potiuk commented on PR #25678:
URL: https://github.com/apache/airflow/pull/25678#issuecomment-1233360504

   I believe this one should be already green and mergable, and it is what I 
wanted it to be. It's WAY nicer comparing to what we had before and as 
explained in #26091 we will be able to further improve the development 
experience of Airlfow running on K8S for everyone as a follow-up.
   
   @dstandish @ephraimbuddy  @jedcunningham  - I think you were one of the 
users there (or maybe non-users because of not-so-great experience).
   
   This Python-Breeze version has a lot of improvements comparing to the legacy 
bash one:
   
   * The CI has 1-1 mapping to the local development commands. You follow 
exactly the same sequence locally to run tests locally as you do in CI 
(screenshot below)
   
   * The commands are consistent with the rest of the commands of breeze , and 
allow both local single-cluster running as well as CI running in parallel 
   
   * The documentation is simplified - some of the duplication was removed and 
it's much clearer now how to approach it IMHO
   
   * Also the commands to setup testable Airlfow instance are pretty much 
"wizard-like" (@rossturk that is the result of our discussions). When you 
create cluster, it tells you "NEXT STEP:" that you should configure it and 
tells you the command, when you configure it it asks you to build k8s as "NEXT 
STEP" and tells the command, and so on. It basically guides the user 
step-by-step on what to do next. You can literally re-run all k8c tests by 
running `breeze k8s create-cluster` and following the "NEXT STEPS" 
instructions. 
   
   * The "K8S upgrade" step is now part of the regular jobs. We save a lot of 
time for setting up the clusters and deploying airlfow again as they are 
already set-up in previous steps and we simply need to add one "upgrade" step 
with some changed valiues (@ephraimbuddy  - you will be happy). This makes the 
tests less resource hungry and faster.
   
   * I found and fixed a few stability issues that plagued us with the Helm 
tests. They should be much more solid now in general. That included necessary 
increase in our CI machine - to add more inotify watchers on system level for 
example (and few other lower-level stuff).
   
   * It's all Python
   
   * The output of CI is really nice and tells exactly what happens
   
   Example succesful run with multiple clusters in parallel:
   
   https://github.com/apache/airflow/runs/8114807752?check_suite_focus=true
   
   Some screenshots:
   
   Steps in the CI job:
   
   
![image](https://user-images.githubusercontent.com/595491/187769369-399bdac4-f1fe-4c78-8c89-fd6294483b69.png)
   
   Job summary:
   
   
![image](https://user-images.githubusercontent.com/595491/187770006-00503ec1-22eb-4d5f-b2f8-ddcba91686d3.png)
   
   
   Job details after unfolding:
   
   
![image](https://user-images.githubusercontent.com/595491/187770172-da012851-877b-4c7c-b31a-09d9ec330828.png)
   
   
   


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]

Reply via email to