Replication Lag on AWS FSx: The Hidden EC2 Single-Flow Bandwidth Limit

A PostgreSQL standby lagging behind its primary turned out to be caused not by the FSx storage layer but by an EC2 network limitation. Even though FSx and EC2 metrics looked healthy at 50% capacity, vmstat and nfsiostat showed sustained I/O wait and queue time. Digging into per-second throughput distributions revealed traffic capped at a consistent ~5 Gbit/s ceiling, matching AWS's documented single-flow (single TCP connection) bandwidth limit that applies regardless of an instance's aggregate network bandwidth. The NFS mount lacked the nconnect option, so all traffic rode a single TCP connection. Setting nconnect=16 resolved the lag, though the investigation suggests a much smaller value (2-3 connections) would have sufficed given FSx's 10.2 Gbps provisioned throughput and the EC2 instance's 25 Gbps aggregate limit.

Source: https://www.percona.com/blog/replication-lag-on-aws-fsx-the-hidden-ec2-single-flow-bandwidth-limit. 8 Sync News only summarizes and links out; content copyright belongs to the authors and original sources.