Oh, I did not write about K8S Executor because we already have K8S executor
support in Breeze. It does not use Breeze docker image - it uses KinD
(Kubernetes-in-Docker) that has all the bells and whistles (and more).

Here is the part of Breeze video when I showcase it:
https://github.com/apache/airflow/blob/master/BREEZE.rst#running-kubernetes-tests

And here the whole chapter describing how to test Airflow with K8S:
https://github.com/apache/airflow/blob/master/TESTING.rst#running-tests-with-kubernetes

What we have built-in with Breeze (automagically):

* configures and start kind cluster (kind is a really nice
development-focus installation of K8S that uses K8S-in-Docker approach).
* deploys airflow with K8S executor
* createss a k8s virtualenv which installs all necessary dependencies that
you can run tests against and allows you to enter it with `shell` command
* it even has nice k9s integration (which is a fantastic tool for
interacting with k8s)
* and it even has a step-by-step screenshot guide on how to debug the K8s
tests with IntelliJ.

So I believe K8S executor is well covered already :)

J.


On Mon, Mar 22, 2021 at 11:08 PM Ryan Hatter <[email protected]> wrote:

> Yeah I'll give that a shot.
>
> Do you think something with KubernetesExecutor with something like
> MiniKube would also make sense?
>
> On Mon, Mar 22, 2021 at 5:51 PM Jarek Potiuk <[email protected]> wrote:
>
>> Actually it is not that strightforward, but possibly we can make it works
>> much more easily
>>
>> In order to make Celery Executor works you need to do a bit more (but
>> should be easy to add as an option to Breeze):
>>
>> * you need to start rabitmq or redis as integration (`--integration
>> rabbitq --integration redis')
>> * you need to start worker(s) (`airflow worker` in the background)
>> * you might want to start flower optionally (the celery monitoring tool)
>>
>> So maybe we could add extra switch to start-airflow command
>> (--use-celery-executor) that could set those integrations and start
>> worker/flower additionally to running webserver/scheduler now in tmux ?
>>
>> WDYT? Maybe Ryan you can check if my recipe works ? I could add it then
>> as an option.
>>
>> BTW. We already have a number of CeleryExecutor tests that use the
>> integrations, so Breeze has all what's needed:
>>
>>
>> https://github.com/apache/airflow/blob/master/tests/executors/test_celery_executor.py#L109
>>
>>  J.
>>
>> On Mon, Mar 22, 2021 at 2:24 PM Ryan Hatter <[email protected]>
>> wrote:
>>
>>> Hmm, maybe I was just getting twisted around with docker then. I’ll have
>>> a look at what you shared.
>>>
>>> Thanks Bin :)
>>>
>>> On Mar 22, 2021, at 01:52, Xinbin Huang <[email protected]> wrote:
>>>
>>> 
>>> Hi Ryan,
>>>
>>> I believe breeze already provides tools for you to do that:
>>>
>>> 1.Would it make sense to allow developers to choose the executor for
>>> their Breeze environment?
>>>
>>> You can set up environment variables and any other custom setup you want
>>> in the file: */files/airflow-breeze-config/variables.env. *To set up
>>> CeleryExecutor, you just need to put `export
>>> AIRFLOW__CORE__EXECUTOR=CeleryExecutor` in the file.
>>>
>>> 2. Developing using Docker
>>> <https://airflow.apache.org/docs/apache-airflow/stable/start/docker.html> 
>>> would
>>> require an update to the image each time you want to make a change to the
>>> codebase
>>>
>>> Breeze automatically mounts the local source to the container unless you
>>> explicitly skip it with the flag *--skip-mounting-local-sources. *You
>>> can find more details here
>>> https://github.com/apache/airflow/blob/master/BREEZE.rst#mounting-local-sources-to-breeze
>>>
>>> Best
>>> Bin
>>>
>>>
>>> On Sun, Mar 21, 2021 at 5:32 PM Ryan Hatter <[email protected]>
>>> wrote:
>>>
>>>> I recently had some trouble trying to fix a bug in the CeleryExecutor
>>>> <https://github.com/apache/airflow/pull/14883>. The code change was
>>>> small, but it was really difficult to set up a development environment
>>>> using the CeleryExecutor. I ultimately had to muck around with the test
>>>> case that covers this situation, default_airflow.cfg, and
>>>> default_celery.py. Developing using Docker
>>>> <https://airflow.apache.org/docs/apache-airflow/stable/start/docker.html>
>>>> would require an update to the image each time you want to make a change to
>>>> the codebase (or maybe `exec`ing into the relevant container?) which is a
>>>> pain.
>>>>
>>>> This led me to two questions:
>>>>
>>>>    1. Would it make sense to allow developers to choose the executor
>>>>    for their Breeze environment?
>>>>    2. If not, how do folks test out changes they make to the
>>>>    CeleryExecutor or KubernetesExecutor?
>>>>
>>>> Thanks!
>>>> Ryan
>>>>
>>>>
>>>>
>>
>> --
>> +48 660 796 129
>>
>

-- 
+48 660 796 129

Reply via email to