NFS Client Mount Options

Recommended NFS client mount options for the HA SAN, organized by network speed and workload type.

VIP Address

The NFS VIP (vip_nfs: 10.30.30.10) follows the active node via Pacemaker. Always mount using the VIP, not the node IP.

NFS Version

The server is configured for NFSv4.1 only (nfs_v41_enabled: true, nfs_v3_enabled: false). Mount with nfsvers=4.1 or nfsvers=4.

NFSv4.1 provides:

• pNFS (parallel NFS) for improved multi-client performance
• Sessions with connection multiplexing
• Delegations for client-side caching with server callbacks
• No portmapper dependency (simpler firewall rules)

Mount Options by Network Speed

1GbE (standard home/lab)

# /etc/fstab
10.30.30.10:/san-pool/nfs  /mnt/san  nfs  \
  nfsvers=4.1,rsize=131072,wsize=131072,hard,timeo=600,retrans=2,\
  _netdev,noatime,nordirplus  0 0

Key parameters:

• rsize=131072,wsize=131072 — 128KB I/O blocks (matches ZFS recordsize=128k)
• hard — keep retrying on timeout (prevents silent data loss on failover)
• timeo=600 — 60 second timeout before retry (1GbE is more latency-sensitive)
• retrans=2 — 2 retries before returning error to application
• noatime — skip access time updates (reduces write amplification)
• nordirplus — disable READDIRPLUS (reduces server load for directory-heavy workloads)

10GbE (moderate performance)

10.30.30.10:/san-pool/nfs  /mnt/san  nfs  \
  nfsvers=4.1,rsize=1048576,wsize=1048576,hard,timeo=150,retrans=3,\
  _netdev,noatime,proto=tcp  0 0

Key changes from 1GbE:

• rsize=1048576,wsize=1048576 — 1MB I/O blocks (saturates 10GbE better)
• timeo=150 — 15 second timeout (lower latency link)
• retrans=3 — more retries for intermittent issues
• proto=tcp — explicit TCP (NFSv4 always uses TCP, but explicit is clear)

25/40GbE (high performance)

10.30.30.10:/san-pool/nfs  /mnt/san  nfs  \
  nfsvers=4.1,rsize=1048576,wsize=1048576,hard,timeo=100,retrans=3,\
  _netdev,noatime,proto=tcp,nconnect=4  0 0

Key additions:

• nconnect=4 — use 4 TCP connections per mount (Linux 5.3+, maximizes bandwidth)
• timeo=100 — 10 second timeout (fast link, quick detection)

Workload-Specific Recommendations

VM/Container Storage (Proxmox, etc.)

10.30.30.10:/san-pool/nfs  /mnt/san  nfs  \
  nfsvers=4.1,rsize=1048576,wsize=1048576,hard,timeo=100,retrans=3,\
  _netdev,noatime,proto=tcp,async  0 0

• async — write-behind caching (improves VM performance; Proxmox handles fsync explicitly)
• For production VMs with databases: use sync instead of async

Database / Synchronous Workloads

10.30.30.10:/san-pool/nfs  /mnt/san  nfs  \
  nfsvers=4.1,rsize=131072,wsize=131072,hard,timeo=100,retrans=3,\
  _netdev,noatime,proto=tcp,sync  0 0

• sync — flush writes to server before returning (data safety for databases)
• Smaller block sizes match database page sizes (128KB for PostgreSQL, InnoDB)
• Consider using iSCSI instead of NFS for database workloads (lower overhead)

General File Sharing (Samba/CIFS is preferred)

For Windows clients and mixed environments, use the SMB VIP (10.30.30.11) instead of NFS.

Failover Behavior

During failover (node failure or pcs resource move):

1. NFS VIP migrates to the surviving node (~5-10 seconds)
2. NFS service restarts on the surviving node
3. NFS grace period allows clients to reclaim locks (nfs_grace_time: 45 seconds)
4. Clients with hard mount option will pause I/O and automatically reconnect
5. Clients with soft mount option may get I/O errors during failover — avoid soft

Estimated client reconnect time

Phase	Duration
Node failure detection (Corosync)	~4s
STONITH fence action	3-15s (depends on method)
ZFS pool import on surviving node	5-15s
NFS service start	2-5s
VIP assignment	1-2s
Total	~15-45s

With timeo=600 (60 seconds), 1GbE clients will typically not experience visible errors for short failovers.

Automount (systemd)

For systemd automount (lazy mount on access):

# /etc/systemd/system/mnt-san.automount
[Unit]
Description=NFS automount for SAN

[Automount]
Where=/mnt/san
TimeoutIdleSec=600

[Install]
WantedBy=multi-user.target

# /etc/systemd/system/mnt-san.mount
[Unit]
Description=NFS mount for SAN
After=network-online.target
Wants=network-online.target

[Mount]
What=10.30.30.10:/san-pool/nfs
Where=/mnt/san
Type=nfs
Options=nfsvers=4.1,rsize=1048576,wsize=1048576,hard,timeo=150,retrans=3,noatime

[Install]
WantedBy=multi-user.target

Troubleshooting

# Check mount options currently in use
mount | grep nfs
cat /proc/mounts | grep nfs

# Check NFS statistics
nfsstat -c    # client stats
nfsstat -s    # server stats (on storage node)

# Check NFS server exports
showmount -e 10.30.30.10

# Debug connection issues
tcpdump -n -i any host 10.30.30.10 and port 2049

# Check for stale file handle errors
dmesg | grep "NFS\|nfs"

# Force cache flush
echo 3 > /proc/sys/vm/drop_caches   # on client