Serverless App Engine:Best practices for advanced parameters in an auto scaling policy

Log on to the PTS console and perform stress testing on the SAE application.

After the stress testing starts, you can view the auto scaling events of the SAE application.

Filter by Source Type and select auto scaling (HorizontalPodAutoscaler). You can view records with the reason SuccessfulRescale. Expand the details to see that the scale-out was triggered because the external metric slb_incall_qps exceeded the target value.

The auto scaling events show that the HPA controller has expanded the SAE application to 5 instances.

Configure the scale-out step size parameter in the metric-based auto scaling policy.

1. On the Basic Information page of the target application, click the Auto Scaling tab, and then click Edit in the Actions column for the added auto scaling policy.

2. In the Edit Auto Scaling Policy panel, click Advanced Settings to expand the advanced parameter configuration area.

3. In the Advanced Settings area, set scale-out step size to 1, leave the other advanced parameters unconfigured, and then click OK.

Use PTS to perform stress testing on the SAE application.

After the stress test starts, go to the Application Events page to view the auto scaling events for the target SAE application.

Auto scaling events show that instances in the SAE application are scaled out one by one, instead of scaling out to five instances at a time. This indicates that the Scale-out Step Size parameter is taking effect and limits the number of instances that are scaled out each time, thus achieving a balance between system stability and response speed by controlling the scale-out rhythm.

Log on to the Performance Testing Service (PTS) console and run a quick stress test on the target SAE application.

After the stress testing starts, you can compare the start time of the stress test and the scale-out time of the SAE application.

The test shows that without the scale-out stabilization window parameter configured, instances scaled out rapidly, about 20 seconds after the SAE application began handling traffic.

Configure the scale-out stabilization window parameter in the metric-based auto scaling policy.

1. On the Basic Information page of the target application, click the Auto Scaling tab, and then click Edit in the Actions column for the added auto scaling policy.

2. In the Edit Auto Scaling Policy panel, click Advanced Settings to expand the advanced parameter configuration area.

3. In the Advanced Settings area, set the scale-out stabilization window to 300 seconds, leave the other advanced parameters unconfigured, and then click OK.

Use the PTS tool to run a quick stress test on the target SAE application.

After the stress testing starts, you can compare the start time of the stress test and the scale-out time of the SAE application.

The test result shows that when the Scale-out Stabilization Window parameter is set to 300, an instance scale-out operation is not immediately triggered after the SAE application receives traffic. Instead, an instance scale-out operation starts after the specified stabilization window ends. This indicates that the Scale-out Stabilization Window parameter can effectively delay scale-out operations and avoid resource waste caused by short-term traffic fluctuations.

Log on to the Performance Testing Service (PTS) console and run a quick stress test on the target SAE application.

After the stress test is complete, go to the Application Events page to view the application's scaling events.

After filtering by Source Type for auto scaling (HorizontalPodAutoscaler), the event list shows two SuccessfulRescale events. The first was a scale-out to 5 instances during the test, triggered when the external metric slb_incall_qps exceeded its target. The second was a scale-in to 1 instance after the test, triggered when all metrics fell below their targets.

The test shows that if the disable scale-in parameter is not enabled, the application scales in to a single instance after the stress test completes.

In the metric-based auto scaling policy, enable disable scale-in.

1. On the Basic Information page of the target application, click the Auto Scaling tab, and then click Edit in the Actions column for the added auto scaling policy.

2. In the Advanced Settings area, enable disable scale-in.

Use the PTS tool to run a quick stress test on the target SAE application.

After the stress testing is complete (the traffic is 0), check whether application instances have been scaled in.

The instance list shows that all five instances are in the Running state. Four instances have been running for about 15 minutes (created during the scale-out), and one instance has been running for about 5 hours (the original instance). None of the instances were scaled in.

The test result shows that the number of application instances remains at 5 within about 5 minutes after the stress testing is complete, and no scale-in operation occurs. This indicates that the Disable Scale-in parameter has taken effect. This can effectively avoid excessive resource release caused by traffic fluctuations or misjudgments, thus ensuring service stability.