Skip to content

Intermittent drop in NewFlowCount / ActiveFlowCount for Global Accelerator in NA

0

Hi, we observed a ~12-hour (October 11th) drop in Global Accelerator metrics for North America in our application:

  • NewFlowCount & ActiveFlowCount dropped ~80% and later recovered
  • ALB behind GA: NewConnectionCount drop similarly
  • BytesProcessedIn remained stable, showing only normal daily fluctuation
  • GA endpoint health remained healthy

Current setup routes traffic either to NA or EU; EU remained unchanged. We checked AWS Health Dashboard — no events were reported. External routing (Cloudflare Radar) shows no ISP outages in NA.

Does anyone have guidance on what could cause fewer new connections reaching GA/ALB while throughput per session remains steady? Are there known GA edge routing or network-layer events that might not trigger Health Dashboard notices?

1 Answer
0

Based on your description, there are several possible explanations for the observed drop in NewFlowCount and ActiveFlowCount while BytesProcessedIn remained stable.

First, this pattern suggests that fewer new connections were being established, but existing connections were transferring more data per connection. This could happen if:

  1. AWS Global Accelerator was performing maintenance events. Global Accelerator occasionally undergoes maintenance that can affect TCP connections and potentially cause TCP resets, which might explain the drop in new connections while existing ones continued to function.

  2. There might have been health check or failover behavior occurring. Global Accelerator continuously monitors endpoint health and automatically redirects traffic when endpoints are unhealthy. If there were intermittent health issues with your NA endpoints, Global Accelerator might have attempted to failover traffic to other regions, even though the endpoints appeared healthy in your monitoring.

  3. TCP termination issues could be involved. When Global Accelerator sets up connections between clients and regional endpoints with TCP termination, there are two segments involved: one between the client and the Global Accelerator POP location, and another between the POP and your regional endpoint. Issues in either segment could cause connection problems without affecting overall throughput.

To investigate further, I recommend:

  • Check CloudWatch metrics for TCP reset counts (TCP_AGA_Reset_Count, TCP_Client_Reset_Count, and TCP_Endpoint_Reset_Count) during the affected period
  • Review flow logs if you have them enabled to see detailed traffic patterns
  • Examine if there were any changes to traffic dials or endpoint weights that might have redirected traffic
  • Monitor for any patterns in connection establishment failures

All CloudWatch metrics for Global Accelerator must be viewed in the US West (Oregon) Region, so ensure you're checking metrics in that region.
Sources
How AWS Global Accelerator works - AWS Global Accelerator
Troubleshooting Global Accelerator TCP reset issues - AWS Global Accelerator
Logging and monitoring in AWS Global Accelerator - AWS Global Accelerator

answered a year ago

You are not logged in. Log in to post an answer.

A good answer clearly answers the question and provides constructive feedback and encourages professional growth in the question asker.