Set up the network for NodeLink
A node mounts a NodeLink by connecting to the file system's mount target: a private IP address in one subnet of the file system's VPC, on TCP port 2049. Jaah sets up the file system; the route from each node to that address is yours to provide. This page shows, for each place a node can run, what to set in Jaah and in your network. How NodeLink works as a whole is in How NodeLink works.
Find your case:
| Where the node runs | What it needs | Section |
|---|---|---|
| Jaah Cloud | Nothing: use Jaah-managed storage | Jaah Cloud nodes |
| The file system's VPC, private subnet | A security group rule | Same VPC, private subnet |
| The file system's VPC, public subnet, with or without a public IP | The same rule; the public IP isn't used | Same VPC, public subnet |
| Another VPC, in your account or another one | Peering or a transit gateway, and the mount target IP | Another VPC |
| On premises or another cloud, through a VPN or Direct Connect | The VPN's routes, and the mount target IP | On premises over a VPN |
| On premises, another cloud or a laptop, behind NAT, no public IP | A Tailscale subnet router, and the mount target IP | Anywhere over Tailscale |
| On premises with a public IP | Still a private route: a VPN or a subnet router | On premises with a public IP |
Before you start
Every case below uses the same few settings. All of them are on the NodeLink's Storage tab, under Directory › Networks › NodeLink.
- New AWS EFS creates the file system, a security group and one mount target in the subnet of the
Connection you pick. The security group admits port 2049 from Allowed CIDR only, and every node is given
that mount target's IP, so nodes need no DNS. Pick a connection whose subnet your nodes can route to. Left
blank, Allowed CIDR is the VPC's own range; set it to every address your nodes' traffic arrives from when
any arrive from outside the VPC (on premises, a peered VPC, a subnet router). It takes one range, no wider
than
/8. - Existing AWS EFS uses your file system and your security groups as they are. Add an inbound rule for TCP 2049 from your nodes to the mount targets' security group yourself. Leave Mount target at None — resolved by DNS only when every node is inside the file system's VPC; as soon as one node is outside it, pick a mount target, so every node is given its IP.
- One NodeLink, one address. All nodes of a NodeLink connect to the same mount target IP when one is set. Nodes in several places can share a NodeLink as long as each place routes to that address.
- Nothing else on the network. A node needs no AWS credential and no inbound port for NodeLink. Besides
port 2049 to the mount target it only needs what every node needs: outbound HTTPS to Jaah. A Linux node also
needs the EFS mount helper,
amazon-efs-utils; without it the node says so on its NodeLinks tab.
A Windows node mounts through its WSL2 engine, so its traffic leaves through the Windows machine's own network: route the machine, as for a Linux node in the same place. A macOS node mounts a NodeLink once it is allowed Full Disk Access: see What a Mac needs.
Jaah Cloud nodes
When every node a NodeLink is attached to runs on a machine Jaah runs, in one location that Jaah-managed storage serves, the Storage tab offers Jaah-managed and picks it for you. A NodeLink that mixes Jaah Cloud nodes with nodes of your own, or nodes in two locations, can't use it. Jaah runs the file system and its network; you set up nothing.
Jaah Cloud machines sit in Jaah's network, not yours, and Jaah doesn't route them into your VPC. Nodes that run in your own network go on a NodeLink on your own EFS, set up as in the sections below.
Same VPC, private subnet
The simplest case: the node is an EC2 instance in the file system's VPC, in a subnet with no route to the internet except through a NAT gateway. Every subnet of a VPC routes to every other, so the node already reaches the mount target.
- On New AWS EFS, set Allowed CIDR to the VPC's range (
10.0.0.0/16here) or to your nodes' subnets. On Existing AWS EFS, allow TCP 2049 from that range on the mount targets' security group. - Make sure the nodes' own security group and the subnets' network ACLs allow outbound TCP 2049 (the AWS defaults do).
- On Existing AWS EFS with every node in this VPC, you may leave Mount target at None — resolved by DNS: each node then uses the mount target in its own Availability Zone. That needs a mount target in every zone your nodes run in, and DNS resolution turned on for the VPC. Otherwise pick one mount target.
Same VPC, public subnet
A node in a public subnet, whether or not it has a public IP, reaches the mount target over the VPC's own network exactly as a private one does. The public IP is never used: a mount target has no public address, and traffic to it never leaves the VPC.
Set it up as for a private subnet: Allowed CIDR (or your security group rule)
must hold the node's private address, 10.0.0.20 here, never its public one. Don't open port 2049 to the
internet: it would reach nothing, and it would only widen the security group.
