dolphinscheduler icon indicating copy to clipboard operation
dolphinscheduler copied to clipboard

[Bug] jvm issue

Open nikusu417 opened this issue 2 months ago • 3 comments

Search before asking

  • [x] I had searched in the issues and found no similar issues.

What happened

Image

我们用mysql替代zk作为注册中心。替换后有6台master,其中一台master的jvm超过,基本每秒都会发生GC,同时在不停的创建新对象。排查下来后发现 如图的AbandonedConnectionCleanupThread持有的map过多

What you expected to happen

希望知道为什么会发生这种情况

How to reproduce

希望知道为什么会发生这种情况

Anything else

No response

Version

dev

Are you willing to submit PR?

  • [ ] Yes I am willing to submit a PR!

Code of Conduct

nikusu417 avatar Nov 13 '25 07:11 nikusu417

Search before asking

  • [x] I had searched in the issues and found no similar issues.

What happened

Image

We use mysql instead of zk as the registration center. After the replacement, there are 6 masters, and the jvm of one master exceeds the limit. GC will occur basically every second, and new objects will be created continuously. After troubleshooting, I found that the AbandonedConnectionCleanupThread in the picture holds too many maps.

What you expected to happen

Hope to know why this happens

How to reproduce

Hope to know why this happens

Anything else

No response

Version

dev

Are you willing to submit PR?

  • [ ] Yes I am willing to submit a PR!

Code of Conduct

github-actions[bot] avatar Nov 13 '25 07:11 github-actions[bot]

You can consider to increase the spring.datasource.hikari.idle-timeout value and observe again to see if there is still the same problem

Mrhs121 avatar Nov 13 '25 07:11 Mrhs121

@nikusu417 Full GC always on 1 of the 6 masters or randomly ? Full GC 是始终发生在其中某台master还是随机出现在各台master上?

qiong-zhou avatar Nov 18 '25 02:11 qiong-zhou

This issue has been automatically marked as stale because it has not had recent activity for 30 days. It will be closed in next 7 days if no further activity occurs.

github-actions[bot] avatar Dec 19 '25 00:12 github-actions[bot]