X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/ffff84bc7c76e3c6581a10559dcfc193768e7b58..3af6db5dc4e2f08b2ebb49a82109c4325ad7fcc4:/sdk/cwl/arvados_cwl/__init__.py diff --git a/sdk/cwl/arvados_cwl/__init__.py b/sdk/cwl/arvados_cwl/__init__.py index 99db9f1026..92be92d6e0 100644 --- a/sdk/cwl/arvados_cwl/__init__.py +++ b/sdk/cwl/arvados_cwl/__init__.py @@ -9,35 +9,47 @@ import os import sys import threading import hashlib +import copy +import json +from functools import partial import pkg_resources # part of setuptools from cwltool.errors import WorkflowException import cwltool.main import cwltool.workflow +import schema_salad import arvados -import arvados.events import arvados.config from .arvcontainer import ArvadosContainer, RunnerContainer from .arvjob import ArvadosJob, RunnerJob, RunnerTemplate +from. runner import Runner, upload_instance from .arvtool import ArvadosCommandTool +from .arvworkflow import ArvadosWorkflow, upload_workflow from .fsaccess import CollectionFsAccess - -from cwltool.process import shortname, UnsupportedRequirement -from cwltool.pathmapper import adjustFileObjs +from .perf import Perf +from .pathmapper import FinalOutputPathMapper +from ._version import __version__ + +from cwltool.pack import pack +from cwltool.process import shortname, UnsupportedRequirement, getListing +from cwltool.pathmapper import adjustFileObjs, adjustDirObjs +from cwltool.draft2tool import compute_checksums from arvados.api import OrderedJsonModel logger = logging.getLogger('arvados.cwl-runner') +metrics = logging.getLogger('arvados.cwl-runner.metrics') logger.setLevel(logging.INFO) + class ArvCwlRunner(object): """Execute a CWL tool or workflow, submit work (using either jobs or containers API), wait for them to complete, and report output. """ - def __init__(self, api_client, work_api=None): + def __init__(self, api_client, work_api=None, keep_client=None, output_name=None): self.api = api_client self.processes = {} self.lock = threading.Lock() @@ -47,18 +59,39 @@ class ArvCwlRunner(object): self.uploaded = {} self.num_retries = 4 self.uuid = None - self.work_api = work_api - - if self.work_api is None: - # todo: autodetect API to use. - self.work_api = "jobs" + self.stop_polling = threading.Event() + self.poll_api = None + self.pipeline = None + self.final_output_collection = None + self.output_name = output_name + self.project_uuid = None - if self.work_api not in ("containers", "jobs"): - raise Exception("Unsupported API '%s'" % self.work_api) + if keep_client is not None: + self.keep_client = keep_client + else: + self.keep_client = arvados.keep.KeepClient(api_client=self.api, num_retries=self.num_retries) + + for api in ["jobs", "containers"]: + try: + methods = self.api._rootDesc.get('resources')[api]['methods'] + if ('httpMethod' in methods['create'] and + (work_api == api or work_api is None)): + self.work_api = api + break + except KeyError: + pass + if not self.work_api: + if work_api is None: + raise Exception("No supported APIs") + else: + raise Exception("Unsupported API '%s'" % work_api) - def arvMakeTool(self, toolpath_object, **kwargs): + def arv_make_tool(self, toolpath_object, **kwargs): + kwargs["work_api"] = self.work_api if "class" in toolpath_object and toolpath_object["class"] == "CommandLineTool": - return ArvadosCommandTool(self, toolpath_object, work_api=self.work_api, **kwargs) + return ArvadosCommandTool(self, toolpath_object, **kwargs) + elif "class" in toolpath_object and toolpath_object["class"] == "Workflow": + return ArvadosWorkflow(self, toolpath_object, **kwargs) else: return cwltool.workflow.defaultMakeTool(toolpath_object, **kwargs) @@ -91,29 +124,168 @@ class ArvCwlRunner(object): try: self.cond.acquire() j = self.processes[uuid] - logger.info("Job %s (%s) is %s", j.name, uuid, event["properties"]["new_attributes"]["state"]) - j.done(event["properties"]["new_attributes"]) + txt = self.work_api[0].upper() + self.work_api[1:-1] + logger.info("%s %s (%s) is %s", txt, j.name, uuid, event["properties"]["new_attributes"]["state"]) + with Perf(metrics, "done %s" % j.name): + j.done(event["properties"]["new_attributes"]) self.cond.notify() finally: self.cond.release() + def poll_states(self): + """Poll status of jobs or containers listed in the processes dict. + + Runs in a separate thread. + """ + + while True: + self.stop_polling.wait(15) + if self.stop_polling.is_set(): + break + with self.lock: + keys = self.processes.keys() + if not keys: + continue + + if self.work_api == "containers": + table = self.poll_api.containers() + elif self.work_api == "jobs": + table = self.poll_api.jobs() + + try: + proc_states = table.list(filters=[["uuid", "in", keys]]).execute(num_retries=self.num_retries) + except Exception as e: + logger.warn("Error checking states on API server: %s", e) + continue + + for p in proc_states["items"]: + self.on_message({ + "object_uuid": p["uuid"], + "event_type": "update", + "properties": { + "new_attributes": p + } + }) + def get_uploaded(self): return self.uploaded.copy() def add_uploaded(self, src, pair): self.uploaded[src] = pair - def arvExecutor(self, tool, job_order, **kwargs): + def check_writable(self, obj): + if isinstance(obj, dict): + if obj.get("writable"): + raise UnsupportedRequirement("InitialWorkDir feature 'writable: true' not supported") + for v in obj.itervalues(): + self.check_writable(v) + if isinstance(obj, list): + for v in obj: + self.check_writable(v) + + def make_output_collection(self, name, outputObj): + outputObj = copy.deepcopy(outputObj) + + files = [] + def capture(fileobj): + files.append(fileobj) + + adjustDirObjs(outputObj, capture) + adjustFileObjs(outputObj, capture) + + generatemapper = FinalOutputPathMapper(files, "", "", separateDirs=False) + + final = arvados.collection.Collection(api_client=self.api, + keep_client=self.keep_client, + num_retries=self.num_retries) + + srccollections = {} + for k,v in generatemapper.items(): + if k.startswith("_:"): + if v.type == "Directory": + continue + if v.type == "CreateFile": + with final.open(v.target, "wb") as f: + f.write(v.resolved.encode("utf-8")) + continue + + if not k.startswith("keep:"): + raise Exception("Output source is not in keep or a literal") + sp = k.split("/") + srccollection = sp[0][5:] + if srccollection not in srccollections: + try: + srccollections[srccollection] = arvados.collection.CollectionReader( + srccollection, + api_client=self.api, + keep_client=self.keep_client, + num_retries=self.num_retries) + except arvados.errors.ArgumentError as e: + logger.error("Creating CollectionReader for '%s' '%s': %s", k, v, e) + raise + reader = srccollections[srccollection] + try: + srcpath = "/".join(sp[1:]) if len(sp) > 1 else "." + final.copy(srcpath, v.target, source_collection=reader, overwrite=False) + except IOError as e: + logger.warn("While preparing output collection: %s", e) + + def rewrite(fileobj): + fileobj["location"] = generatemapper.mapper(fileobj["location"]).target + for k in ("basename", "listing", "contents"): + if k in fileobj: + del fileobj[k] + + adjustDirObjs(outputObj, rewrite) + adjustFileObjs(outputObj, rewrite) + + with final.open("cwl.output.json", "w") as f: + json.dump(outputObj, f, sort_keys=True, indent=4, separators=(',',': ')) + + final.save_new(name=name, owner_uuid=self.project_uuid, ensure_unique_name=True) + + logger.info("Final output collection %s \"%s\" (%s)", final.portable_data_hash(), + final.api_response()["name"], + final.manifest_locator()) + + def