Question

A fleet of 200 instances in a private subnet uses a single Cloud NAT gateway to reach an external payment API. During peak traffic, application logs show many outbound connections to the same API endpoint failing intermittently. Cloud Monitoring shows the 'dropped_sent_packets' metric for the NAT gateway spiking, while CPU and bandwidth on the instances remain low. What is the most likely cause and the correct remediation?

Accepted Answer

NAT port exhaustion because too few ports are allocated per VM for many connections to the same destination; increase minimum ports per VM or enable Dynamic Port Allocation.

Answer

The NAT gateway has reached its bandwidth limit; add a second Cloud NAT gateway in the same region to load-balance egress.

Answer

MTU mismatch on the egress path; lower the interface MTU to 1460 on all instances to prevent fragmentation drops.

Answer

The default route to the internet gateway was deleted; recreate the 0.0.0.0/0 route pointing to the default internet gateway.