Data Loss in Site to Site BGP Tunnel 1

If the ping works when you bring down one of the VPN tunnels (as you described in one of your replies), then the problem is likely related to what is called asymmetric routing. If briefly, it happens when your packets enter VPN Tunnel 1 on AWS side but exit VPN Tunnel 2 on AWS side, in which case your Customer Gateway (CGW) "sees" a ping reply coming from Tunnel 2 when there was no ping request sent through Tunnel 2. Some CGWs may drop such traffic depending on their configuration and capabilities. This behaviour is described in that article. It also has step-by-step instructions on how to check in CloudWatch if your traffic is routed asymmetrically.

When you turn one of the tunnels down, the CGW "sees" ping requests and replies exiting and entering the same tunnel, in which case the CGW may allow that traffic through depending on its configuration.

The first article is provided only to better describe the concept of asymmetric routing and how to check it in CloudWatch. The concept is the same for static and dynamic VPN. Because the first article is about static VPN, here is another article that talks about dynamic VPN. It describes how to configure a CGW to make AWS VPN endpoint(s) "prefer" a certain VPN tunnel when sending traffic from AWS to the CGW.

In summary, to avoid problems related to asymmetric routing, you can either configure your CGW to allow certain "reply" traffic through a tunnel where the CGW didn't "see" any "request” traffic; or you can make AWS VPN endpoint(s) "prefer" a certain tunnel when forwarding traffic to the CGW (as described in the second article) and at the same time make your CGW use the same tunnel to send traffic to AWS.