Clients mount over NFS 4.1, from inside the file system’s VPC. Mount by the mount target’s
hostname, or by its ipv4Address. The path is the file system’s ID (/fsy_…), or an
access point’s (/fap_…).
The mount command
The instance needs an NFS client. On an image without one, install nfs-common (Ubuntu, Debian:
sudo apt-get install -y nfs-common) or nfs-utils (Rocky Linux: sudo dnf install -y nfs-utils).
sudo mount -t nfs4 -o nfsvers=4.1,xprtsec=tls,hard,noresvport,rsize=1048576,wsize=1048576,timeo=600,retrans=2 \
<hostname>:/<fsy or fap> /mnt/<name>Without TLS, drop xprtsec=tls; you can then mount by the address instead of the name:
sudo mount -t nfs4 -o nfsvers=4.1,hard,noresvport,rsize=1048576,wsize=1048576,timeo=600,retrans=2 \
<ipv4Address>:/<fsy or fap> /mnt/<name>A file system with requireTls accepts only the first. TLS needs a few things on the client first:
see Encrypt with TLS.
| Option | Why |
|---|---|
nfsvers=4.1 |
Files serves NFS 4.1 |
xprtsec=tls |
Encrypts the connection with TLS |
hard |
If the file system is briefly unreachable, reads and writes wait and then carry on, instead of failing |
noresvport |
Reconnects from a new source port, so a reconnect isn’t held up |
rsize=1048576,wsize=1048576 |
Reads and writes in 1 MiB requests |
timeo=600,retrans=2 |
Waits 60 seconds before retrying a request, twice before logging that the server isn’t responding |
Never use soft: it turns a short wait into an I/O error in your application. nconnect=<n> (up to
16 connections) is fine.
Mount at boot
Add a line to /etc/fstab. _netdev waits for the network:
dfc6fl64neocrciu5ogrl4kccq.fs.us-west-1.ezghcompute.com:/fsy_4k2m9x0a7qpe /mnt/shared nfs4 nfsvers=4.1,xprtsec=tls,hard,noresvport,rsize=1048576,wsize=1048576,timeo=600,retrans=2,_netdev 0 0sudo mkdir -p /mnt/shared
sudo mount /mnt/sharedIn user data, cloud-init can do the same at first boot:
#cloud-config
packages:
- nfs-common
runcmd:
- mkdir -p /mnt/shared
- echo "dfc6fl64neocrciu5ogrl4kccq.fs.us-west-1.ezghcompute.com:/fsy_4k2m9x0a7qpe /mnt/shared nfs4 nfsvers=4.1,xprtsec=tls,hard,noresvport,rsize=1048576,wsize=1048576,timeo=600,retrans=2,_netdev 0 0" >> /etc/fstab
- mount /mnt/sharedThis mounts with TLS, which works as it is on the optimized images that support it.
On other images, set up TLS in the same user data first, or drop
xprtsec=tls.
The hostname resolves in public DNS. An instance that can’t reach the internet can’t resolve it:
mount by address, or, for TLS, add the name to /etc/hosts.
Allowed sources
A mount target’s allowedSourceCidrs are the IPv4 CIDRs that may mount through it, 1 to 50 of them.
Left out at create, they’re the VPC’s ranges. Change them with
UpdateMountTarget; the list replaces the previous one:
curl -X PATCH -H "Authorization: Bearer $EZGH_API_KEY" -H "Content-Type: application/json" \
-d '{"allowedSourceCidrs": ["10.0.1.0/24", "10.0.2.0/24"]}' \
"https://files-api.ezghcloud.com/v1/organizations/$ORG_ID/projects/$PROJECT_ID/mount-targets/fmt_7hq2m0x9ka3c"Firewall rules
Your instances’ firewalls must allow outbound TCP to port 2049 of the mount
target’s address. A firewall that allows all outbound traffic, as a new one does, needs nothing
more. No inbound rule is needed. A rule can name the mount target as its peer, { "mountTarget": "fmt_…" }, instead of
its address.
Mount targets
| Field | Meaning |
|---|---|
fileSystem |
Its file system (fsy_…). Fixed |
subnet |
The subnet its address is in (sbn_…), in the file system’s VPC. Fixed |
ipv4Address, hostname |
Its address and DNS name; null until it’s available. Fixed |
allowedSourceCidrs |
Who may mount through it |
A file system has at most one mount target per subnet; another is 409 mount_target_exists. A subnet
in another VPC is 409 vpc_mismatch. In a subnet with no free address, the create answers 202, its
operation fails with subnet_has_no_room, and the mount target ends failed: delete it, and create
one in a subnet with room.
Deleting a mount target (DeleteMountTarget) frees its address. The file system and its files stay; clients that mounted through it lose access until they mount through another one.