Another VPC
A node in a different VPC, in your AWS account or another one, in the same region or another, needs a private route into the file system's VPC: VPC peering for two VPCs, or a transit gateway for several. The file system's DNS name doesn't resolve outside its own VPC, so the node must be given the mount target IP.
- Create the peering connection (accept it in the other account) or attach both VPCs to a transit gateway. The two VPCs' ranges must not overlap.
- In the nodes' VPC, route the file system's VPC range (or just the mount target's address) to the peering connection or transit gateway. In the file system's VPC, route the nodes' range back the same way, in the route table of the mount target's subnet.
- On New AWS EFS, set Allowed CIDR to the nodes' VPC range (
10.1.0.0/16here). On Existing AWS EFS, allow TCP 2049 from it on the mount targets' security group. - On Existing AWS EFS, pick a Mount target: Change beside Mount target IP on the Storage tab. Pick one in the Availability Zone most of your nodes use; traffic from other zones still works, at AWS's cross-zone data rate. New AWS EFS gives nodes its mount target's IP already.
AWS's guides: VPC peering and transit gateways.
On premises or another cloud over a VPN
A node on your own machine in your data centre, office or another cloud, with no public IP, reaches the file system over an AWS Site-to-Site VPN or AWS Direct Connect into the file system's VPC.
- Set up the VPN (or Direct Connect) between your network and the file system's VPC, through a virtual private gateway or a transit gateway.
- On your side, route the VPC's range, or just the mount target's address, into the VPN. In the VPC, route your on-premises range to the gateway in the mount target's subnet's route table (turning on route propagation does this for you).
- On New AWS EFS, set Allowed CIDR to your on-premises range (
192.168.10.0/24here). On Existing AWS EFS, allow TCP 2049 from it on the mount targets' security group. - On Existing AWS EFS, pick a Mount target, as for another VPC.
- The node still needs outbound HTTPS to Jaah, through your own internet connection.
AWS's guides: Site-to-Site VPN and Direct Connect.
Anywhere over a Tailscale subnet router
When a VPN to AWS isn't an option — a node behind NAT at home or in an office, a laptop, a machine in another cloud — run a Tailscale subnet router in the file system's VPC. It's a small instance that joins your organization's own tailnet and passes tailnet traffic on into the VPC. Nodes join the same tailnet and reach the mount target through it, from anywhere, with no public IP and no inbound port.
- Launch a small Linux instance in the file system's VPC, install Tailscale on it and turn on IP forwarding
(
net.ipv4.ip_forward=1), as Tailscale's guide shows. Advertise the mount target's address as a route (tailscale up --advertise-routes=10.0.2.40/32here; you may advertise the whole VPC range instead), and approve the route in your Tailscale admin console. - Install Tailscale on each node, join it to the same tailnet and accept routes (
tailscale up --accept-routes; Linux doesn't accept them by default). - The subnet router forwards traffic as if from its own address. On New AWS EFS, set Allowed CIDR to the
router's address (
10.0.1.50/32here) or its subnet; on Existing AWS EFS, allow TCP 2049 from it. - On Existing AWS EFS, pick a Mount target, as for another VPC.
Because traffic arrives from the router's address, this also helps when nodes in many places don't fit into one Allowed CIDR. Tailscale's guide: subnet routers.
This tailnet is yours, and Jaah doesn't manage it. A machine's Tailscale client is in one tailnet at a time: if a node is already in another tailnet, route it with a VPN instead.
On premises with a public IP
A public IP on an on-premises machine doesn't give it a way to the file system: the mount target answers only on its private address inside the VPC. Use a VPN or a Tailscale subnet router, exactly as for a machine without one, and set Allowed CIDR to the address the traffic arrives from over that route, never the public IP.
Check that it works
- From the node, check that port 2049 answers on the mount target IP (shown on the NodeLink's Storage tab
as Mount target IP), for example
nc -vz 10.0.2.40 2049. A timeout means a missing route, security group rule or network ACL on the way. - Attach the NodeLink to the node on its Nodes tab. The node usually shows Mounted within a minute; give it up to five minutes. A node that is offline mounts it when it next starts.
- If it shows Not mounted, read the reason beside it:
- The node can't reach the file system, or a
mount … failedline ending in a timeout — the route, a security group rule or a network ACL is missing: go through your case's steps again. - Outside the file system's VPC — the node tried to find the file system by name and couldn't. On Existing AWS EFS, pick a Mount target on the Storage tab.
- The node can't reach the file system, or a
Every field on the Storage and Nodes tabs is described in Networks › NodeLink.