The cluster size is 400 applications, 4,000 pods, and 3 billion segments are generated every day. Looking for advice.

[liuxinagxiang]

My project generates about 3 billion segments a day, and the current architecture is skywalking-agent --> kafka ---> skywalking-L1-L2 ---> ES.

The scale is as follows: 400 microservices, nearly 4,000 pods, three Kafka nodes, skywalking-L1 6 nodes: Xmx8G Xms8G Xmn3G, skywalking-L2 6 nodes: Xmx8G Xms8G Xmn3G, elasticsearch 6 nodes: Xmx16G Xms16G Xmn6G
#kafka3.7.0 skywalking9.2 elasticsearch7.17.18

The current problem encountered is that kafka cluster and elaticsearch cluster are normal but skywalking L1 and L2 clusters continue to report errors,skywalking oap service consumption is very slow . After each restart, kafka data is consumed normally within ten minutes and various timeout errors are reported. so I have moved L2 has been moved to k8s cluster container deployment to avoid manually restarting oap every time 😅

I want to know how to plan the cluster according to the size of my project. Which configurations can be optimized? Does anyone have any suggestions?

I was referring to this article recently: https://skywalking.apache.org/zh/2022-08-30-pingan-jiankang/

[wu-sheng]

I want to know how to plan the cluster according to the size of my project. Which configurations can be optimized? Does anyone have any suggestions?

Elasticsearch healthy doesn't mean it is powerful enough. Check self-observability data, especially OAP flush metrics. I believe it is too slow, then everything goes to be blocked eventually.

[linshenfanying]

Is your SkyWalking deployed in Kubernetes? If so, could you provide the relevant documentation? I'm also making relevant modifications recently.