X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/8c593d84f88aa5cd87de0acedffdf867deca51f7..58b12d248ed05f8b75b16cee33b0e153e7be71f6:/sdk/cwl/arvados_cwl/executor.py diff --git a/sdk/cwl/arvados_cwl/executor.py b/sdk/cwl/arvados_cwl/executor.py index eab12efaac..c358426166 100644 --- a/sdk/cwl/arvados_cwl/executor.py +++ b/sdk/cwl/arvados_cwl/executor.py @@ -5,6 +5,7 @@ from __future__ import division from builtins import next from builtins import object +from builtins import str from future.utils import viewvalues import argparse @@ -58,6 +59,7 @@ 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 @@ -65,22 +67,27 @@ 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 @@ -360,8 +367,8 @@ http://doc.arvados.org/install/install-api-server.html#disable_api_methods 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 @@ -392,9 +399,9 @@ 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): @@ -467,8 +474,8 @@ 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) @@ -505,8 +512,9 @@ 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={ @@ -730,8 +738,11 @@ 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)