Please explain advantages of LXD cluster over regular LXD

[igoryonya]

I’ve read the topic:

Basic questions about LXD cluster

I red the documentation about the cluster mode, but I am still a bit confused. Do container instances still live in specific nodes? What is the advantage of using a cluster instead of just running LXD on every machine? Is there any extra abstraction build of top of the cluster that allows me to just spin up containers and not worry about which node they end up on? I want to automate the provision of a DNS record per container on a cloud environment, say AWS. I believe a setup like this won’t w…

, but the only advantage, that I’ve gathered from there is that it will start on any node.
Are there more advantages for using LXD cluster?

[freeekanayaka]

From the top of my head:

• If you use ceph as storage driver, in case you lose a node you can restart all the containers it had on another node, in a very easy way.

• You have a N machines that act as a single LXD instance: if manage it like you would with a single LXD, but if you need more capacity (CPU, memory etc) you can just add more machines to the cluster.

• Images are retrieved from other nodes if they are not available on the local node, so in most cases you avoid one remote download per node and just have one remote download per cluster.

In general, LXD clustering is pretty much a replacement for vSphere, which instead of using virtual machines uses containers. The advantages of LXD clustering are similar to the ones you get from vSphere compared to a standalone vmware host.

[igoryonya]

@freeekanayaka

If you use ceph as storage driver, in case you lose a node you can restart all the containers it had on another node, in a very easy way.

So, it’s a manual restart? not automatic?
So, if I am using the same ceph storage on 2 independent (not clustered) servers with LXD installed and one of the servers dies, it’s not possible to launch the container, located on the ceph storage from an LXD server, that is still alive?

[freeekanayaka]

It’s a manual “restart” (technically you run "lxc move --target " and then "lxc start "). It’s not possible to do that easily if 1 of 2 independent LXD servers dies, because the server that remains alive does not have the needed container metadata. You can of course somehow recover the container somehow by poking with the ceph volume, but you would have to do it by yourself.

[lbg74 (Lee Byung Gon)]

— freekanayaka —
You have a N machines that act as a single LXD instance: if manage it like you would with a single LXD, but if you need more capacity (CPU, memory etc) you can just add more machines to the cluster.

If not share cpu and memory between nodes in the pool like IBM Power Enterprise Pool (PEP), what isthe difference between adding a node to the lxd cluster for distributing and adding vm or containers and adding a single lxd host ?

— freekanayaka —
Images are retrieved from other nodes if they are not available on the local node, so in most cases you avoid one remote download per node and just have one remote download per cluster.

Isn’t it possible to have the remote connection between single lxd hosts explained above?

Thank you