X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/be57c52f8ea23852d9d00ab1981b41354f59f27d..e838828374ceed5ef6da260939251e86f72b6f27:/sdk/cwl/arvados_cwl/executor.py diff --git a/sdk/cwl/arvados_cwl/executor.py b/sdk/cwl/arvados_cwl/executor.py index 14ea8c534c..190ea35e2a 100644 --- a/sdk/cwl/arvados_cwl/executor.py +++ b/sdk/cwl/arvados_cwl/executor.py @@ -5,8 +5,8 @@ from __future__ import division from builtins import next from builtins import object -from future.utils import viewvalues -from past.utils import old_div +from builtins import str +from future.utils import viewvalues, viewitems import argparse import logging @@ -45,6 +45,7 @@ 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') @@ -59,6 +60,7 @@ class RuntimeStatusLoggingHandler(logging.Handler): def __init__(self, runtime_status_update_func): super(RuntimeStatusLoggingHandler, self).__init__() self.runtime_status_update = runtime_status_update_func + self.updatingRuntimeStatus = False def emit(self, record): kind = None @@ -66,22 +68,27 @@ class RuntimeStatusLoggingHandler(logging.Handler): kind = 'error' elif record.levelno >= logging.WARNING: kind = 'warning' - if kind is not None: - log_msg = record.getMessage() - if '\n' in log_msg: - # If the logged message is multi-line, use its first line as status - # and the rest as detail. - status, detail = log_msg.split('\n', 1) - self.runtime_status_update( - kind, - "%s: %s" % (record.name, status), - detail - ) - else: - self.runtime_status_update( - kind, - "%s: %s" % (record.name, record.getMessage()) - ) + if kind is not None and self.updatingRuntimeStatus is not True: + self.updatingRuntimeStatus = True + try: + log_msg = record.getMessage() + if '\n' in log_msg: + # If the logged message is multi-line, use its first line as status + # and the rest as detail. + status, detail = log_msg.split('\n', 1) + self.runtime_status_update( + kind, + "%s: %s" % (record.name, status), + detail + ) + else: + self.runtime_status_update( + kind, + "%s: %s" % (record.name, record.getMessage()) + ) + finally: + self.updatingRuntimeStatus = False + class ArvCwlExecutor(object): """Execute a CWL tool or workflow, submit work (using either jobs or @@ -124,6 +131,8 @@ 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 @@ -145,7 +154,7 @@ class ArvCwlExecutor(object): num_retries=self.num_retries) self.work_api = None - expected_api = ["jobs", "containers"] + expected_api = ["containers", "jobs"] for api in expected_api: try: methods = self.api._rootDesc.get('resources')[api]['methods'] @@ -343,7 +352,7 @@ http://doc.arvados.org/install/install-api-server.html#disable_api_methods if self.stop_polling.is_set(): break with self.workflow_eval_lock: - keys = list(self.processes.keys()) + keys = list(self.processes) if not keys: remain_wait = self.poll_interval continue @@ -358,11 +367,10 @@ http://doc.arvados.org/install/install-api-server.html#disable_api_methods 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 as e: - logger.warning("Error checking states on API server: %s", e) + except Exception: + logger.exception("Error checking states on API server: %s") remain_wait = self.poll_interval continue @@ -374,6 +382,8 @@ 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: @@ -393,12 +403,12 @@ http://doc.arvados.org/install/install-api-server.html#disable_api_methods for i in self.intermediate_output_collections: try: self.api.collections().delete(uuid=i).execute(num_retries=self.num_retries) - except: + except Exception: logger.warning("Failed to delete intermediate output: %s", sys.exc_info()[1], exc_info=(sys.exc_info()[1] if self.debug else False)) - if sys.exc_info()[0] is KeyboardInterrupt or sys.exc_info()[0] is SystemExit: + 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") @@ -412,12 +422,15 @@ http://doc.arvados.org/install/install-api-server.html#disable_api_methods "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) @@ -435,18 +448,17 @@ http://doc.arvados.org/install/install-api-server.html#disable_api_methods keep_client=self.keep_client, num_retries=self.num_retries) - for k,v in list(generatemapper.items()): - if k.startswith("_:"): - if v.type == "Directory": + for k,v in generatemapper.items(): + 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) @@ -456,7 +468,8 @@ 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 @@ -468,8 +481,8 @@ http://doc.arvados.org/install/install-api-server.html#disable_api_methods adjustFileObjs(outputObj, rewrite) with final.open("cwl.output.json", "w") as f: - res = json.dumps(outputObj, sort_keys=True, indent=4, separators=(',',': '), ensure_ascii=False).encode('utf-8').decode() - f.write(res) + res = str(json.dumps(outputObj, sort_keys=True, indent=4, separators=(',',': '), ensure_ascii=False)) + f.write(res) final.save_new(name=name, owner_uuid=self.project_uuid, storage_classes=storage_classes, ensure_unique_name=True) @@ -506,8 +519,9 @@ http://doc.arvados.org/install/install-api-server.html#disable_api_methods body={ 'is_trashed': True }).execute(num_retries=self.num_retries) - except Exception as e: - logger.info("Setting container output: %s", e) + 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={ @@ -516,6 +530,18 @@ http://doc.arvados.org/install/install-api-server.html#disable_api_methods 'progress':1.0 }).execute(num_retries=self.num_retries) + 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, tool, job_order, runtimeContext, logger=None): self.debug = runtimeContext.debug @@ -542,25 +568,38 @@ http://doc.arvados.org/install/install-api-server.html#disable_api_methods if not runtimeContext.name: runtimeContext.name = self.name = tool.tool.get("label") or tool.metadata.get("label") or os.path.basename(tool.tool["id"]) + # Upload local file references in the job order. + job_order = upload_job_order(self, "%s input" % runtimeContext.name, + tool, job_order) + + submitting = (runtimeContext.update_workflow or + runtimeContext.create_workflow or + (runtimeContext.submit and not + (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 original document, not the auto-updated one. + tool = load_tool(tool.tool["id"], loadingContext) + # 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: @@ -585,6 +624,8 @@ 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 @@ -620,7 +661,7 @@ http://doc.arvados.org/install/install-api-server.html#disable_api_methods visited.add(m.group(1)) estimated_size[0] += int(m.group(2)) visit_class(job_order, ("File", "Directory"), estimate_collection_cache) - runtimeContext.collection_cache_size = max((old_div((estimated_size[0]*192), (1024*1024)))+1, 256) + runtimeContext.collection_cache_size = max(((estimated_size[0]*192) // (1024*1024))+1, 256) self.collection_cache.set_cap(runtimeContext.collection_cache_size*1024*1024) logger.info("Using collection cache size %s MiB", runtimeContext.collection_cache_size) @@ -731,8 +772,11 @@ http://doc.arvados.org/install/install-api-server.html#disable_api_methods except: if sys.exc_info()[0] is KeyboardInterrupt or sys.exc_info()[0] is SystemExit: logger.error("Interrupted, workflow will be cancelled") + elif isinstance(sys.exc_info()[1], WorkflowException): + logger.error("Workflow execution failed:\n%s", sys.exc_info()[1], exc_info=(sys.exc_info()[1] if self.debug else False)) else: - logger.error("Execution failed:\n%s", sys.exc_info()[1], exc_info=(sys.exc_info()[1] if self.debug else False)) + logger.exception("Workflow execution failed") + if self.pipeline: self.api.pipeline_instances().update(uuid=self.pipeline["uuid"], body={"state": "Failed"}).execute(num_retries=self.num_retries)