Miretpl commented on code in PR #69034:
URL: https://github.com/apache/airflow/pull/69034#discussion_r3488418944


##########
chart/files/pod-template-file.kubernetes-helm-yaml:
##########
@@ -106,6 +106,15 @@ spec:
       env:
         - name: AIRFLOW__CORE__EXECUTOR
           value: {{ .Values.executor | quote }}
+        # Deliver pod-termination signals only to the task supervisor 
(dumb-init's
+        # direct child), not to the whole process group. This lets a running 
task
+        # finish on graceful pod shutdown -- the same warm-shutdown mechanism 
Celery
+        # workers use -- instead of being killed by dumb-init's group-wide 
broadcast,
+        # and removes the boot-time race where a SIGTERM could reach the task 
before
+        # its handler is installed. Hard kills (heartbeat loss / overtime / the
+        # post-grace SIGKILL) are unaffected.

Review Comment:
   ```suggestion
   ```
   For me, it is like it's partially how things work + knowledge regarding 
executors, so probably, if it is not already, it should be documented in the 
executor documentation instead of here.



##########
chart/files/pod-template-file.kubernetes-helm-yaml:
##########
@@ -106,6 +106,15 @@ spec:
       env:
         - name: AIRFLOW__CORE__EXECUTOR
           value: {{ .Values.executor | quote }}
+        # Deliver pod-termination signals only to the task supervisor 
(dumb-init's
+        # direct child), not to the whole process group. This lets a running 
task
+        # finish on graceful pod shutdown -- the same warm-shutdown mechanism 
Celery
+        # workers use -- instead of being killed by dumb-init's group-wide 
broadcast,
+        # and removes the boot-time race where a SIGTERM could reach the task 
before
+        # its handler is installed. Hard kills (heartbeat loss / overtime / the
+        # post-grace SIGKILL) are unaffected.
+        - name: DUMB_INIT_SETSID
+          value: "0"

Review Comment:
   This changes the default behaviour - could you add the newsfragment to 
`chart/newsframgnets` for it? Also, we could consider backporting it to the 
chart 1.2x line (on main we have 2.0 version), which I wouldn't do - users can 
add it via extra envs, and it would require manual backporting as it changes 
the default behaviour. 



-- 
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