finalcollection(fileobj): + fileobj["location"] = "keep:%s/%s" % (final.portable_data_hash(), fileobj["location"]) + + adjustDirObjs(outputObj, finalcollection) + adjustFileObjs(outputObj, finalcollection) + + return (outputObj, final) + + def set_crunch_output(self): + if self.work_api == "containers": + try: + current = self.api.containers().current().execute(num_retries=self.num_retries) + self.api.containers().update(uuid=current['uuid'], + body={ + 'output': self.final_output_collection.portable_data_hash(), + }).execute(num_retries=self.num_retries) + except Exception as e: + logger.info("Setting container output: %s", e) + elif self.work_api == "jobs" and "TASK_UUID" in os.environ: + self.api.job_tasks().update(uuid=os.environ["TASK_UUID"], + body={ + 'output': self.final_output_collection.portable_data_hash(), + 'success': self.final_status == "success", + 'progress':1.0 + }).execute(num_retries=self.num_retries) + + def arv_executor(self, tool, job_order, **kwargs): self.debug = kwargs.get("debug") - if kwargs.get("quiet"): - logger.setLevel(logging.WARN) - logging.getLogger('arvados.arv-run').setLevel(logging.WARN) + tool.visit(self.check_writable) useruuid = self.api.users().current().execute()["uuid"] self.project_uuid = kwargs.get("project_uuid") if kwargs.get("project_uuid") else useruuid self.pipeline = None - self.fs_access = CollectionFsAccess(kwargs["basedir"], api_client=self.api) + make_fs_access = kwargs.get("make_fs_access") or partial(CollectionFsAccess, + api_client=self.api, + keep_client=self.keep_client) + self.fs_access = make_fs_access(kwargs["basedir"]) if kwargs.get("create_template"): tmpl = RunnerTemplate(self, tool, job_order, kwargs.get("enable_reuse")) @@ -121,10 +293,12 @@ class ArvCwlRunner(object): # cwltool.main will write our return value to stdout. return tmpl.uuid - self.debug = kwargs.get("debug") + if kwargs.get("create_workflow") or kwargs.get("update_workflow"): + return upload_workflow(self, tool, job_order, self.project_uuid, kwargs.get("update_workflow")) + self.ignore_docker_for_reuse = kwargs.get("ignore_docker_for_reuse") - kwargs["fs_access"] = self.fs_access + kwargs["make_fs_access"] = make_fs_access kwargs["enable_reuse"] = kwargs.get("enable_reuse") kwargs["use_container"] = True kwargs["tmpdir_prefix"] = "tmp" @@ -135,11 +309,14 @@ class ArvCwlRunner(object): kwargs["outdir"] = "/var/spool/cwl" kwargs["docker_outdir"] = "/var/spool/cwl" kwargs["tmpdir"] = "/tmp" + kwargs["docker_tmpdir"] = "/tmp" elif self.work_api == "jobs": kwargs["outdir"] = "$(task.outdir)" kwargs["docker_outdir"] = "$(task.outdir)" kwargs["tmpdir"] = "$(task.tmpdir)" + upload_instance(self, shortname(tool.tool["id"]), tool, job_order) + runnerjob = None if kwargs.get("submit"): if self.work_api == "containers": @@ -148,9 +325,9 @@ class ArvCwlRunner(object): self.output_callback, **kwargs).next() else: - runnerjob = RunnerContainer(self, tool, job_order, kwargs.get("enable_reuse")) + runnerjob = RunnerContainer(self, tool, job_order, kwargs.get("enable_reuse"), self.output_name) else: - runnerjob = RunnerJob(self, tool, job_order, kwargs.get("enable_reuse")) + runnerjob = RunnerJob(self, tool, job_order, kwargs.get("enable_reuse"), self.output_name) if not kwargs.get("submit") and "cwl_runner_job" not in kwargs and not self.work_api == "containers": # Create pipeline for local run @@ -163,15 +340,12 @@ class ArvCwlRunner(object): logger.info("Pipeline instance %s", self.pipeline["uuid"]) if runnerjob and not kwargs.get("wait"): - runnerjob.run() + runnerjob.run(wait=kwargs.get("wait")) return runnerjob.uuid - arvados.config.settings()["ARVADOS_DISABLE_WEBSOCKETS"] = "1" - - if self.work_api == "containers": - events = arvados.events.subscribe(arvados.api('v1'), [["object_uuid", "is_a", "arvados#container"]], self.on_message) - if self.work_api == "jobs": - events = arvados.events.subscribe(arvados.api('v1'), [["object_uuid", "is_a", "arvados#job"]], self.on_message) + self.poll_api = arvados.api('v1') + self.polling_thread = threading.Thread(target=self.poll_states) + self.polling_thread.start() if runnerjob: jobiter = iter((runnerjob,)) @@ -188,20 +362,25 @@ class ArvCwlRunner(object): # except when in cond.wait(), at which point on_message can update # job state and process output callbacks. + loopperf = Perf(metrics, "jobiter") + loopperf.__enter__() for runnable in jobiter: + loopperf.__exit__() if runnable: - runnable.run(**kwargs) + with Perf(metrics, "run"): + runnable.run(**kwargs) else: if self.processes: self.cond.wait(1) else: logger.error("Workflow is deadlocked, no runnable jobs and not waiting on any pending jobs.") break + loopperf.__enter__() + loopperf.__exit__() while self.processes: self.cond.wait(1) - events.close() except UnsupportedRequirement: raise except: @@ -217,28 +396,29 @@ class ArvCwlRunner(object): body={"priority": "0"}).execute(num_retries=self.num_retries) finally: self.cond.release() + self.stop_polling.set() + self.polling_thread.join() if self.final_status == "UnsupportedRequirement": raise UnsupportedRequirement("Check log for details.") - if self.final_status != "success": - raise WorkflowException("Workflow failed.") - if self.final_output is None: raise WorkflowException("Workflow did not return a result.") + if kwargs.get("submit") and isinstance(runnerjob, Runner): + logger.info("Final output collection %s", runnerjob.final_output) + else: + if self.output_name is None: + self.output_name = "Output of %s" % (shortname(tool.tool["id"])) + self.final_output, self.final_output_collection = self.make_output_collection(self.output_name, self.final_output) + self.set_crunch_output() + + if self.final_status != "success": + raise WorkflowException("Workflow failed.") + if kwargs.get("compute_checksum"): - def compute_checksums(fileobj): - if "checksum" not in fileobj: - checksum = hashlib.sha1() - with self.fs_access.open(fileobj["location"], "rb") as f: - contents = f.read(1024*1024) - while contents != "": - checksum.update(contents) - contents = f.read(1024*1024) - fileobj["checksum"] = "sha1$%s" % checksum.hexdigest() - - adjustFileObjs(self.final_output, compute_checksums) + adjustDirObjs(self.final_output, partial(getListing, self.fs_access)) + adjustFileObjs(self.final_output, partial(compute_checksums, self.fs_access)) return self.final_output @@ -250,7 +430,7 @@ def versionstring(): arvpkg = pkg_resources.require("arvados-python-client") cwlpkg = pkg_resources.require("cwltool") - return "%s %s, %s %s, %s %s" % (sys.argv[0], arvcwlpkg[0].version, + return "%s %s %s, %s %s, %s %s" % (sys.argv[0], __version__, arvcwlpkg[0].version, "arvados-python-client", arvpkg[0].version, "cwltool", cwlpkg[0].version) @@ -274,6 +454,8 @@ def arg_parser(): # type: () -> argparse.ArgumentParser exgroup.add_argument("--quiet", action="store_true", help="Only print warnings and errors.") exgroup.add_argument("--debug", action="store_true", help="Print even more logging") + parser.add_argument("--metrics", action="store_true", help="Print timing metrics") + parser.add_argument("--tool-help", action="store_true", help="Print command line help for tool") exgroup = parser.add_mutually_exclusive_group() @@ -284,7 +466,8 @@ def arg_parser(): # type: () -> argparse.ArgumentParser default=True, dest="enable_reuse", help="") - parser.add_argument("--project-uuid", type=str, help="Project that will own the workflow jobs, if not provided, will go to home