Remove member before killing container(s)

[CRACKBOY]

Description

It would be better if the controller removes the member before killing the container(s).
Use case:

When marthon try to kill a container(s), it sends a task_killing state (SIGTERM) to gracefully shutdown it, then the container(s) enter in taskKillGracePeriodSeconds before sending SIGKILL (Source).
During this period and in case of a web server like Nginx which handle system signals, the F5 member should be unregistered in order to avoid redirection for all new connections and let the container terminate current connections before dying (due to taskKillGracePeriodS econds ).

Mesos Version

any

Marathon Version

any

Controller Version

v1.1.0

BIG-IP Version

any

[CRACKBOY]

@sjsharma2001, thank you for the enhancement label.
Otherwise, any information about how to handle the current situation with the actual system?

Kind Regards,
Slim.

[CRACKBOY]

@edarzins @larkinkevin any suggestion Guys ?

Regards,
Slim.

[amudukutore]

@CRACKBOY - our feature list for the upcoming feature release of the controller (v1.3.0 - mid-April timeframe) is pretty full and finalized. We do see value in adding this enhancement but would not be able to work on it until the following release (v1.4.0 in mid-July timeframe). If there is an urgency on your side to have this feature sooner, feel free to contribute a patch. Our contributing guidelines are listed on: https://github.com/F5Networks/marathon-bigip-ctlr/blob/master/CONTRIBUTING.md. Thanks.

[CRACKBOY]

@amudukutore thank you for your quick answer and I see the strategy for releasing...
Actually, the problem is very critical because when we scale down the containers... the f5 still forwarding users to killed members for some moments util the update happen... so I guess that's very danger for a production case no?

Regards,
Slim.