Can not GracefulShutdown application

[liaozan]

Can not GracefulShutdown application

[wilkinsona]

Thanks for the report. Unfortunately, as things stand, there are too many unknowns for us to efficiently investigate the problem. Can you please provide a complete example that reproduces it?

[liaozan]

I also wish I could provide an EXAMPLE, unfortunately this error it is not 100% reproducible.
As it stands, the problem is more likely to occur when the application is running a little longer. If it's just restarted right after a short period of time, it's hardly ever reproducible!

[liaozan]

I'll try to see if I can provide a 100% reproduction of the demo
Is there anything else I should provide other than a demo?

[wilkinsona]

I'm afraid we really need a demo as it's very hard to diagnose a problem that we cannot reproduce.

[liaozan]

Okay, I'll try to provide a minimal reproduction of the example

[liaozan]

I did my best to try to provide an example that could be reproduced consistently, but it is completely unreproducible on my own computer. But on CentOS 7.9, it will fail to terminate properly without any regularity

[liaozan]

I think my problem is very similar to #38930, the difference is that I'm using the latest SpringBoot 3.2.4

[wilkinsona]

Yes, it does look similar to #39930. Unfortunately, that doesn't help us here as no reproducer was provided there either. The class loading failures would suggest that either the class loader has been closed (that should not have happened at this point in the shutdown process) or the jar file has been deleted while the app is running. The latter's out of Spring Boot's control and, judging by your script above, should not be happening here. That leaves the class loader being closed too soon. To confirm or eliminate that possibility, I'm afraid we'll need something that reproduces the problem.

[jocull]

I am seeing a similar problem in an application that interrupts threads when workers need to rebalance. I am highly suspicious of #37668 but don't know where to start.

• We already migrated to JDK 17 and didn't have issues with Spring Boot 2.7.x
• When migrating to 3.2.x now we see ClassNotFoundException or NoClassDefFoundError, but this is highly correlated to the timings of the interrupts.
• Nothing very helpful shows in the logs, but the interrupts can be traced and we see these errors appear immediately after.

Is it possible that the interrupts are hitting the classloaders and breaking definition lookups? It would be a similar case to what is described here: https://stackoverflow.com/q/56866573/97964

How can I help trace this further?