How to migrate container online ?

[LnxBil]

Hmm ... if you mean HA migration, yes Proxmox VE supports this; if you mean resetting the machine by a push on the reset button or a power outage (really hard shutdown), then no. This is no migration, this is failover. The VM will restart on the other side (the same as "current" LXC migration).

The case of a "critical HA" (as you wrote), so a fault-tolerant VM, this is currently being implemented on upstream QEMU/KVM and will be included into PVE, yet it is not finally released yet. This enables a kind of a "shadow" of a VM on another machine so if the one fails hard (reset, fire, meltdown...) the other will just jump in (with a small maybe negligible loss). IMHO this is already available in VMware, yet it is very hard to do correctly and almost impossible to do very fast, yet "critical HA" is normally not very fast as all other commercial products show.

 

[twinsen]

IMHO this is already available in VMware, yet it is very hard to do correctly and almost impossible to do very fast, yet "critical HA" is normally not very fast as all other commercial products show.

What I mean by that HA is, that consider a situation where LXC container is serving the Samba share, and on the normal Windows PC connected to this share, I'm doing a backup from Acronis True Image, or just uploading some _very-big-file_. Using for example Total Commander, but it doesn't matter. Then node 1 of 3 fails. What I expect, that transfer will still go, neither the IO on the Windows side, or the Samba server should notice that something happened. But it is good to hear that some like this is in development. As I said somewhere here, I saw vMotion in action, where Youtube video was still running smoothly even the node was powered off (I was at that time in this server room, where friend of mine was demonstrating this to me). I would love such a behaviour in Proxmox

 

[LnxBil]

I saw vMotion in action, where Youtube video was still running smoothly even the node was powered off (I was at that time in this server room, where friend of mine was demonstrating this to me).

Power off as in "shut down properly" or unplugged it while working? The former case is working with Proxmox VE and is called switch-over (in comparison to fail-over in the failing case)

What I mean by that HA is,

The problem is the HA terminology itself and, unfortunately, the English Wikipedia does not clarify this very well, although, the German one does and names the Harvard Research Group , which defines the Availability Environment Classification:
https://serverfault.com/a/693138

KVM on Proxmox VE is AE2, for everything else, you need to create a real cluster like Oracle does with their Real Application Cluster (RAC) Database, which is AE4. The downside of AE3 and AE4 is most certainly the price or its proprietariness.

AE3 and AE4 cannot work with LXC and does currently not work with KVM. In your scenario, it is very hard to have this with a TCP-based connection in a "dumb" environment, NFS could work, yet the file is most probably in a "strange" state and the application will hang. With "dumb" I mean that the service (in your case a file service) does not know about multiple paths or some failover mechanism. This problem can only be solved properly by using software that is built to fail and can recover from it, e.g. automatic rollback, session recovery etc. For Samba or CIFS-based filesystems, SMB transparent failover from Microsoft exists to do exactly what you want:

https://blogs.technet.microsoft.com...er-making-file-shares-continuously-available/

 

[twinsen]

For Samba or CIFS-based filesystems, SMB transparent failover from Microsoft exists to do exactly what you want:

https://blogs.technet.microsoft.com...er-making-file-shares-continuously-available/

Just read it and while it looks amazing, it could be also overkill to have a cluster of Windows Servers on the Proxmox cluster togeher with CEPH cluster... But, for home use I would probably use it like in my VM lab - LXC failover with shutdown and start. A probability for failure during any personal stuff transfer is rather low. I know, that we here usually are professionals and use Proxmox/CEPH in very large scale production environments, but I tend also to open such a amazing technologies to home users, that's why I'm considering some probably "strange" usecases ;-) But thanks for the link, it is worth to implement and try.

 

[NStorm]

Bumping this thread because I still don't get if it would be possible for live migration with LXC? Technically, I understand that it doesn't implemented right now in Proxmox. Because Proxmox 3.x allowed for online migration of OpenVZ containers even without shared storage with nearly zero downtime. Isn't it possible to do something like this for LXC?

 

[oguz]

Isn't it possible to do something like this for LXC?

it's dependent on criu, and it just doesn't work reliably