X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/c60bff923fdd7616eeb1bd8f5998331cbd262970..060d38d627bd1e51dd2b3c6e7de9af6aa7d7b6f3:/sdk/cwl/arvados_cwl/arvcontainer.py diff --git a/sdk/cwl/arvados_cwl/arvcontainer.py b/sdk/cwl/arvados_cwl/arvcontainer.py index 413435db23..4ab65d9d87 100644 --- a/sdk/cwl/arvados_cwl/arvcontainer.py +++ b/sdk/cwl/arvados_cwl/arvcontainer.py @@ -1,18 +1,33 @@ +# Copyright (C) The Arvados Authors. All rights reserved. +# +# SPDX-License-Identifier: Apache-2.0 + import logging import json import os +import urllib +import time +import datetime +import ciso8601 + +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.pathmapper import adjustFileObjs, adjustDirObjs +from cwltool.utils import aslist import arvados.collection from .arvdocker import arv_docker_get_image from . import done -from .runner import Runner +from .runner import Runner, arvados_jobs_image, packed_workflow, trim_anonymous_location +from .fsaccess import CollectionFetcher +from .pathmapper import NoFollowPathMapper, trim_listing +from .perf import Perf logger = logging.getLogger('arvados.cwl-runner') +metrics = logging.getLogger('arvados.cwl-runner.metrics') class ArvadosContainer(object): """Submit and manage a Crunch container request for executing a CWL CommandLineTool.""" @@ -33,44 +48,98 @@ class ArvadosContainer(object): "output_path": self.outdir, "cwd": self.outdir, "priority": 1, - "state": "Committed" + "state": "Committed", + "properties": {}, } runtime_constraints = {} + + resources = self.builder.resources + if resources is not None: + runtime_constraints["vcpus"] = resources.get("cores", 1) + runtime_constraints["ram"] = resources.get("ram") * 2**20 + mounts = { self.outdir: { - "kind": "tmp" + "kind": "tmp", + "capacity": resources.get("outdirSize", 0) * 2**20 + }, + self.tmpdir: { + "kind": "tmp", + "capacity": resources.get("tmpdirSize", 0) * 2**20 } } - - 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("Generate files not supported") + scheduling_parameters = {} + + rf = [self.pathmapper.mapper(f) for f in self.pathmapper.referenced_files] + rf.sort(key=lambda k: k.resolved) + prevdir = None + for resolved, target, tp, stg in rf: + if not stg: + continue + if prevdir and target.startswith(prevdir): + continue + if tp == "Directory": + targetdir = target + else: + targetdir = os.path.dirname(target) + sp = resolved.split("/", 1) + pdh = sp[0][5:] # remove "keep:" + mounts[targetdir] = { + "kind": "collection", + "portable_data_hash": pdh + } + if len(sp) == 2: + if tp == "Directory": + path = sp[1] + else: + path = os.path.dirname(sp[1]) + if path and path != "/": + mounts[targetdir]["path"] = path + prevdir = targetdir + "/" + + 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) + generatemapper = NoFollowPathMapper([self.generatefiles], "", "", + separateDirs=False) + + with Perf(metrics, "createfiles %s" % self.name): + for f, p in generatemapper.items(): + if not p.target: + pass + elif p.type in ("File", "Directory"): + source, path = self.arvrunner.fs_access.get_collection(p.resolved) + vwd.copy(path, p.target, source_collection=source) + elif 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 not p.target: + continue + mountpoint = "%s/%s" % (self.outdir, p.target) + mounts[mountpoint] = {"kind": "collection", + "portable_data_hash": vwd.portable_data_hash(), + "path": p.target} container_request["environment"] = {"TMPDIR": self.tmpdir, "HOME": self.outdir} if self.environment: container_request["environment"].update(self.environment) if self.stdin: - raise UnsupportedRequirement("Stdin redirection currently not suppported") + sp = self.stdin[6:].split("/", 1) + mounts["stdin"] = {"kind": "collection", + "portable_data_hash": sp[0], + "path": sp[1]} if self.stderr: - raise UnsupportedRequirement("Stderr redirection currently not suppported") + mounts["stderr"] = {"kind": "file", + "path": "%s/%s" % (self.outdir, self.stderr)} if self.stdout: mounts["stdout"] = {"kind": "file", @@ -85,33 +154,81 @@ class ArvadosContainer(object): pull_image, self.arvrunner.project_uuid) - resources = self.builder.resources - if resources is not None: - runtime_constraints["vcpus"] = resources.get("cores", 