Summary

  • AWS has extended synthetic-monitor health attribution to Transit Gateway inter-Region peering, with an explicit boundary at the peering connection.
  • A zero means no AWS-network degradation was observed within scope; it is not proof of application health or an instruction to fail over.

The expensive part of a network incident is often the hand-off: one team has latency, another has no alarm, and neither yet has grounds to act. AWS's September 10 addition to CloudWatch Network Monitoring gives that conversation another piece of evidence. Synthetic monitors can now publish a network health indicator, or NHI, for paths that cross Transit Gateway inter-Region peering. But the announcement draws a specific line: the indicator describes the AWS network path up to the peering connection. That is less than an assurance about the entire journey to a working application.

This is a change to synthetic monitoring, not the first appearance of cross-Region network visibility in CloudWatch. AWS added that capability to flow monitors in September 2025. Treating the two as interchangeable would obscure what an operator has actually deployed and what evidence an incident record contains.

The distinction matters most when the indicator is quiet. The synthetic-monitor guide defines two values: 100 for observed degradation within the AWS-controlled path and 0 when such degradation was not observed. These are not percentages of working customers or a count of failed requests. Nor does a zero settle the condition of everything outside the stated observation boundary.

The same guide describes ICMP echoes and TCP connection-handshake responses as the basis for latency and packet-loss measurements. A reply to a probe is useful transport evidence. It is not a completed checkout, database query or authenticated application session. If the customer experience remains poor, a team still needs application evidence and a path description that matches the affected workload. Otherwise, a clean-looking network panel can prematurely end the wrong investigation.

There is a timing qualification too. NHI can take a few hours to become available when a monitor is created, a probe is added or a probe is reactivated. A newly instrumented route therefore has a learning period, not a documented instant verdict. Recording an unavailable indicator as zero would erase the difference between lacking an observation and observing no degradation.

The commercial calculation is similarly narrower than “more visibility”. The pricing guide identifies monitored-resource hourly charges and CloudWatch metrics fees; source subnets form the resource coverage. Reducing that coverage may trim expenditure while removing evidence from a path that behaves differently during an incident. The relevant comparison is the cost of representative coverage against the decisions it improves, not the number of dashboard tiles purchased.

Finally, the service's limitations explicitly exclude automatic network failover. An alarm can shorten the route to an investigation; it does not supply a replacement path, approve a traffic move or demonstrate that the move is safe. No measured reduction in restoration time or guaranteed financial saving follows from the launch alone.