Why no 2 node shared nothing?

[wimp9849]

Hi, i did some research about 2 node cluster wihtout shared storage and the answer is always the same: "is not possible to use 2 node for a cluster, minimum 3 with gluster or ceph"

So, how some vendor sell solution starting from 2 node? Starwind, nodeweaver, simplivity, ecc? Is not really possible to make an hyperconverged system with only 2 node in the open source world?

 

[tom]

No one can solve a split-brain situation with two nodes only (automatically), this is by design. If you want HA and reliable service, at least 3 nodes are needed. If someone tells you the opposite, think twice if you should go with it.

E.g. I read on Starwind FAQ:

...After both nodes of the HA cluster were down simultaneously in some cases StarWind is not able to determine which node holds the most recent data. So it blocks all the incoming connections...

You will find countless posts and here and on the internet.

 

[udo]

Hi, i did some research about 2 node cluster wihtout shared storage and the answer is always the same: "is not possible to use 2 node for a cluster, minimum 3 with gluster or ceph"

So, how some vendor sell solution starting from 2 node? Starwind, nodeweaver, simplivity, ecc? Is not really possible to make an hyperconverged system with only 2 node in the open source world?

Hi,
how you will build "hyper-converged infrastructure" (imho bullshitbingo marketing speech) without shared storage?!

You can use two systems as an cluster, but without HA! If one node die, all VMs on that node are gone...
But if you build the HA inside your VMs (like two DB-Server active/active plus web-server behind an loadbalancer) you can reach your goal perhaps with an two-node-cluster too.

Easier it's with three nodes (even without shared storage you can use an very small system as quorum node).

Udo

 

[wimp9849]

Hi,
how you will build "hyper-converged infrastructure" (imho bullshitbingo marketing speech) without shared storage?!

You can use two systems as an cluster, but without HA! If one node die, all VMs on that node are gone...
But if you build the HA inside your VMs (like two DB-Server active/active plus web-server behind an loadbalancer) you can reach your goal perhaps with an two-node-cluster too.

Easier it's with three nodes (even without shared storage you can use an very small system as quorum node).

Udo

Without shared Storage i mean without a San, replicating internal Storage between node.

 

[Net Runner]

I am not sure about other vendors but as far as i know Starwinds use additional interconnection link (mostly direct one) for heartbeat between the hosts that eliminates a need for third instance plus internal node majority mechanisms, however for the cases where direct redundant link is not possible they also implemented a witness option.

 

[MewBie]

No one can solve a split-brain situation with two nodes only (automatically), this is by design. If you want HA and reliable service, at least 3 nodes are needed. If someone tells you the opposite, think twice if you should go with it.

E.g. I read on Starwind FAQ:

...After both nodes of the HA cluster were down simultaneously in some cases StarWind is not able to determine which node holds the most recent data. So it blocks all the incoming connections...

You will find countless posts and here and on the internet.

Corosync 2 has some nice features to prevent a split brain and recover from a completely shut down cluster. It will never be proper HA, but it should increase availability in a lot of cases.

 

[jim.bond.9862]

I am not sure about other vendors but as far as i know Starwinds use additional interconnection link (mostly direct one) for heartbeat between the hosts that eliminates a need for third instance plus internal node majority mechanisms, however for the cases where direct redundant link is not possible they also implemented a witness option.

I am suing StarWind vSAN on my 2-node Server 2012 R1 Hyper-V cluster. been using it for about 1+ years.

I never had total cluster shutdown, but had total cluster lockup once. it was a combination of MS updates going bunkers and Starwind issue with some of the security patches that was installed as well as my free license lapsing without me noticing.
what didn't help was that all my AD/DC where VMs on said cluster as well.
so after I got all back up and running I moved one of the DC VMs onto standalone backup server just in case.
but it really works.