Best Practices Guide for NFS on Debian/Ubuntu

Linux
Audience
Public
Product
FlashBlade
FlashArray
Technology Integrations
Linux
Source Type
Documentation

Architecture Overview

A production-grade NFS deployment consists of:

  • Debian/Ubuntu Hosts: Multiple hosts with redundant network connectivity

  • Dedicated Storage Network: Isolated network for NFS traffic

  • NFS Storage Array: Everpure FlashArray with NFSv4.1 support

  • Redundant Network Paths: Bonded NICs for failover

NFS Virtual IP (VIP) and Controller Failover

Everpure FlashArray uses a Virtual IP (VIP) for NFS services:

  1. Single Active Controller: The NFS VIP is hosted on one controller at a time

  2. Automatic Failover: If the active controller fails, the VIP migrates to the standby

  3. Client Transparency: NFS clients connect to the VIP-failover is transparent

  4. TCP Session Recovery: NFSv4.1 supports session recovery after failover

Why `hard` Mount is Critical

  • Soft mount: Failover > Timeout > I/O ERROR > Data corruption risk

  • Hard Mount: Failover > I/O pauses > VIP migrates > I/O resumes automatically

With soft mounts, if the VIP isn't available within the timeout, NFS returns errors to applications-this can corrupt databases, crash VMs, or lose data. Always use hard mounts for production.

Timeout Calculation

With timeo=300 (30 seconds) and retrans=2:

Phase Duration Cumulative
Initial request - 0s
First timeout 30s 30s
First retransmit + timeout 30s 60s
Second retransmit + timeout 30s 90s
Major timeout reached - ~90 seconds

After major timeout, the client enters exponential backoff but continues retrying indefinitely with hard mount. This ensures VMs survive even extended maintenance windows.

Note:

The 90-second window comfortably exceeds FlashArray's <30 second failover target, providing margin for network reconvergence.

Impact on VM I/O During Failover

Phase VM Behavior Duration
Pre-failover Normal I/O -
Failover in progress I/O pauses (queued in kernel) 10-30 seconds typical
Post-failover Queued I/O completes, normal operation resumes Automatic

What happens inside the VM:

  • Processes performing I/O will block (appear frozen)

  • No I/O errors are returned to applications

  • CPU-bound processes continue normally

  • Network connections unrelated to NFS are unaffected

  • After failover, all queued I/O completes in order

Guest OS behavior:

  • Linux guests: Processes in "D" (un-interruptible sleep) state temporarily

  • Windows guests: Applications may show "Not Responding" briefly

  • Databases: Transactions pause but do not fail (with hard mount)

Warning:

Using soft mounts converts the I/O pause into I/O errors, which can cause VM filesystem corruption, database crashes, or application failures.