X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/9a7e2a24f5f3d261e554ac3815b7e2a4c2e24503..48c38895200cdafaaeca37299bf8352878389a77:/sdk/cwl/arvados_cwl/executor.py diff --git a/sdk/cwl/arvados_cwl/executor.py b/sdk/cwl/arvados_cwl/executor.py index c358426166..99d4c4e9a1 100644 --- a/sdk/cwl/arvados_cwl/executor.py +++ b/sdk/cwl/arvados_cwl/executor.py @@ -6,7 +6,7 @@ from __future__ import division from builtins import next from builtins import object from builtins import str -from future.utils import viewvalues +from future.utils import viewvalues, viewitems import argparse import logging @@ -31,7 +31,6 @@ from arvados.errors import ApiError import arvados_cwl.util from .arvcontainer import RunnerContainer -from .arvjob import RunnerJob, RunnerTemplate from .runner import Runner, upload_docker, upload_job_order, upload_workflow_deps from .arvtool import ArvadosCommandTool, validate_cluster_target, ArvadosExpressionTool from .arvworkflow import ArvadosWorkflow, upload_workflow @@ -45,6 +44,7 @@ from ._version import __version__ from cwltool.process import shortname, UnsupportedRequirement, use_custom_schema from cwltool.pathmapper import adjustFileObjs, adjustDirObjs, get_listing, visit_class from cwltool.command_line_tool import compute_checksums +from cwltool.load_tool import load_tool logger = logging.getLogger('arvados.cwl-runner') metrics = logging.getLogger('arvados.cwl-runner.metrics') @@ -90,8 +90,8 @@ class RuntimeStatusLoggingHandler(logging.Handler): class ArvCwlExecutor(object): - """Execute a CWL tool or workflow, submit work (using either jobs or - containers API), wait for them to complete, and report output. + """Execute a CWL tool or workflow, submit work (using containers API), + wait for them to complete, and report output. """ @@ -130,6 +130,8 @@ class ArvCwlExecutor(object): self.poll_interval = 12 self.loadingContext = None self.should_estimate_cache_size = True + self.fs_access = None + self.secret_store = None if keep_client is not None: self.keep_client = keep_client @@ -151,7 +153,7 @@ class ArvCwlExecutor(object): num_retries=self.num_retries) self.work_api = None - expected_api = ["jobs", "containers"] + expected_api = ["containers"] for api in expected_api: try: methods = self.api._rootDesc.get('resources')[api]['methods'] @@ -169,19 +171,11 @@ class ArvCwlExecutor(object): raise Exception("Unsupported API '%s', expected one of %s" % (arvargs.work_api, expected_api)) if self.work_api == "jobs": - logger.warning(""" + logger.error(""" ******************************* -Using the deprecated 'jobs' API. - -To get rid of this warning: - -Users: read about migrating at -http://doc.arvados.org/user/cwl/cwl-style.html#migrate -and use the option --api=containers - -Admins: configure the cluster to disable the 'jobs' API as described at: -http://doc.arvados.org/install/install-api-server.html#disable_api_methods +The 'jobs' API is no longer supported. *******************************""") + exit(1) self.loadingContext = ArvLoadingContext(vars(arvargs)) self.loadingContext.fetcher_constructor = self.fetcher_constructor @@ -256,7 +250,11 @@ http://doc.arvados.org/install/install-api-server.html#disable_api_methods activity statuses, for example in the RuntimeStatusLoggingHandler. """ with self.workflow_eval_lock: - current = arvados_cwl.util.get_current_container(self.api, self.num_retries, logger) + current = None + try: + current = arvados_cwl.util.get_current_container(self.api, self.num_retries, logger) + except Exception as e: + logger.info("Couldn't get current container: %s", e) if current is None: return runtime_status = current.get('runtime_status', {}) @@ -336,7 +334,7 @@ http://doc.arvados.org/install/install-api-server.html#disable_api_methods return "[%s %s]" % (self.work_api[0:-1], obj.name) def poll_states(self): - """Poll status of jobs or containers listed in the processes dict. + """Poll status of containers listed in the processes dict. Runs in a separate thread. """ @@ -357,14 +355,11 @@ http://doc.arvados.org/install/install-api-server.html#disable_api_methods begin_poll = time.time() if