project.") + parser.add_argument("--project-uuid", type=str, metavar="UUID", help="Project that will own the workflow jobs, if not provided, will go to home project.") + parser.add_argument("--output-name", type=str, help="Name to use for collection that stores the final output.", default=None) parser.add_argument("--ignore-docker-for-reuse", action="store_true", help="Ignore Docker image version when deciding whether to reuse past jobs.", default=False) @@ -295,6 +478,8 @@ def arg_parser(): # type: () -> argparse.ArgumentParser exgroup.add_argument("--local", action="store_false", help="Run workflow on local host (submits jobs to Arvados).", default=True, dest="submit") exgroup.add_argument("--create-template", action="store_true", help="Create an Arvados pipeline template.") + exgroup.add_argument("--create-workflow", action="store_true", help="Create an Arvados workflow.") + exgroup.add_argument("--update-workflow", type=str, metavar="UUID", help="Update existing Arvados workflow with uuid.") exgroup = parser.add_mutually_exclusive_group() exgroup.add_argument("--wait", action="store_true", help="After submitting workflow runner job, wait for completion.", @@ -304,7 +489,7 @@ def arg_parser(): # type: () -> argparse.ArgumentParser parser.add_argument("--api", type=str, default=None, dest="work_api", - help="Select work submission API, one of 'jobs' or 'containers'.") + help="Select work submission API, one of 'jobs' or 'containers'. Default is 'jobs' if that API is available, otherwise 'containers'.") parser.add_argument("--compute-checksum", action="store_true", default=False, help="Compute checksum of contents while collecting outputs", @@ -315,30 +500,55 @@ def arg_parser(): # type: () -> argparse.ArgumentParser return parser - -def main(args, stdout, stderr, api_client=None): +def add_arv_hints(): + cache = {} + res = pkg_resources.resource_stream(__name__, 'arv-cwl-schema.yml') + cache["http://arvados.org/cwl"] = res.read() + res.close() + document_loader, cwlnames, _, _ = cwltool.process.get_schema("v1.0") + _, extnames, _, _ = schema_salad.schema.load_schema("http://arvados.org/cwl", cache=cache) + for n in extnames.names: + if not cwlnames.has_name("http://arvados.org/cwl#"+n, ""): + cwlnames.add_name("http://arvados.org/cwl#"+n, "", extnames.get_name(n, "")) + document_loader.idx["http://arvados.org/cwl#"+n] = {} + +def main(args, stdout, stderr, api_client=None, keep_client=None): parser = arg_parser() job_order_object = None arvargs = parser.parse_args(args) - if arvargs.create_template and not arvargs.job_order: + if (arvargs.create_template or arvargs.create_workflow or arvargs.update_workflow) and not arvargs.job_order: job_order_object = ({}, "") + add_arv_hints() + try: if api_client is None: api_client=arvados.api('v1', model=OrderedJsonModel()) - runner = ArvCwlRunner(api_client, work_api=arvargs.work_api) + runner = ArvCwlRunner(api_client, work_api=arvargs.work_api, keep_client=keep_client, output_name=arvargs.output_name) except Exception as e: logger.error(e) return 1 + if arvargs.debug: + logger.setLevel(logging.DEBUG) + + if arvargs.quiet: + logger.setLevel(logging.WARN) + logging.getLogger('arvados.arv-run').setLevel(logging.WARN) + + if arvargs.metrics: + metrics.setLevel(logging.DEBUG) + logging.getLogger("cwltool.metrics").setLevel(logging.DEBUG) + arvargs.conformance_test = None arvargs.use_container = True return cwltool.main.main(args=arvargs, stdout=stdout, stderr=stderr, - executor=runner.arvExecutor, - makeTool=runner.arvMakeTool, + executor=runner.arv_executor, + makeTool=runner.arv_make_tool, versionfunc=versionstring, - job_order_object=job_order_object) + job_order_object=job_order_object, + make_fs_access=partial(CollectionFsAccess, api_client=api_client))