NFS 深入：server / client / K8s NFS / 经典坑

$ mount | grep nfs
nfs.example.com:/exports/data on /mnt/data type nfs4 (...,vers=4.2,...)
                                                    ^^^^^^^^^
                                                    协议版本

mount -t nfs -o vers=4.2 nfs.example.com:/exports/data /mnt/data

# 1. 装 NFS server
apt install -y nfs-kernel-server

# 2. 创建导出目录
mkdir -p /exports/data
chown nobody:nogroup /exports/data
chmod 0777 /exports/data                    # 简化（实际要细粒度）

# 3. 配置导出
cat > /etc/exports <<'EOF'
/exports/data 10.0.24.0/24(rw,sync,no_subtree_check,no_root_squash,fsid=0)
EOF

# 4. 应用
exportfs -ra
systemctl enable --now nfs-kernel-server

# 5. 验证
exportfs -v
showmount -e localhost

exportfs -v                       # 看当前 exports
exportfs -ra                       # reload 配置
exportfs -u <client>:<path>        # 卸下某客户端
exportfs -au                        # 全卸

# 本机
$ showmount -e localhost
Export list for localhost:
/exports/data    10.0.24.0/24

# 远端
$ showmount -e nfs.example.com

# 临时挂
mount -t nfs -o vers=4.2,rw,hard,timeo=600,retrans=2 \
  nfs.example.com:/exports/data /mnt/data

# 持久（/etc/fstab）
echo "nfs.example.com:/exports/data /mnt/data nfs vers=4.2,rw,hard,timeo=600,retrans=2,_netdev,nofail 0 0" \
  >> /etc/fstab

mount -t nfs -o \
  vers=4.2,proto=tcp,\
  hard,timeo=600,retrans=2,\
  rsize=1048576,wsize=1048576,\
  noatime,nodiratime,\
  _netdev,nofail \
  nfs.example.com:/exports/data /mnt/data

# 装 csi-driver-nfs
helm install csi-driver-nfs csi-driver-nfs/csi-driver-nfs \
  --namespace kube-system \
  --set kubeletDir=/var/lib/kubelet

$ kubectl apply -f pvc.yaml
# StorageClass 自动:
#   - 在 NFS server 的 /exports/data 下创建子目录 default-my-pvc/
#   - 创建 PV 包装这子目录
#   - bind PVC

# 节点上
$ ps aux | awk '$8 ~ /D/ {print}'
USER    PID  ... STAT  ...  COMMAND
root    1234 ... D     ...  java -jar app.jar
                     ^ 卡 D

# 看进程在等啥
$ cat /proc/1234/wchan
nfs_wait_on_request                          ← NFS 在等

$ umount /mnt/data
umount: /mnt/data: target is busy

$ umount -f /mnt/data
umount.nfs: ...  Stale file handle

$ umount -l /mnt/data                       # lazy umount
# 立即"断开"、但已经持有 fd 的进程继续等 NFS
# server 恢复后这些进程才能完成

mount -o rsize=1048576,wsize=1048576 ...

# /etc/sysctl.d/99-nfs.conf
sunrpc.tcp_max_slot_table_entries = 128       # 默认 65、调大并发
sunrpc.tcp_slot_table_entries = 128

# 应用
sysctl --system

nfsstat -m                                    # 看每个挂载点的统计
$ nfsstat -m
/mnt/data from nfs.example.com:/exports/data
  Flags: rw,relatime,vers=4.2,...
  ...

nfsstat -c                                    # client 操作统计
$ nfsstat -c
Client rpc stats:
calls      retrans    authrefrsh
123456     45         123456
            ^^^
            重传数大 → 网络问题

Client nfs v4:
null     read    write   commit  open  ...
0%       45%     30%     5%      2%   ...

$ ls /mnt/data
ls: cannot access '/mnt/data': Stale file handle

umount /mnt/data
mount -t nfs ... /mnt/data

umount -l /mnt/data
mount -t nfs ... /mnt/data

# 1. PV / PVC 状态
$ kubectl get pv,pvc
# Bound ?

# 2. NFS server 通吗
$ ssh m4 'showmount -e nfs.example.com'
Export list for nfs.example.com:
/exports/data    10.0.24.0/24

# 3. 节点能挂吗（手动测）
$ ssh m4 'mkdir -p /tmp/nfs-test && mount -t nfs nfs.example.com:/exports/data /tmp/nfs-test && ls /tmp/nfs-test'

# 4. K8s describe pod 看具体错
$ kubectl describe pod my-pod
Events:
  FailedMount  ...  mount failed: ...

# 5. kubelet 日志
$ ssh m4 'journalctl -u kubelet --since "5 min ago" | grep -i nfs'

# 1. 网络
$ mtr -rn -c 30 nfs.example.com

# 2. client 重传率
$ nfsstat -c | grep retrans

# 3. server 线程满
$ ssh nfs-server 'cat /proc/net/rpc/nfsd | head -1'
# th 8 1234 ...     ← 1234 次满载

# 4. mount options
$ mount | grep nfs
# 看 rsize / wsize / vers

# 5. server 磁盘 / CPU
$ ssh nfs-server 'iostat -x 1 5'

# 客户端
$ ls /mnt/data
ls: Stale file handle

# server 端
$ ssh nfs-server 'ls -li /exports/data'   # 看 inode 还在不

# 重新挂
$ umount -l /mnt/data
$ mount -t nfs ... /mnt/data