self.work_api == "containers": table = self.poll_api.container_requests() - elif self.work_api == "jobs": - table = self.poll_api.jobs() pageSize = self.poll_api._rootDesc.get('maxItemsPerResponse', 1000) while keys: page = keys[:pageSize] - keys = keys[pageSize:] try: proc_states = table.list(filters=[["uuid", "in", page]]).execute(num_retries=self.num_retries) except Exception: @@ -380,6 +375,8 @@ http://doc.arvados.org/install/install-api-server.html#disable_api_methods "new_attributes": p } }) + keys = keys[pageSize:] + finish_poll = time.time() remain_wait = self.poll_interval - (finish_poll - begin_poll) except: @@ -404,26 +401,22 @@ http://doc.arvados.org/install/install-api-server.html#disable_api_methods except (KeyboardInterrupt, SystemExit): break - def check_features(self, obj): + def check_features(self, obj, parentfield=""): if isinstance(obj, dict): - if obj.get("writable") and self.work_api != "containers": - raise SourceLine(obj, "writable", UnsupportedRequirement).makeError("InitialWorkDir feature 'writable: true' not supported with --api=jobs") if obj.get("class") == "DockerRequirement": if obj.get("dockerOutputDirectory"): - if self.work_api != "containers": - raise SourceLine(obj, "dockerOutputDirectory", UnsupportedRequirement).makeError( - "Option 'dockerOutputDirectory' of DockerRequirement not supported with --api=jobs.") if not obj.get("dockerOutputDirectory").startswith('/'): raise SourceLine(obj, "dockerOutputDirectory", validate.ValidationException).makeError( "Option 'dockerOutputDirectory' must be an absolute path.") - if obj.get("class") == "http://commonwl.org/cwltool#Secrets" and self.work_api != "containers": - raise SourceLine(obj, "class", UnsupportedRequirement).makeError("Secrets not supported with --api=jobs") - for v in viewvalues(obj): - self.check_features(v) + if obj.get("class") == "InplaceUpdateRequirement": + if obj["inplaceUpdate"] and parentfield == "requirements": + raise SourceLine(obj, "class", UnsupportedRequirement).makeError("InplaceUpdateRequirement not supported for keep collections.") + for k,v in viewitems(obj): + self.check_features(v, parentfield=k) elif isinstance(obj, list): for i,v in enumerate(obj): with SourceLine(obj, i, UnsupportedRequirement, logger.isEnabledFor(logging.DEBUG)): - self.check_features(v) + self.check_features(v, parentfield=parentfield) def make_output_collection(self, name, storage_classes, tagsString, outputObj): outputObj = copy.deepcopy(outputObj) @@ -442,17 +435,16 @@ http://doc.arvados.org/install/install-api-server.html#disable_api_methods num_retries=self.num_retries) for k,v in generatemapper.items(): - if k.startswith("_:"): - if v.type == "Directory": + if v.type == "Directory" and v.resolved.startswith("_:"): continue - if v.type == "CreateFile": - with final.open(v.target, "wb") as f: - f.write(v.resolved.encode("utf-8")) + if v.type == "CreateFile" and (k.startswith("_:") or v.resolved.startswith("_:")): + with final.open(v.target, "wb") as f: + f.write(v.resolved.encode("utf-8")) continue - if not k.startswith("keep:"): + if not v.resolved.startswith("keep:"): raise Exception("Output source is not in keep or a literal") - sp = k.split("/") + sp = v.resolved.split("/") srccollection = sp[0][5:] try: reader = self.collection_cache.get(srccollection) @@ -462,7 +454,8 @@ http://doc.arvados.org/install/install-api-server.html#disable_api_methods logger.error("Creating CollectionReader for '%s' '%s': %s", k, v, e) raise except IOError as e: - logger.warning("While preparing output collection: %s", e) + logger.error("While preparing output collection: %s", e) + raise def rewrite(fileobj): fileobj["location"] = generatemapper.mapper(fileobj["location"]).target @@ -515,18 +508,23 @@ http://doc.arvados.org/install/install-api-server.html#disable_api_methods except Exception: logger.exception("Setting container output") return - 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, runtimeContext, logger=None): + + def apply_reqs(self, job_order_object, tool): + if "https://w3id.org/cwl/cwl#requirements" in job_order_object: + if