bug: In version 3.6.0, a slow memory leak phenomenon occurs.

[wang243216388]

Current Behavior

During the use of version 3.6.0, it was discovered that there is a slow memory leak phenomenon, with a very small rate of increase but it keeps growing. I conducted a comparative test and observed between versions 3.6.0 and 3.8.0, and found that the memory usage in version 3.8.0 is normal, while in version 3.6.0, it continues to grow.

The image below illustrates the memory fluctuation trend of version 3.6.0.

The image below illustrates the memory fluctuation trend of version 3.8.0.

The image below displays the memory size of the worker process in version 3.6.0.

The image below displays the memory size of the worker process in version 3.8.0.

Why does the memory keep growing slightly in version 3.6.0?
How to investigate the specific content of the memory leak in version 3.6.0?

Expected Behavior

In version 3.6.0, there should be periodic memory release actions.

Error Logs

No response

Steps to Reproduce

Install versions 3.6.0 and 3.8.0 normally, use the same etcd cluster, add two routes, and perform external pressure testing with the same load.

Environment

NAME="EulerOS"
VERSION="2.0 (SP8)"
ID="euleros"
ID_LIKE="rhel fedora centos"
VERSION_ID="2.0"
PRETTY_NAME="EulerOS 2.0 (SP8)"
ANSI_COLOR="0;31"

Linux v3 4.19.36-vhulk1907.1.0.h1524.eulerosv2r8.aarch64

[wklken]

maybe relate to #10655

[wang243216388]

maybe relate to #10655

I haven't used any plugins. Are you saying that using the Prometheus plugin could cause it?

[wang243216388]

I referred to this #10614
This defect still exists in version 3.6.0 even after directly replacing the apisix/apisix/core/config_etcd.lua script with the one from version 3.8.0.
The image below shows the corresponding memory growth situation.