[ https://issues.apache.org/jira/browse/AIRFLOW-151?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Siddharth Anand updated AIRFLOW-151: ------------------------------------ Description: Porting from https://github.com/apache/incubator-airflow/issues/1521 Dear Airflow Maintainers, *Environment* {panel} Airflow version: 1.7.0rc3 Airflow components: webserver, scheduler, worker, postgres database, CeleryExecutor Relevant airflow.cfg settings: nothing special here; mostly defaults Python Version: 3.4.3 Operating System: Centos 6.7 Python packages: virtualenv with standard airflow install {panel} *Background* We are constructing a workflow to automate standard business processes around the creation and maintenance of virtual machines. After creation, we verify several information points on the VM to ensure that it is a viable machine and that no configuration errors occurred. If it fails verification and is not running, then it should be deleted. If it fails verification and is running, then we stop it first, then delete it. *What did you expect to happen?* After researching the BranchPythonOperator, I found that I should be using trigger_rule='one_success' to allow a task at a join point downstream of the branch(es) to be triggered, as mentioned in #1078. So, I defined the task as follows: {code} delete_vm = PythonOperator( task_id='delete_vm', trigger_rule=TriggerRule.ONE_SUCCESS, python_callable=_delete_vm, provide_context=True, dag=dag) {code} delete_vm.set_upstream({poll_vm_stop, verify_vm}) What happened instead? Rather than executing correctly, the delete_vm task is marked as skipped and is not re-evaluated following poll_vm_stop. There is no stack trace available, as the task simply does not execute. Sidenote: the PythonSensor you see in the picture below is a sensor which evaluates the truthy- or falsey-ness of a Python callable. It has been tested extensively and works as intended. image Any help would be greatly appreciated. I've tested various ways of linking the dag, providing DummyOperators as buffers, using a second BranchPythonOperator to explicitly call the task; all of these have failed. Am I missing something obvious here? was: Porting from https://github.com/apache/incubator-airflow/issues/1521 Dear Airflow Maintainers, *Environment* {format} Airflow version: 1.7.0rc3 Airflow components: webserver, scheduler, worker, postgres database, CeleryExecutor Relevant airflow.cfg settings: nothing special here; mostly defaults Python Version: 3.4.3 Operating System: Centos 6.7 Python packages: virtualenv with standard airflow install Description *Description of Issue* {format} *Background* We are constructing a workflow to automate standard business processes around the creation and maintenance of virtual machines. After creation, we verify several information points on the VM to ensure that it is a viable machine and that no configuration errors occurred. If it fails verification and is not running, then it should be deleted. If it fails verification and is running, then we stop it first, then delete it. *What did you expect to happen?* After researching the BranchPythonOperator, I found that I should be using trigger_rule='one_success' to allow a task at a join point downstream of the branch(es) to be triggered, as mentioned in #1078. So, I defined the task as follows: {code} delete_vm = PythonOperator( task_id='delete_vm', trigger_rule=TriggerRule.ONE_SUCCESS, python_callable=_delete_vm, provide_context=True, dag=dag) {code} delete_vm.set_upstream({poll_vm_stop, verify_vm}) What happened instead? Rather than executing correctly, the delete_vm task is marked as skipped and is not re-evaluated following poll_vm_stop. There is no stack trace available, as the task simply does not execute. Sidenote: the PythonSensor you see in the picture below is a sensor which evaluates the truthy- or falsey-ness of a Python callable. It has been tested extensively and works as intended. image Any help would be greatly appreciated. I've tested various ways of linking the dag, providing DummyOperators as buffers, using a second BranchPythonOperator to explicitly call the task; all of these have failed. Am I missing something obvious here? > trigger_rule='one_success' not allowing tasks downstream of a > BranchPythonOperator to be executed > ------------------------------------------------------------------------------------------------- > > Key: AIRFLOW-151 > URL: https://issues.apache.org/jira/browse/AIRFLOW-151 > Project: Apache Airflow > Issue Type: Bug > Reporter: Siddharth Anand > Assignee: Siddharth Anand > > Porting from https://github.com/apache/incubator-airflow/issues/1521 > Dear Airflow Maintainers, > *Environment* > {panel} > Airflow version: 1.7.0rc3 > Airflow components: webserver, scheduler, worker, postgres database, > CeleryExecutor > Relevant airflow.cfg settings: nothing special here; mostly defaults > Python Version: 3.4.3 > Operating System: Centos 6.7 > Python packages: virtualenv with standard airflow install > {panel} > *Background* > We are constructing a workflow to automate standard business processes around > the creation and maintenance of virtual machines. After creation, we verify > several information points on the VM to ensure that it is a viable machine > and that no configuration errors occurred. If it fails verification and is > not running, then it should be deleted. If it fails verification and is > running, then we stop it first, then delete it. > *What did you expect to happen?* > After researching the BranchPythonOperator, I found that I should be using > trigger_rule='one_success' to allow a task at a join point downstream of the > branch(es) to be triggered, as mentioned in #1078. So, I defined the task as > follows: > {code} > delete_vm = PythonOperator( > task_id='delete_vm', > trigger_rule=TriggerRule.ONE_SUCCESS, > python_callable=_delete_vm, > provide_context=True, > dag=dag) > {code} > delete_vm.set_upstream({poll_vm_stop, verify_vm}) > What happened instead? > Rather than executing correctly, the delete_vm task is marked as skipped and > is not re-evaluated following poll_vm_stop. There is no stack trace > available, as the task simply does not execute. Sidenote: the PythonSensor > you see in the picture below is a sensor which evaluates the truthy- or > falsey-ness of a Python callable. It has been tested extensively and works as > intended. > image > Any help would be greatly appreciated. I've tested various ways of linking > the dag, providing DummyOperators as buffers, using a second > BranchPythonOperator to explicitly call the task; all of these have failed. > Am I missing something obvious here? -- This message was sent by Atlassian JIRA (v6.3.4#6332)