We have a Spring Boot application (webflux, r2dbc, spring data) using the version 2.5.3. After upgrading to 2.5.4 the application halts at startup. From the thread dump it seems that a thread is blocked during the initialization of AutoTimer in metrics.
"reactor-tcp-epoll-1" #32 daemon prio=5 os_prio=0 cpu=122,91ms elapsed=8,35s tid=0x00007f8404f851e0 nid=0xa098 waiting for monitor entry [0x00007f84747d8000]
java.lang.Thread.State: BLOCKED (on object monitor)
at org.springframework.beans.factory.support.DefaultSingletonBeanRegistry.getSingleton(DefaultSingletonBeanRegistry.java:217)
- waiting to lock <0x000000060ec1ef78> (a java.util.concurrent.ConcurrentHashMap)
at org.springframework.beans.factory.support.AbstractBeanFactory.doGetBean(AbstractBeanFactory.java:333)
at org.springframework.beans.factory.support.AbstractBeanFactory.getBean(AbstractBeanFactory.java:208)
at org.springframework.beans.factory.config.DependencyDescriptor.resolveCandidate(DependencyDescriptor.java:276)
at org.springframework.beans.factory.support.DefaultListableBeanFactory.doResolveDependency(DefaultListableBeanFactory.java:1380)
at org.springframework.beans.factory.support.DefaultListableBeanFactory$DependencyObjectProvider.getObject(DefaultListableBeanFactory.java:1979)
at org.springframework.boot.actuate.autoconfigure.metrics.data.RepositoryMetricsAutoConfiguration$$Lambda$840/0x00000008011e8b20.get(Unknown Source)
at org.springframework.util.function.SingletonSupplier.get(SingletonSupplier.java:97)
- locked <0x000000061115e180> (a org.springframework.util.function.SingletonSupplier)
at org.springframework.boot.actuate.metrics.data.MetricsRepositoryMethodInvocationListener.lambda$afterInvocation$0(MetricsRepositoryMethodInvocationListener.java:87)
at org.springframework.boot.actuate.metrics.data.MetricsRepositoryMethodInvocationListener$$Lambda$1194/0x0000000801315700.accept(Unknown Source)
at org.springframework.boot.actuate.metrics.AutoTimer.apply(AutoTimer.java:109)
at org.springframework.boot.actuate.metrics.data.MetricsRepositoryMethodInvocationListener.afterInvocation(MetricsRepositoryMethodInvocationListener.java:86)
at org.springframework.data.repository.core.support.RepositoryInvocationMulticaster$DefaultRepositoryInvocationMulticaster.notifyListeners(RepositoryInvocationMulticaster.java:76)
at org.springframework.data.repository.core.support.RepositoryMethodInvoker$ReactiveInvocationListenerDecorator.lambda$decorate$7(RepositoryMethodInvoker.java:248)
at org.springframework.data.repository.core.support.RepositoryMethodInvoker$ReactiveInvocationListenerDecorator$$Lambda$1047/0x0000000801233a30.apply(Unknown Source)
Drilling down a bit more, it seems that the problem is specifically with the AutoTimer support in spring data repositories, so if you disable that, the application starts normally.
management.metrics.data.repository.autotime.enabled= false
Unfortunatelly i could not reproduce that in a clean project with the same depedencies. Also the same issue appears in similar applications that have spring data but with cassandra connectivity this time.
Comment From: wilkinsona
Thanks for the report. Unfortunately, if you're unable to reproduce the problem with a complete application to base your attempt on, I think it's quite unlikely that we'll be able to figure it out from just the stack of a single thread.
Making a somewhat educated guess, it appears that you're using a Spring Data Repository during application startup which is unusual. If you'd like us to spend some more time investigating, a sample that reproduces the problem really would be ideal. Failing that, we might be able to figure things out from a full thread dump during the hang along with a detailed description of what the application does and how it uses Spring Data repositories during startup.
Comment From: dimcookies
Hi @wilkinsona , thank you for your prompt reply. You are right, the problem has to do with some caching that is performed during the startup. Based on that i was able to reproduce that in a clean project (attached). You can check the stacktrace of the thread dump and that if you change the version to 2.5.3 or disable autotime, the application starts successfully
The db schema: ``` CREATE TABLE configuration.config_param ( external_id character varying(100) NOT NULL, value character varying(100) NOT NULL, last_update timestamp without time zone ); ````
Comment From: wilkinsona
The change in behaviour in 2.5.4 is due to https://github.com/spring-projects/spring-boot/issues/27591.
This is a very similar to https://github.com/spring-projects/spring-framework/issues/20904. In this case it's the deferred bean lookup in MetricsRepositoryMethodInvocationListener that's the trigger rather than event publication, but it's essentially the same problem.
Generally speaking, you should avoid doing anything multi-threaded during application context refresh where there's any chance that one of those threads will interact with the bean factory. In this case, that includes calling a reactive Spring Data repository that performs network IO. You may want to consider using a CommandLineRunner rather than @PostConstruct or using a non-reactive repository for this sort of initialization during application context refresh.
We can take another look at the fix for #27591 as there was an alternative fix that we rejected at the time. I'm not sure if it'll fix this problem, and even if it did we may not want to switch to it, but it's worth considering at least.
Comment From: dimcookies
@wilkinsona thank you again for your enlightening answer. We will consider our options and delay these calls after startup has been completed
Comment From: wilkinsona
We discussed this today and decided that we don't want to risk destabilising things by trying to workaround something that is really a Framework problem. Using a CommandLineRunner or a non-reactive repository is your best bet for now.