[lxc-users] LXD move, how to reduce downtime without live migration

[Jäkel, Guido]

Dear Fajar,

with respect, imho this time you go too far. It's definite not in the responsibility of any common application to be data-consistent at every point in time. In the opposite, there are well-known and well-respected signaling mechanisms  at both sides -- application and operating system -- to deal in an appropriate way with an "abnormal halt" condition. At the Unix platform, a termination signal is send to the program and the program have to should down as fast and as good as possible.

Exactly this is defined to treat as the "correct behavior" and common application following this are "correct"! Of course, every developer and every operator will be well advised to have a plan about to deal with the situation where even this ungly but controlled shutdowns failed.

But it's an absolute no-go to *build* normal workflows on instruments that should be reservered for exceptional cases. With another picuture: You can try to repair your car in case of a traffic accident. But I hope you don't want to vote here to align the common driving stile on this.

To my eyes, a snapshot of anything is just a "better than nothing", but far away from something that should be called a resilient solution to satisfy requirements of safety. One may use it as a comfortable solution to restore things by hand, but it's not suitable for any automatism. I use and rely on this feature too: Ups, I just forget to backup a configuration file before editing. No matter, it wasn't changed before last automatic snapshot and I may draw for the mandatory backup from this source.


Greetings

Guido


>-----Original Message-----
>From: lxc-users [mailto:lxc-users-bounces at lists.linuxcontainers.org] On Behalf Of Fajar A. Nugraha
>Sent: Tuesday, May 02, 2017 4:00 AM
>To: LXC users mailing-list
>Subject: Re: [lxc-users] LXD move, how to reduce downtime without live migration
>If you have one of those apps, I highly recommend you fix it, or find another one that behave correctly. You don't want to be
>left out cold when a real power interruption (or server crash) happens.
>
>--
>Fajar