X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/70f8cb8fbc5cfce0f64adcfac181042390fcb3d6..c1429bc60d45525469ee9f73d5215814f405d8d7:/sdk/cwl/arvados_cwl/__init__.py diff --git a/sdk/cwl/arvados_cwl/__init__.py b/sdk/cwl/arvados_cwl/__init__.py index 4bff093d6b..7bfdba8b80 100644 --- a/sdk/cwl/arvados_cwl/__init__.py +++ b/sdk/cwl/arvados_cwl/__init__.py @@ -8,6 +8,8 @@ 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 @@ -15,15 +17,18 @@ 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 .perf import Perf 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') @@ -46,6 +51,8 @@ class ArvCwlRunner(object): self.num_retries = 4 self.uuid = None 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. @@ -56,7 +63,8 @@ class ArvCwlRunner(object): def arvMakeTool(self, toolpath_object, **kwargs): if "class" in toolpath_object and toolpath_object["class"] == "CommandLineTool": - return ArvadosCommandTool(self, toolpath_object, work_api=self.work_api, **kwargs) + kwargs["work_api"] = self.work_api + return ArvadosCommandTool(self, toolpath_object, **kwargs) else: return cwltool.workflow.defaultMakeTool(toolpath_object, **kwargs) @@ -90,28 +98,80 @@ class ArvCwlRunner(object): 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"]) + with Perf(logger, "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 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) + if self.debug: + logger.setLevel(logging.DEBUG) + 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) + self.fs_access = make_fs_access(kwargs["basedir"]) if kwargs.get("create_template"): tmpl = RunnerTemplate(self, tool, job_order, kwargs.get("enable_reuse")) @@ -119,20 +179,23 @@ 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 make_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" kwargs["on_error"] = "continue" - kwargs["compute_checksum"] = False + kwargs["compute_checksum"] = kwargs.get("compute_checksum") if self.work_api == "containers": 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)" @@ -164,12 +227,9 @@ class ArvCwlRunner(object): runnerjob.run() 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,)) @@ -188,7 +248,8 @@ class ArvCwlRunner(object): for runnable in jobiter: if runnable: - runnable.run(**kwargs) + with Perf(logger, "run"): + runnable.run(**kwargs) else: if self.processes: self.cond.wait(1) @@ -199,7 +260,6 @@ class ArvCwlRunner(object): while self.processes: self.cond.wait(1) - events.close() except UnsupportedRequirement: raise except: @@ -215,6 +275,8 @@ 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.") @@ -225,6 +287,9 @@ class ArvCwlRunner(object): if self.final_output is None: 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 @@ -269,7 +334,7 @@ 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("--ignore-docker-for-reuse", action="store_true", help="Ignore Docker image version when deciding whether to reuse past jobs.", default=False) @@ -280,6 +345,8 @@ 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.", @@ -291,6 +358,10 @@ def arg_parser(): # type: () -> argparse.ArgumentParser default=None, dest="work_api", help="Select work submission API, one of 'jobs' or 'containers'.") + 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.") @@ -302,7 +373,7 @@ 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: @@ -322,4 +393,5 @@ def main(args, stdout, stderr, api_client=None): 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))