tool.metadata.get("http://commonwl.org/cwltool#original_cwlVersion") == 'v1.0': + raise WorkflowException( + "`cwl:requirements` in the input object is not part of CWL " + "v1.0. You can adjust to use `cwltool:overrides` instead; or you " + "can set the cwlVersion to v1.1 or greater and re-run with " + "--enable-dev.") + job_reqs = job_order_object["https://w3id.org/cwl/cwl#requirements"] + for req in job_reqs: + tool.requirements.append(req) + + def arv_executor(self, updated_tool, job_order, runtimeContext, logger=None): self.debug = runtimeContext.debug - tool.visit(self.check_features) + updated_tool.visit(self.check_features) self.project_uuid = runtimeContext.project_uuid self.pipeline = None @@ -547,43 +545,52 @@ http://doc.arvados.org/install/install-api-server.html#disable_api_methods raise Exception("--submit-request-uuid requires containers API, but using '{}' api".format(self.work_api)) if not runtimeContext.name: - runtimeContext.name = self.name = tool.tool.get("label") or tool.metadata.get("label") or os.path.basename(tool.tool["id"]) + runtimeContext.name = self.name = updated_tool.tool.get("label") or updated_tool.metadata.get("label") or os.path.basename(updated_tool.tool["id"]) + + # Upload local file references in the job order. + job_order = upload_job_order(self, "%s input" % runtimeContext.name, + updated_tool, job_order) + + # the last clause means: if it is a command line tool, and we + # are going to wait for the result, and always_submit_runner + # is false, then we don't submit a runner process. + + submitting = (runtimeContext.update_workflow or + runtimeContext.create_workflow or + (runtimeContext.submit and not + (updated_tool.tool["class"] == "CommandLineTool" and + runtimeContext.wait and + not runtimeContext.always_submit_runner))) + + loadingContext = self.loadingContext.copy() + loadingContext.do_validate = False + loadingContext.do_update = False + if submitting: + # Document may have been auto-updated. Reload the original + # document with updating disabled because we want to + # submit the document with its original CWL version, not + # the auto-updated one. + tool = load_tool(updated_tool.tool["id"], loadingContext) + else: + tool = updated_tool # Upload direct dependencies of workflow steps, get back mapping of files to keep references. # Also uploads docker images. merged_map = upload_workflow_deps(self, tool) - # Reload tool object which may have been updated by - # upload_workflow_deps - # Don't validate this time because it will just print redundant errors. - loadingContext = self.loadingContext.copy() + # Recreate process object (ArvadosWorkflow or + # ArvadosCommandTool) because tool document may have been + # updated by upload_workflow_deps in ways that modify + # inheritance of hints or requirements. loadingContext.loader = tool.doc_loader loadingContext.avsc_names = tool.doc_schema loadingContext.metadata = tool.metadata - loadingContext.do_validate = False - - tool = self.arv_make_tool(tool.doc_loader.idx[tool.tool["id"]], - loadingContext) - - # Upload local file references in the job order. - job_order = upload_job_order(self, "%s input" % runtimeContext.name, - tool, job_order) + tool = load_tool(tool.tool, loadingContext) existing_uuid = runtimeContext.update_workflow if existing_uuid or runtimeContext.create_workflow: # Create a pipeline template or workflow record and exit. - if self.work_api == "jobs": - tmpl = RunnerTemplate(self, tool, job_order, - runtimeContext.enable_reuse, - uuid=existing_uuid, - submit_runner_ram=runtimeContext.submit_runner_ram, - name=runtimeContext.name, - merged_map=merged_map, - loadingContext=loadingContext) - tmpl.save() - # cwltool.main will write our return value to stdout. - return (tmpl.uuid, "success") - elif self.work_api == "containers": + if self.work_api == "containers": return (upload_workflow(self, tool, job_order, self.project_uuid, uuid=existing_uuid, @@ -592,6 +599,8 @@ http://doc.arvados.org/install/install-api-server.html#disable_api_methods merged_map=merged_map), "success") + self.apply_reqs(job_order, tool) + self.ignore_docker_for_reuse = runtimeContext.ignore_docker_for_reuse self.eval_timeout = runtimeContext.eval_timeout @@ -607,12 +616,6 @@ http://doc.arvados.org/install/install-api-server.html#disable_api_methods runtimeContext.docker_outdir = "/var/spool/cwl" runtimeContext.tmpdir = "/tmp" runtimeContext.docker_tmpdir = "/tmp" - elif self.work_api == "jobs": - if runtimeContext.priority != DEFAULT_PRIORITY: - raise Exception("--priority not implemented for jobs API.") - runtimeContext.outdir = "$(task.outdir)" - runtimeContext.docker_outdir = "$(task.outdir)" - runtimeContext.tmpdir = "$(task.tmpdir)" if runtimeContext.priority < 1 or runtimeContext.priority > 1000: raise Exception("--priority must be in the range 1..1000.") @@ -636,40 +639,23 @@ http://doc.arvados.org/install/install-api-server.html#disable_api_methods if runtimeContext.submit: # Submit a runner job to run the workflow for us. if self.work_api == "containers": - if tool.tool["class"] == "CommandLineTool" and runtimeContext.wait and (not runtimeContext.always_submit_runner): - runtimeContext.runnerjob = tool.tool["id"] + if submitting: + tool = RunnerContainer(self, updated_tool, + tool, loadingContext, runtimeContext.enable_reuse, + self.output_name, + self.output_tags, + submit_runner_ram=runtimeContext.submit_runner_ram, + name=runtimeContext.name, + on_error=runtimeContext.on_error, + submit_runner_image=runtimeContext.submit_runner_image, + intermediate_output_ttl=runtimeContext.intermediate_output_ttl, + merged_map=merged_map, + priority=runtimeContext.priority, + secret_store=self.secret_store, + collection_cache_size=runtimeContext.collection_cache_size, + collection_cache_is_default=self.should_estimate_cache_size) else: - tool = RunnerContainer(self, tool, loadingContext, runtimeContext.enable_reuse, - self.output_name, - self.output_tags, - submit_runner_ram=runtimeContext.submit_runner_ram, - name=runtimeContext.name, - on_error=runtimeContext.on_error, - submit_runner_image=runtimeContext.submit_runner_image, - intermediate_output_ttl=runtimeContext.intermediate_output_ttl, - merged_map=merged_map, - priority=runtimeContext.priority, - secret_store=self.secret_store, - collection_cache_size=runtimeContext.collection_cache_size, - collection_cache_is_default=self.should_estimate_cache_size) - elif self.work_api == "jobs": - tool = RunnerJob(self, tool, loadingContext, runtimeContext.enable_reuse, - self.output_name, - self.output_tags, - submit_runner_ram=runtimeContext.submit_runner_ram, - name=runtimeContext.name, - on_error=runtimeContext.on_error, - submit_runner_image=runtimeContext.submit_runner_image, - merged_map=merged_map) - elif runtimeContext.cwl_runner_job is None and self.work_api == "jobs": - # Create pipeline for local run - self.pipeline = self.api.pipeline_instances().create( - body={ - "owner_uuid": self.project_uuid, - "name": runtimeContext.name if runtimeContext.name else shortname(tool.tool["id"]), - "components": {}, - "state": "RunningOnClient"}).execute(num_retries=self.num_retries) - logger.info("Pipeline instance %s", self.pipeline["uuid"]) + runtimeContext.runnerjob = tool.tool["id"] if runtimeContext.cwl_runner_job is not None: self.uuid = runtimeContext.cwl_runner_job.get('uuid') @@ -746,11 +732,16 @@ http://doc.arvados.org/install/install-api-server.html#disable_api_methods if self.pipeline: self.api.pipeline_instances().update(uuid=self.pipeline["uuid"], body={"state": "Failed"}).execute(num_retries=self.num_retries) - if runtimeContext.submit and isinstance(tool, Runner): - runnerjob = tool - if runnerjob.uuid and self.work_api == "containers": - self.api.container_requests().update(uuid=runnerjob.uuid, - body={"priority": "0"}).execute(num_retries=self.num_retries) + + if self.work_api == "containers" and not current_container: + # Not running in a crunch container, so cancel any outstanding processes. + for p in self.processes: + try: + self.api.container_requests().update(uuid=p, + body={"priority": "0"} + ).execute(num_retries=self.num_retries) + except Exception: + pass finally: self.workflow_eval_lock.release() self.task_queue.drain()