Hi all. I ask for the help of specialists, as my head is already spinning. There are two Centos 7 servers, the Carp protocol is configured on them. It works and ip goes from one server to another when the main one goes down. On the local network everything works perfectly. The problem begins when there is a communication in other subnet. For clarity, I give an example:
Master:
VIP_ADDRESS="192.168.98.210"
ID=001
PASSFILE=/etc/ucarp/vip-001.pwd
BIND_INTERFACE="eth0"
SOURCE_ADDRESS="192.168.98.214"
OPTIONS="-k 2 --shutdown --preempt"

Slave:
VIP_ADDRESS="192.168.98.210"
ID=001
PASSFILE=/etc/ucarp/vip-001.pwd
BIND_INTERFACE="eth0"
SOURCE_ADDRESS="192.168.98.215"

Added net.ipv4.ip_nonlocal_bind=1 option to kernel

There is a server that has ip 192.168.120.15, we start ping it from 192.168.98.214 (Master)
10:38:38.315935 IP 192.168.98.214 > 192.168.121.15: ICMP echo request, id 17900, seq 2, length 64
10:38:38.315996 IP 192.168.121.15 > 192.168.98.214: ICMP echo reply, id 17900, seq 2, length 64

Trying a reverse ping to the cluster ip
10:40:26.227616 IP 192.168.121.15 > 192.168.98.210: ICMP echo request, id 17969, seq 2, length 64
10:40:26.227658 IP 192.168.98.210 > 192.168.121.15: ICMP echo reply, id 17969, seq 2, length 64

On reverse ping everything goes right. How to make it so that the routing would go from the cluster ip?????

I found a solution
iptables -t nat -A POSTROUTING -o eth0 -m state --state NEW ! -s 192.168.98.210 -j SNAT --to-source 192.168.98.210