You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@airflow.apache.org by "Aizhamal Nurmamat kyzy (JIRA)" <ji...@apache.org> on 2019/05/18 03:03:00 UTC

[jira] [Updated] (AIRFLOW-4385) Add namespace into pod's yaml

     [ https://issues.apache.org/jira/browse/AIRFLOW-4385?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Aizhamal Nurmamat kyzy updated AIRFLOW-4385:
--------------------------------------------
    Labels: kubernetes  (was: )

> Add namespace into pod's yaml
> -----------------------------
>
>                 Key: AIRFLOW-4385
>                 URL: https://issues.apache.org/jira/browse/AIRFLOW-4385
>             Project: Apache Airflow
>          Issue Type: Improvement
>          Components: executor, worker
>            Reporter: Nikolay
>            Assignee: Nikolay
>            Priority: Major
>              Labels: kubernetes
>         Attachments: pods_logs.txt
>
>
> During integration phase between Airflow and VerticalPodAutoscaler (link: [https://github.com/kubernetes/autoscaler/tree/master/vertical-pod-autoscaler]) i stumbled into an issue when VerticalPodAutoscaler library doesn't work for Airflow worker pods. 
> How VerticalPodAutoscaler produces resource recommendations:
> - monitor newly created pods
> - compare newly created pods with existing configuration (compare by label and namespace)
> - provide new recommendations if configuration exists for pod
>  VerticalPodAutoscaler finds all created pods by label, but namespace comparison fails every time. I did some investigation and realise that when airflow calls a kubernetes api, it passes yaml file without +metadata.namespace+ property. Go libraries can deserialise this yaml, but there is no information about namespace.
> Attaching logs from VerticalPodAutoscaler, there you can see how airflow pod was deserialised (namespace is missing). Works for custom pod.



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