Check Adding New Node Process

Question

Having never added a node to an existing cluster i thought i would check if my thinking is correct here. We have an existing 8 node cluster and adding some more nodes. These are identical model and spec nodes to the existing ones so not having to worry about compatibility issues there. But from what i can find its pretty straight forward but almost seems too straight forward, hence asking the question. Pre-Implementation/Checks Internal IPs - Int-a and int-b has an IP range of 40 addresses. And believe you only need 2 per node so that's all good as we are not going up to 20 nodes. External IP Range in subnets - Edit these to ensure have enough IP's if allocated as Static or Dynamic so each node can have at least one IP allocated. Implementation Rack and Cable up to Inifiniband switches and external network switches. Power on each node and either use a serial connection or front display to select option to join existing cluster. Wait while OneFS on the new nodes is upgraded or downgraded to match the version of the cluster. Post Implementation If i look in the WebGUI i should see on the Dashboard the new nodes listed after a couple minutes if all has gone well? The nodes will join the node pools automatically as they are the same model and spec as existing nodes wont they? All i would need to do now is edit the IP pool in each External network subnet and add in this node and that should be it, shouldn't it? Make sure that run either the Multiscan job or the Autobalance & Collect jobs to balance out the disk usage across the cluster. Questions Does that look right? Is it really that simple in theory or have i missed something obvious? And if you were adding multiple nodes to a cluster i assume you can add them all in one go? So can power one up, select join the cluster, power on the next, select join the cluster and repeat? If you wouldn't mind letting me know if i've missed anything obvious or not that would be appreciated. Thanks

GKi2drzsN11172279451969 · Accepted Answer

You've pretty much covered it and generally, yes, it's that easy. Pulling some of your comments in:

Internal IPs - Int-a and int-b has an IP range of 40 addresses. And believe you only need 2 per node so that's all good as we are not going up to 20 nodes.

It pulls one IP address per node added per pool. There are three assigned (int-a, int-b and failover), but it's only using one address from each of the three pools.

External IP Range in subnets - Edit these to ensure have enough IP's if allocated as Static or Dynamic so each node can have at least one IP allocated.

If you want addresses assigned, then this is correct. You don't need addresses, though, if you don't want them on the network. If you're not assigning addresses, make sure you remove interfaces from any pools that are automatically configured via the pool rules.

Power on each node and either use a serial connection or front display to select option to join existing cluster.

Wait while OneFS on the new nodes is upgraded or downgraded to match the version of the cluster.

You can also add them through the WebUI.

Other things to watch for: Make sure you do NOT have a Node Firmware Package (NFP) installed. If it is installed and you try to add the node, it will fail. Drive Support Package (DSP) should be fine, but I've had issues a couple of times with that.

If you want to go into advanced installation... The recommendation I make to the field support before installing a new node is to stand the node up as a single-node cluster (Infiniband and front end network cables MUST be unplugged). Once the node is up, install NFP and DSP, update firmware for the node and drives, then reimage it to the base level the cluster is running, boot to the install wizard and plug in the network cables. That way, you don't have to go through the firmware update procedure if it's down-rev from current while the node is active in the cluster.

AndrewP27 · Answer

Thanks for the reply. Fingers crossed it all goes to plan then.

Isilon

Was this post helpful?