X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/bb48c3155b90e4bf1994a66b8fab858f12a1f4f4..86af9b2ebc9240e9041870786e87021c94f11eea:/sdk/cwl/arvados_cwl/__init__.py diff --git a/sdk/cwl/arvados_cwl/__init__.py b/sdk/cwl/arvados_cwl/__init__.py index 9b92e6ffe7..80734d2f7c 100644 --- a/sdk/cwl/arvados_cwl/__init__.py +++ b/sdk/cwl/arvados_cwl/__init__.py @@ -1,12 +1,15 @@ #!/usr/bin/env python -# Implement cwl-runner interface for submitting and running jobs on Arvados. +# Implement cwl-runner interface for submitting and running work on Arvados, using +# either the Crunch jobs API or Crunch containers API. import argparse import logging import os import sys import threading +import hashlib +from functools import partial import pkg_resources # part of setuptools from cwltool.errors import WorkflowException @@ -14,26 +17,31 @@ import cwltool.main import cwltool.workflow import arvados -import arvados.events +import arvados.config from .arvcontainer import ArvadosContainer, RunnerContainer from .arvjob import ArvadosJob, RunnerJob, RunnerTemplate from .arvtool import ArvadosCommandTool from .fsaccess import CollectionFsAccess +from .arvworkflow import make_workflow from cwltool.process import shortname, UnsupportedRequirement +from cwltool.pathmapper import adjustFileObjs +from cwltool.draft2tool import compute_checksums from arvados.api import OrderedJsonModel logger = logging.getLogger('arvados.cwl-runner') logger.setLevel(logging.INFO) class ArvCwlRunner(object): - """Execute a CWL tool or workflow, submit crunch jobs, wait for them to - complete, and report output.""" + """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, crunch2=False): + """ + + def __init__(self, api_client, work_api=None): self.api = api_client - self.jobs = {} + self.processes = {} self.lock = threading.Lock() self.cond = threading.Condition(self.lock) self.final_output = None @@ -41,22 +49,31 @@ class ArvCwlRunner(object): self.uploaded = {} self.num_retries = 4 self.uuid = None - self.crunch2 = crunch2 + self.work_api = work_api + self.stop_polling = threading.Event() + self.poll_api = None + + if self.work_api is None: + # todo: autodetect API to use. + self.work_api = "jobs" + + if self.work_api not in ("containers", "jobs"): + raise Exception("Unsupported API '%s'" % self.work_api) def arvMakeTool(self, toolpath_object, **kwargs): if "class" in toolpath_object and toolpath_object["class"] == "CommandLineTool": - return ArvadosCommandTool(self, toolpath_object, crunch2=self.crunch2, **kwargs) + return ArvadosCommandTool(self, toolpath_object, work_api=self.work_api, **kwargs) else: return cwltool.workflow.defaultMakeTool(toolpath_object, **kwargs) def output_callback(self, out, processStatus): if processStatus == "success": - logger.info("Overall job status is %s", processStatus) + logger.info("Overall process status is %s", processStatus) if self.pipeline: self.api.pipeline_instances().update(uuid=self.pipeline["uuid"], body={"state": "Complete"}).execute(num_retries=self.num_retries) else: - logger.warn("Overall job status is %s", processStatus) + logger.warn("Overall process status is %s", processStatus) if self.pipeline: self.api.pipeline_instances().update(uuid=self.pipeline["uuid"], body={"state": "Failed"}).execute(num_retries=self.num_retries) @@ -65,11 +82,11 @@ class ArvCwlRunner(object): def on_message(self, event): if "object_uuid" in event: - if event["object_uuid"] in self.jobs and event["event_type"] == "update": - if event["properties"]["new_attributes"]["state"] == "Running" and self.jobs[event["object_uuid"]].running is False: + if event["object_uuid"] in self.processes and event["event_type"] == "update": + if event["properties"]["new_attributes"]["state"] == "Running" and self.processes[event["object_uuid"]].running is False: uuid = event["object_uuid"] with self.lock: - j = self.jobs[uuid] + j = self.processes[uuid] logger.info("Job %s (%s) is Running", j.name, uuid) j.running = True j.update_pipeline_component(event["properties"]["new_attributes"]) @@ -77,22 +94,69 @@ class ArvCwlRunner(object): uuid = event["object_uuid"] try: self.cond.acquire() - j = self.jobs[uuid] + 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"]) 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 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 arvExecutor(self, tool, job_order, **kwargs): self.debug = kwargs.get("debug") + tool.visit(self.check_writable) + if kwargs.get("quiet"): logger.setLevel(logging.WARN) logging.getLogger('arvados.arv-run').setLevel(logging.WARN) @@ -100,6 +164,8 @@ class ArvCwlRunner(object): useruuid = self.api.users().current().execute()["uuid"] self.project_uuid = kwargs.get("project_uuid") if kwargs.get("project_uuid") else useruuid self.pipeline = None + make_fs_access = kwargs.get("make_fs_access") or partial(CollectionFsAccess, api_client=self.api) + self.fs_access = make_fs_access(kwargs["basedir"]) if kwargs.get("create_template"): tmpl = RunnerTemplate(self, tool, job_order, kwargs.get("enable_reuse")) @@ -107,23 +173,32 @@ class ArvCwlRunner(object): # cwltool.main will write our return value to stdout. return tmpl.uuid + if kwargs.get("create_workflow") or kwargs.get("update_workflow"): + return make_workflow(self, tool, job_order, self.project_uuid, kwargs.get("update_workflow")) + self.debug = kwargs.get("debug") self.ignore_docker_for_reuse = kwargs.get("ignore_docker_for_reuse") - self.fs_access = CollectionFsAccess(kwargs["basedir"]) - 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" + kwargs["on_error"] = "continue" + kwargs["compute_checksum"] = kwargs.get("compute_checksum") - if self.crunch2: + if self.work_api == "containers": kwargs["outdir"] = "/var/spool/cwl" + kwargs["docker_outdir"] = "/var/spool/cwl" kwargs["tmpdir"] = "/tmp" - else: + kwargs["docker_tmpdir"] = "/tmp" + elif self.work_api == "jobs": kwargs["outdir"] = "$(task.outdir)" + kwargs["docker_outdir"] = "$(task.outdir)" kwargs["tmpdir"] = "$(task.tmpdir)" runnerjob = None if kwargs.get("submit"): - if self.crunch2: + if self.work_api == "containers": if tool.tool["class"] == "CommandLineTool": runnerjob = tool.job(job_order, self.output_callback, @@ -133,7 +208,7 @@ class ArvCwlRunner(object): else: runnerjob = RunnerJob(self, tool, job_order, kwargs.get("enable_reuse")) - if not kwargs.get("submit") and "cwl_runner_job" not in kwargs and not self.crunch2: + if not kwargs.get("submit") and "cwl_runner_job" not in kwargs and not self.work_api == "containers": # Create pipeline for local run self.pipeline = self.api.pipeline_instances().create( body={ @@ -147,10 +222,9 @@ class ArvCwlRunner(object): runnerjob.run() return runnerjob.uuid - if self.crunch2: - events = arvados.events.subscribe(arvados.api('v1'), [["object_uuid", "is_a", "arvados#container"]], self.on_message) - else: - events = arvados.events.subscribe(arvados.api('v1'), [["object_uuid", "is_a", "arvados#job"]], self.on_message) + self.poll_api = arvados.api('v1') + self.polling_thread = threading.Thread(target=self.poll_states) + self.polling_thread.start() if runnerjob: jobiter = iter((runnerjob,)) @@ -159,7 +233,6 @@ class ArvCwlRunner(object): self.uuid = kwargs.get("cwl_runner_job").get('uuid') jobiter = tool.job(job_order, self.output_callback, - docker_outdir="$(task.outdir)", **kwargs) try: @@ -172,16 +245,15 @@ class ArvCwlRunner(object): if runnable: runnable.run(**kwargs) else: - if self.jobs: + if self.processes: self.cond.wait(1) else: logger.error("Workflow is deadlocked, no runnable jobs and not waiting on any pending jobs.") break - while self.jobs: + while self.processes: self.cond.wait(1) - events.close() except UnsupportedRequirement: raise except: @@ -192,17 +264,25 @@ class ArvCwlRunner(object): if self.pipeline: self.api.pipeline_instances().update(uuid=self.pipeline["uuid"], body={"state": "Failed"}).execute(num_retries=self.num_retries) - if runnerjob and runnerjob.uuid and self.crunch2: + if runnerjob and runnerjob.uuid and self.work_api == "containers": self.api.container_requests().update(uuid=runnerjob.uuid, 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 cwltool.workflow.WorkflowException("Workflow did not return a result.") + raise WorkflowException("Workflow did not return a result.") + + if kwargs.get("compute_checksum"): + adjustFileObjs(self.final_output, partial(compute_checksums, self.fs_access)) return self.final_output @@ -222,7 +302,6 @@ def versionstring(): def arg_parser(): # type: () -> argparse.ArgumentParser parser = argparse.ArgumentParser(description='Arvados executor for Common Workflow Language') - parser.add_argument("--conformance-test", action="store_true") parser.add_argument("--basedir", type=str, help="Base directory used to resolve relative references in the input, default to directory of input object file or current directory (if inputs piped/provided on command line).") parser.add_argument("--outdir", type=str, default=os.path.abspath('.'), @@ -260,6 +339,8 @@ 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, help="Update Arvados workflow.") exgroup = parser.add_mutually_exclusive_group() exgroup.add_argument("--wait", action="store_true", help="After submitting workflow runner job, wait for completion.", @@ -268,11 +349,15 @@ def arg_parser(): # type: () -> argparse.ArgumentParser default=True, dest="wait") parser.add_argument("--api", type=str, - default=None, + default=None, dest="work_api", help="Select work submission API, one of 'jobs' or 'containers'.") - parser.add_argument("workflow", type=str, nargs="?", default=None) - parser.add_argument("job_order", nargs=argparse.REMAINDER) + parser.add_argument("--compute-checksum", action="store_true", default=False, + help="Compute checksum of contents while collecting outputs", + dest="compute_checksum") + + parser.add_argument("workflow", type=str, nargs="?", default=None, help="The workflow to execute") + parser.add_argument("job_order", nargs=argparse.REMAINDER, help="The input object to the workflow.") return parser @@ -282,21 +367,25 @@ def main(args, stdout, stderr, api_client=None): 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 = ({}, "") try: if api_client is None: api_client=arvados.api('v1', model=OrderedJsonModel()) - runner = ArvCwlRunner(api_client, crunch2=(arvargs.api == "containers")) + runner = ArvCwlRunner(api_client, work_api=arvargs.work_api) except Exception as e: logger.error(e) return 1 + arvargs.conformance_test = None + arvargs.use_container = True + return cwltool.main.main(args=arvargs, stdout=stdout, stderr=stderr, executor=runner.arvExecutor, makeTool=runner.arvMakeTool, versionfunc=versionstring, - job_order_object=job_order_object) + job_order_object=job_order_object, + make_fs_access=partial(CollectionFsAccess, api_client=api_client))