1) - runtime_constraints["ram"] = resources.get("ram") * 2**20 - + 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"] * 2**20 + if "outputDirType" in runtime_req: + if runtime_req["outputDirType"] == "local_output_dir": + # Currently the default behavior. + pass + elif runtime_req["outputDirType"] == "keep_output_dir": + mounts[self.outdir]= { + "kind": "collection", + "writable": True + } + + partition_req, _ = get_feature(self, "http://arvados.org/cwl#PartitionRequirement") + if partition_req: + scheduling_parameters["partitions"] = aslist(partition_req["partition"]) + + intermediate_output_req, _ = get_feature(self, "http://arvados.org/cwl#IntermediateOutput") + if intermediate_output_req: + self.output_ttl = intermediate_output_req["outputTTL"] + else: + self.output_ttl = self.arvrunner.intermediate_output_ttl + + if self.output_ttl < 0: + raise WorkflowError("Invalid value %d for output_ttl, cannot be less than zero" % container_request["output_ttl"]) + + container_request["output_ttl"] = self.output_ttl container_request["mounts"] = mounts container_request["runtime_constraints"] = runtime_constraints + container_request["scheduling_parameters"] = scheduling_parameters + + 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"] + container_request["use_existing"] = enable_reuse + + 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.processes[response["container_uuid"]] = self - - logger.info("Container %s (%s) request state is %s", self.name, response["uuid"], response["state"]) + self.uuid = response["uuid"] + self.arvrunner.processes[self.uuid] = self if response["state"] == "Final": + logger.info("%s reused container %s", self.arvrunner.label(self), response["container_uuid"]) self.done(response) + else: + logger.info("%s %s state is %s", self.arvrunner.label(self), response["uuid"], response["state"]) except Exception as e: - logger.error("Got error %s" % str(e)) + logger.error("%s got error %s" % (self.arvrunner.label(self), str(e))) self.output_callback({}, "permanentFail") def done(self, record): + outputs = {} try: - if record["state"] == "Complete": - rcode = record["exit_code"] + 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: @@ -125,20 +242,37 @@ class ArvadosContainer(object): else: processStatus = "permanentFail" - try: - outputs = {} - if record["output"]: - outputs = done.done(self, record, "/tmp", self.outdir, "/keep") - except WorkflowException as e: - logger.error("Error while collecting container 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" - - self.output_callback(outputs, processStatus) + if processStatus == "permanentFail": + logc = arvados.collection.CollectionReader(container["log"], + api_client=self.arvrunner.api, + keep_client=self.arvrunner.keep_client, + num_retries=self.arvrunner.num_retries) + done.logtail(logc, logger, "%s error log:" % self.arvrunner.label(self)) + + if record["output_uuid"]: + if self.arvrunner.trash_intermediate or self.arvrunner.intermediate_output_ttl: + # Compute the trash time to avoid requesting the collection record. + trash_at = ciso8601.parse_datetime_unaware(record["modified_at"]) + datetime.timedelta(0, self.arvrunner.intermediate_output_ttl) + aftertime = " at %s" % trash_at.strftime("%Y-%m-%d %H:%M:%S UTC") if self.arvrunner.intermediate_output_ttl else "" + orpart = ", or" if self.arvrunner.trash_intermediate and self.arvrunner.intermediate_output_ttl else "" + oncomplete = " upon successful completion of the workflow" if self.arvrunner.trash_intermediate else "" + logger.info("%s Intermediate output %s (%s) will be trashed%s%s%s." % ( + self.arvrunner.label(self), record["output_uuid"], container["output"], aftertime, orpart, oncomplete)) + self.arvrunner.add_intermediate_output(record["output_uuid"]) + + if container["output"]: + outputs = done.done_outputs(self, container, "/tmp", self.outdir, "/keep") + except WorkflowException as e: + logger.error("%s unable to collect output from %s:\n%s", + self.arvrunner.label(self), container["output"], e, exc_info=(e if self.arvrunner.debug else False)) + processStatus = "permanentFail" + except Exception as e: + logger.exception("%s while getting output object: %s", self.arvrunner.label(self), e) + 