[ 
https://issues.apache.org/jira/browse/AIRFLOW-2236?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16415045#comment-16415045
 ] 

David Klosowski commented on AIRFLOW-2236:
------------------------------------------

I agree, this is an issue that can be resolved.  The big question is if to 
handle the backfill case akin to the non-backfill (current) case or if treating 
them separately makes sense. The feeling is that this should be SLA_MISS = if 
(task_end_datetime - task_start_datetime) > sla_interval.  I'd say that the 
non-backfill case has an argument for this to be based on execution_date + 
schedule_interval (intended start_date, so just substitute for 
task_start_date).  I'll ponder this more but not a fan of logical bifurcation 
as it creates more complexity and potential confusion.

> Airflow SLA is triggered for all backfilled tasks
> -------------------------------------------------
>
>                 Key: AIRFLOW-2236
>                 URL: https://issues.apache.org/jira/browse/AIRFLOW-2236
>             Project: Apache Airflow
>          Issue Type: Bug
>          Components: backfill
>    Affects Versions: Airflow 1.8, 1.8.1, 1.8.0, 1.9.0, 1.8.2
>            Reporter: barak schoster
>            Assignee: David Klosowski
>            Priority: Major
>
> While executing a task with a historical execution date and a schedule 
> interval of 1 hour, and SLA of 1 hours - all backfill instances appear as 
> sla_miss - though the duration of each task was not exceeded.
>  
> I think that this is a result of comparing the tasks end time to utc.now at 
> https://github.com/apache/incubator-airflow/blob/7cc6d8a5645b8974f07e132cc3c4820e880fd3ce/airflow/jobs.py#L624



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)

Reply via email to