Is the deployment reasonable?

Note:
This topic has been translated from a Chinese forum by GPT and might contain errors.

Original topic: 部署是否合理

| username: neihan241

【TiDB Usage Environment】Production Environment
【TiDB Version】v5.4.0
Here is the instance:
image
Here is the server configuration:


Not sure if it’s reasonable.

The current issue is that it crashes from time to time. The company is considering upgrading the server. It’s a mixed deployment, and the crash issue has been addressed as needed, but there’s no improvement.

| username: Mark | Original post link

The mixed deployment performance in the picture is affected. Non-critical node services can be replaced with slightly lower-spec machines. TiKV and TiFlash should be deployed separately, and it is best to separate TiDB monitoring and similar services.

| username: wisdom | Original post link

It is recommended not to mix deployments. The number of nodes should be arranged according to the actual business scenario and data volume for better management.

| username: ddhe9527 | Original post link

Do not deploy TiKV, TiFlash, and TiDB together unless you have a thorough understanding of the memory and CPU resource usage and allocation for each component, and the parameter configuration is reasonable enough. Otherwise, with the default configuration, it is easy to encounter OOM (Out of Memory) issues or fully utilize the CPU.

| username: neihan241 | Original post link

Yes, now the CPUs of all three servers are fully utilized. I don’t understand resource allocation and I’m not a dedicated operations person. I just set it up at the time, and changing the configuration during OOM didn’t help.

| username: neihan241 | Original post link

Okay, thank you for the guidance.

| username: neihan241 | Original post link

It was initially a hybrid deployment, not sure how to change it.

| username: Mark | Original post link

You’re welcome. If the request volume for TiDB is not high, a single machine will suffice. If you are not using something like TIFlash, you can remove it for now.

| username: Kongdom | Original post link

Server reallocation can be achieved through scaling in and out.

| username: ablewang_xiaobo | Original post link

The number of machines is too small, so try to separate TiKV. You can put PD and TiDB together.

| username: ablewang_xiaobo | Original post link

You can add machines by scaling out, and then remove the roles from the original machines by scaling in.

| username: neihan241 | Original post link

Okay, thank you, I roughly understand.

| username: system | Original post link

This topic was automatically closed 1 minute after the last reply. No new replies are allowed.