After Downgrading from V7.6 to TiDB7.5, Overall Performance Has Significantly Decreased

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

Original topic: 从V7.6 退回到TiDB7.5 发现 整体性能都下降了很多

| username: 郑旭东石家庄

[TiDB Usage Environment] Production Environment
[TiDB Version]
[Reproduction Path]
[Encountered Problem:] All SQL executions are slow, creating a table takes 6 minutes, and creating an index on an empty table also takes 5 minutes.
[Resource Configuration] Go to TiDB Dashboard - Cluster Info - Hosts and take a screenshot of this page
[Attachments: Screenshots/Logs/Monitoring]


Attached is the architecture diagram and configuration

The hard disk is a 10,000 RPM mechanical drive. Previously, using V7.6 was fine, but now switching to the stable version 7.5.1, everything is not working.

| username: TiDBer_jYQINSnf | Original post link

Why switch to 7.5? If you can use 7.6, just use 7.6. I’ve encountered a situation where performance dropped significantly when upgrading from 4.0.16 to 6.5.2, but it got better after a few days.

| username: h5n1 | Original post link

Does “rollback” mean that a cluster was created in version 7.5.1 after having a cluster in version 7.6?

| username: zhanggame1 | Original post link

Once there is data scheduling on a mechanical disk, it becomes very slow.

| username: 托马斯滑板鞋 | Original post link

Please provide the dashboard SQL latency and host monitoring, especially for 50.24. I suspect resource contention. :upside_down_face:

| username: 蓝胖爸爸 | Original post link

May I ask how exactly the rollback is performed? Did you create an empty cluster?

| username: 郑旭东石家庄 | Original post link

Back up the data, uninstall 7.6, and reinstall 7.5.1.

| username: zhaokede | Original post link

Is this the same issue as the one in this post?

| username: 像风一样的男子 | Original post link

Version 7.6 is a development version, not an official release version, and it is not recommended for use in a production environment.

| username: DBAER | Original post link

You have both TiKV and TiFlash in this 50.24 setup, and the configuration is low. It feels like resource enforcement.

| username: paulli | Original post link

Compare the resource usage of the performance panels before and after.

| username: 郑旭东石家庄 | Original post link

It’s not the same issue.

| username: Kongdom | Original post link

Is it possible to single out a slow SQL in the dashboard and take a look? To see exactly which step is slow?
However, I tend to agree with the above analysis of resource contention.

| username: WalterWj | Original post link

It’s not possible to roll back, right? :thinking: Could it be that a new cluster was created and the data was exported and imported logically?

| username: 郑旭东石家庄 | Original post link

Yes, a full backup of the data, then delete the original cluster, create a new one, and import the data. With limited resources, we have no choice but to delete. If resources were sufficient, we could keep both and migration would be easier.

| username: 小于同学 | Original post link

Still not enough resources.

| username: dba远航 | Original post link

The same hard drive shouldn’t have such a big difference with just a minor version change. Could it be that version 7.6 has significant optimizations for hard drives?

| username: 郑旭东石家庄 | Original post link

From the perspective of SQL execution efficiency, version 7.6 is faster than version 7.5, especially when handling multiple tasks simultaneously.

| username: 郑旭东石家庄 | Original post link

Indeed, resources are a significant limitation.

| username: redgame | Original post link

Is it that obvious? I’ll try version 7.6.