Shreyas Joshi created AIRFLOW-735:
-------------------------------------
Summary: Dag instance status still "running" if a task instance in
a previous dag instance is still running
Key: AIRFLOW-735
URL: https://issues.apache.org/jira/browse/AIRFLOW-735
Project: Apache Airflow
Issue Type: Bug
Components: scheduler
Affects Versions: Airflow 1.7.1.3
Reporter: Shreyas Joshi
Attachments: bad-tree.jpg
Dag instance: A dag running at a particular time
Let us say we have a Dag instance D ~T~ that got kicked off at time T. A task
instance in D ~T~ takes unusually long such that subsequent dag instances D
~T+1~, D ~T+2~ and D ~T+3~ are kicked off. All the task instances in D ~T+1~, D
~T+2~ and D ~T+3~ finish, but there's one long running task instance in D ~T~
that still isn't done.
In this case:
* Even though all the task instances in dag instances D ~T+1~, D ~T+2~ and D
~T+3~ are done, the dag instances are still shown as running.
* When the long running task instance in D ~T~ finishes D ~T+1~, D ~T+2~ and D
~T+3~ are marked as successful.
In my case the task was probably hung. I killed the task manually and airflow
restarted it. However, I find it odd that a task instance in a given dag
instance should affect the status of subsequent dag instances. Is this expected
behavior?
[^bad-tree.jpg] shows the what the tree looked like before the long running
task was successful.
Nothing depends on previous runs. We are using the CelereyExecutor with three
worker nodes.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