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 RunnerContainer(Runner): @@ -151,41 +285,22 @@ class RunnerContainer(Runner): the +body+ argument to container_requests().create(). """ - workflowmapper = super(RunnerContainer, self).arvados_job_spec(dry_run=dry_run, pull_image=pull_image, **kwargs) - - with arvados.collection.Collection(api_client=self.arvrunner.api) as jobobj: - with jobobj.open("cwl.input.json", "w") as f: - json.dump(self.job_order, f, sort_keys=True, indent=4) - jobobj.save_new(owner_uuid=self.arvrunner.project_uuid) - - workflowname = os.path.basename(self.tool.tool["id"]) - workflowpath = "/var/lib/cwl/workflow/%s" % workflowname - workflowcollection = workflowmapper.mapper(self.tool.tool["id"])[1] - workflowcollection = workflowcollection[5:workflowcollection.index('/')] - jobpath = "/var/lib/cwl/job/cwl.input.json" + adjustDirObjs(self.job_order, trim_listing) + adjustFileObjs(self.job_order, trim_anonymous_location) + adjustDirObjs(self.job_order, trim_anonymous_location) - container_image = arv_docker_get_image(self.arvrunner.api, - {"dockerImageId": "arvados/jobs"}, - pull_image, - self.arvrunner.project_uuid) - - return { - "command": ["arvados-cwl-runner", "--local", "--api=containers", workflowpath, jobpath], + 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": container_image, + "container_image": arvados_jobs_image(self.arvrunner, self.jobs_image), "mounts": { - "/var/lib/cwl/workflow": { - "kind": "collection", - "portable_data_hash": "%s" % workflowcollection - }, - jobpath: { - "kind": "collection", - "portable_data_hash": "%s/cwl.input.json" % jobobj.portable_data_hash() + "/var/lib/cwl/cwl.input.json": { + "kind": "json", + "content": self.job_order }, "stdout": { "kind": "file", @@ -198,11 +313,63 @@ class RunnerContainer(Runner): }, "runtime_constraints": { "vcpus": 1, - "ram": 1024*1024*256, + "ram": 1024*1024 * self.submit_runner_ram, "API": True - } + }, + "properties": {} } + if self.tool.tool.get("id", "").startswith("keep:"): + sp = self.tool.tool["id"].split('/') + workflowcollection = sp[0][5:] + workflowname = "/".join(sp[1:]) + workflowpath = "/var/lib/cwl/workflow/%s" % workflowname + container_req["mounts"]["/var/lib/cwl/workflow"] = { + "kind": "collection", + "portable_data_hash": "%s" % workflowcollection + } + else: + packed = packed_workflow(self.arvrunner, self.tool) + workflowpath = "/var/lib/cwl/workflow.json#main" + container_req["mounts"]["/var/lib/cwl/workflow.json"] = { + "kind": "json", + "content": packed + } + if self.tool.tool.get("id", "").startswith("arvwf:"): + container_req["properties"]["template_uuid"] = self.tool.tool["id"][6:33] + + command = ["arvados-cwl-runner", "--local", "--api=containers", "--no-log-timestamps"] + if self.output_name: + command.append("--output-name=" + self.output_name) + container_req["output_name"] = self.output_name + + if self.output_tags: + command.append("--output-tags=" + self.output_tags) + + if kwargs.get("debug"): + command.append("--debug") + + if self.enable_reuse: + command.append("--enable-reuse") + else: + command.append("--disable-reuse") + + if self.on_error: + command.append("--on-error=" + self.on_error) + + if self.intermediate_output_ttl: + command.append("--intermediate-output-ttl=%d" % self.intermediate_output_ttl) + + if self.arvrunner.trash_intermediate: + command.append("--trash-intermediate") + + 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) @@ -213,9 +380,23 @@ class RunnerContainer(Runner): ).execute(num_retries=self.arvrunner.num_retries) self.uuid = response["uuid"] - self.arvrunner.processes[response["container_uuid"]] = self + self.arvrunner.processes[self.uuid] = self - logger.info("Submitted container %s", response["uuid"]) + logger.info("%s submitted container %s", self.arvrunner.label(self), response["uuid"]) - if response["state"] in ("Complete", "Failed", "Cancelled"): + 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("%s while getting runner container: %s", self.arvrunner.label(self), e) + self.arvrunner.output_callback({}, "permanentFail") + else: + super(RunnerContainer, self).done(container) + finally: + if record["uuid"] in self.arvrunner.processes: + del self.arvrunner.processes[record["uuid"]]