Why Does the Traffic Throttling Count of a Cluster Show Non-Zero Values When the Cluster TPS Has Not Reached the Threshold?
Traffic throttling is implemented on a per-second basis, whereas the displayed average TPS is calculated on a per-minute basis. When a cluster experiences significant traffic spikes, it is possible for the TPS to exceed the threshold within a specific second during the minute, triggering traffic throttling, even though the minute-level average TPS does not reach the threshold. For more information, see Traffic Throttling Mechanism. To view second-level TPS metrics, change the statistical method from avg to max. If traffic throttling occurs only occasionally, you can increase the producer timeout period on the client side. If traffic throttling occurs frequently, it is recommended to evaluate whether the current cluster needs to be upgraded. Additionally, the default TPS allocation for cluster specifications assigns 50% to production and 50% to consumption. For example, a cluster with 1,000 TPS has default limits of 500 for production and 500 for consumption. If your usage scenario requires a different production-to-consumption ratio, you need to adjust the TPS allocation in the console.