosJob(object): """Submit and manage a Crunch job for executing a CWL CommandLineTool.""" @@ -36,21 +50,30 @@ 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] = "$(task.keep)/%s/%s" % (vwd.portable_data_hash(), p.target) + with Perf(metrics, "generatefiles %s" % self.name): + if self.generatefiles["listing"]: + vwd = arvados.collection.Collection(api_client=self.arvrunner.api, + keep_client=self.arvrunner.keep_client, + num_retries=self.arvrunner.num_retries) + script_parameters["task.vwd"] = {} + generatemapper = VwdPathMapper([self.generatefiles], "", "", + separateDirs=False) + + with Perf(metrics, "createfiles %s" % self.name): + for f, p in generatemapper.items(): + if p.type == "CreateFile": + with vwd.open(p.target, "w") as n: + n.write(p.resolved.encode("utf-8")) + + if vwd: + with Perf(metrics, "generatefiles.save_new %s" % self.name): + vwd.save_new() + + for f, p in generatemapper.items(): + if p.type == "File": + script_parameters["task.vwd"][p.target] = p.resolved + if p.type == "CreateFile": + script_parameters["task.vwd"][p.target] = "$(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 +95,15 @@ 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: + if docker_req.get("dockerOutputDirectory"): + raise SourceLine(docker_req, "dockerOutputDirectory", UnsupportedRequirement).makeError( + "Option 'dockerOutputDirectory' of DockerRequirement not supported.") + 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: @@ -84,45 +111,79 @@ 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"] + runtime_constraints["min_ram_mb_per_node"] += 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"]] + ["script_version", "in git", crunchrunner_git_commit]] if not self.arvrunner.ignore_docker_for_reuse: filters.append(["docker_image_locator", "in docker", runtime_constraints["docker_image"]]) + enable_reuse = kwargs.get("enable_reuse", True) + if enable_reuse: + reuse_req, _ = get_feature(self, "http://arvados.org/cwl#ReuseRequirement") + if reuse_req: + enable_reuse = reuse_req["enableReuse"] + 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, "script": "crunchrunner", "repository": "arvados", "script_version": "master", - "minimum_script_version": "9e5b98e8f5f4727856b53447191f9c06e3da2ba6", + "minimum_script_version": crunchrunner_git_commit, "script_parameters": {"tasks": [script_parameters]}, "runtime_constraints": runtime_constraints }, filters=filters, - find_or_create=kwargs.get("enable_reuse", True) + find_or_create=enable_reuse ).execute(num_retries=self.arvrunner.num_retries) self.arvrunner.processes[response["uuid"]] = self self.update_pipeline_component(response) - 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): + if response["state"] == "Complete": + logger.info("%s reused job %s", self.arvrunner.label(self), response["uuid"]) + # Give read permission to the desired project on reused jobs + if response["owner_uuid"] != self.arvrunner.project_uuid: + try: + self.arvrunner.api.links().create(body={ + 'link_class': 'permission', + 'name': 'can_read', + 'tail_uuid': self.arvrunner.project_uuid, + 'head_uuid': response["uuid"], + }).execute(num_retries=self.arvrunner.num_retries) + except ApiError as e: + # The user might not have "manage" access on the job: log + # a message and continue. + logger.info("Creating read permission on job %s: %s", + response["uuid"], + e) + + with Perf(metrics, "done %s" % self.name): self.done(response) + else: + logger.info("%s %s is %s", self.arvrunner.label(self), response["uuid"], response["state"]) except Exception as e: - logger.error("Got error %s" % str(e)) + logger.exception("%s error" % (self.arvrunner.label(self))) self.output_callback({}, "permanentFail") 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"] @@ -155,9 +216,13 @@ 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]) + dirs = {} tmpdir = None outdir = None keepdir = None @@ -169,30 +234,34 @@ class ArvadosJob(object): # 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) + g = crunchrunner_re.match(l) if g: - outdir = g.group(1) - g = keepre.match(l) - if g: - keepdir = g.group(1) + dirs[g.group(1)] = g.group(2) + + if processStatus == "permanentFail": + done.logtail(logc, logger, "%s error log:" % self.arvrunner.label(self)) - with Perf(logger, "output collection %s" % self.name): - outputs = done.done(self, record, tmpdir, outdir, keepdir) + with Perf(metrics, "output collection %s" % self.name): + outputs = done.done(self, record, dirs["tmpdir"], + dirs["outdir"], dirs["keep"]) except WorkflowException as e: - logger.error("Error while collecting job outputs:\n%s", e, exc_info=(e if self.arvrunner.debug else False)) + logger.error("%s unable to collect output from %s:\n%s", + self.arvrunner.label(self), record["output"], 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:") + logger.exception("Got unknown exception while collecting output for job %s:", self.name) processStatus = "permanentFail" - self.output_callback(outputs, processStatus) + # Note: Currently, on error output_callback is expecting an empty dict, + # anything else will fail. + if not isinstance(outputs, dict): + logger.error("Unexpected output type %s '%s'", type(outputs), outputs) + outputs = {} + processStatus = "permanentFail" finally: - del self.arvrunner.processes[record["uuid"]] - + self.output_callback(outputs, processStatus) + if record["uuid"] in self.arvrunner.processes: + del self.arvrunner.processes[record["uuid"]] class RunnerJob(Runner): """Submit and manage a Crunch job that runs crunch_scripts/cwl-runner.""" @@ -205,43 +274,73 @@ class RunnerJob(Runner): a pipeline template or pipeline instance. """ - workflowmapper = super(RunnerJob, self).arvados_job_spec(dry_run=dry_run, pull_image=pull_image, **kwargs) + if self.tool.tool["id"].startswith("keep:"): + self.job_order["cwl:tool"] = self.tool.tool["id"][5:] + else: + packed = packed_workflow(self.arvrunner, self.tool) + wf_pdh = upload_workflow_collection(self.arvrunner, self.name, packed) + self.job_order["cwl:tool"] = "%s/workflow.cwl#main" % wf_pdh + + adjustDirObjs(self.job_order, trim_listing) + visit_class(self.job_order, ("File", "Directory"), trim_anonymous_location) + visit_class(self.job_order, ("File", "Directory"), remove_redundant_fields) + + if self.output_name: + self.job_order["arv:output_name"] = self.output_name + + if self.output_tags: + self.job_order["arv:output_tags"] = self.output_tags + + self.job_order["arv:enable_reuse"] = self.enable_reuse + + if self.on_error: + self.job_order["arv:on_error"] = self.on_error - self.job_order["cwl:tool"] = workflowmapper.mapper(self.tool.tool["id"])[1] return { "script": "cwl-runner", "script_version": "master", + "minimum_script_version": "570509ab4d2ef93d870fd2b1f2eab178afb1bad9", "repository": "arvados", "script_parameters": self.job_order, "runtime_constraints": { - "docker_image": "arvados/jobs" + "docker_image": arvados_jobs_image(self.arvrunner, self.jobs_image), + "min_ram_mb_per_node": self.submit_runner_ram } } def run(self, *args, **kwargs): job_spec = self.arvados_job_spec(*args, **kwargs) + job_spec.setdefault("owner_uuid", self.arvrunner.project_uuid) - response = self.arvrunner.api.jobs().create( + job = self.arvrunner.api.jobs().create( body=job_spec, find_or_create=self.enable_reuse ).execute(num_retries=self.arvrunner.num_retries) - self.uuid = response["uuid"] + for k,v in job_spec["script_parameters"].items(): + if v is False or v is None or isinstance(v, dict): + job_spec["script_parameters"][k] = {"value": v} + + del job_spec["owner_uuid"] + job_spec["job"] = job + self.arvrunner.pipeline = self.arvrunner.api.pipeline_instances().create( + body={ + "owner_uuid": self.arvrunner.project_uuid, + "name": self.name, + "components": {"cwl-runner": job_spec }, + "state": "RunningOnServer"}).execute(num_retries=self.arvrunner.num_retries) + logger.info("Created pipeline %s", self.arvrunner.pipeline["uuid"]) + + if kwargs.get("wait") is False: + self.uuid = self.arvrunner.pipeline["uuid"] + return + + self.uuid = job["uuid"] self.arvrunner.processes[self.uuid] = self - logger.info("Submitted job %s", response["uuid"]) - - if kwargs.get("submit"): - self.arvrunner.pipeline = self.arvrunner.api.pipeline_instances().create( - body={ - "owner_uuid": self.arvrunner.project_uuid, - "name": shortname(self.tool.tool["id"]), - "components": {"cwl-runner": {"job": {"uuid": self.uuid, "state": response["state"]} } }, - "state": "RunningOnClient"}).execute(num_retries=self.arvrunner.num_retries) - - if response["state"] in ("Complete", "Failed", "Cancelled"): - self.done(response) + if job["state"] in ("Complete", "Failed", "Cancelled"): + self.done(job) class RunnerTemplate(object): @@ -250,19 +349,26 @@ class RunnerTemplate(object): type_to_dataclass = { 'boolean': 'boolean', 'File': 'File', + 'Directory': 'Collection', 'float': 'number', 'int': 'number', 'string': 'text', } - def __init__(self, runner, tool, job_order, enable_reuse): + def __init__(self, runner, tool, job_order, enable_reuse, uuid, + submit_runner_ram=0, name=None): self.runner = runner self.tool = tool self.job = RunnerJob( runner=runner, tool=tool, job_order=job_order, - enable_reuse=enable_reuse) + enable_reuse=enable_reuse, + output_name=None, + output_tags=None, + submit_runner_ram=submit_runner_ram, + name=name) + self.uuid = uuid def pipeline_component_spec(self): """Return a component that Workbench and a-r-p-i will understand. @@ -270,6 +376,7 @@ class RunnerTemplate(object): Specifically, translate CWL input specs to Arvados pipeline format, like {"dataclass":"File","value":"xyz"}. """ + spec = self.job.arvados_job_spec() # Most of the component spec is exactly the same as the job @@ -290,10 +397,12 @@ class RunnerTemplate(object): if not isinstance(types, list): types = [types] param['required'] = 'null' not in types - non_null_types = set(types) - set(['null']) + non_null_types = [t for t in types if t != "null"] if len(non_null_types) == 1: the_type = [c for c in non_null_types][0] - dataclass = self.type_to_dataclass.get(the_type) + dataclass = None + if isinstance(the_type, basestring): + dataclass = self.type_to_dataclass.get(the_type) if dataclass: param['dataclass'] = dataclass # Note: If we didn't figure out a single appropriate @@ -316,21 +425,29 @@ class RunnerTemplate(object): pass elif not isinstance(value, dict): param['value'] = value - elif param.get('dataclass') == 'File' and value.get('location'): - param['value'] = value['location'] + elif param.get('dataclass') in ('File', 'Collection') and value.get('location'): + param['value'] = value['location'][5:] spec['script_parameters'][param_id] = param spec['script_parameters']['cwl:tool'] = job_params['cwl:tool'] return spec def save(self): - job_spec = self.pipeline_component_spec() - response = self.runner.api.pipeline_templates().create(body={ + body = { "components": { - self.job.name: job_spec, + self.job.name: self.pipeline_component_spec(), }, "name": self.job.name, - "owner_uuid": self.runner.project_uuid, - }, ensure_unique_name=True).execute(num_retries=self.runner.num_retries) - self.uuid = response["uuid"] - logger.info("Created template %s", self.uuid) + } + if self.runner.project_uuid: + body["owner_uuid"] = self.runner.project_uuid + if self.uuid: + self.runner.api.pipeline_templates().update( + uuid=self.uuid, body=body).execute( + num_retries=self.runner.num_retries) + logger.info("Updated template %s", self.uuid) + else: + self.uuid = self.runner.api.pipeline_templates().create( + body=body, ensure_unique_name=True).execute( + num_retries=self.runner.num_retries)['uuid'] + logger.info("Created template %s", self.uuid)