
AEM is constantly evolving, offering more new features with each new release. But, with so many new features, sometimes new platform features can affect existing home-grown functionality, even when they don’t seem to be related to each other.
Here’s an example. One day, after migrating to AEM 6.4, one of our public services available at /services/acme/myservice.json
stopped working, and the following message appeared in the logs:
1 2 |
11.09.2018 17:59:24.403 *ERROR* [12.345.6.7 [1536688764343] GET /services/acme/myservice.json HTTP/1.1] org.apache.jackrabbit.oak.segment.SegmentNotFoundExceptionListener Segment not found: e851f378-0da9-4f99-a34b-2d2e036a40b1. SegmentId age=344865269ms,segment-generation=GCGeneration{generation=7,fullGeneration=2,isCompacted=false} org.apache.jackrabbit.oak.segment.SegmentNotFoundException: Segment e851f378-0da9-4f99-a34b-2d2e036a40b1 not found |
After some investigation it appeared that:
- The issue happens after AEM has performed an online revision cleanup
- The issue goes away after an AEM server restart
In fact a Sling Servlet behind /services/acme/myservice.json
was referencing and using the com.acme.www.service.impl.MyServiceImpl
service. We searched through the log files and found the following messages related to that MyServiceImpl
service:
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 |
11.09.2018 18:57:05.949 *WARN* [qtp1796865675-2034] org.apache.jackrabbit.oak.jcr.session.RefreshStrategy This session has been idle for 23 minutes and might be out of date. Consider using a fresh session or explicitly refresh the session. java.lang.Exception: The session was created here: at org.apache.jackrabbit.oak.jcr.session.RefreshStrategy$LogOnce.(RefreshStrategy.java:170) [org.apache.jackrabbit.oak-jcr:1.8.2] at org.apache.jackrabbit.oak.jcr.repository.RepositoryImpl.login(RepositoryImpl.java:285) [org.apache.jackrabbit.oak-jcr:1.8.2] at org.apache.jackrabbit.oak.jcr.repository.RepositoryImpl.login(RepositoryImpl.java:226) [org.apache.jackrabbit.oak-jcr:1.8.2] at org.apache.jackrabbit.oak.jcr.session.SessionImpl.impersonate(SessionImpl.java:275) [org.apache.jackrabbit.oak-jcr:1.8.2] at com.adobe.granite.repository.impl.CRX3SessionImpl.impersonate(CRX3SessionImpl.java:149) [com.adobe.granite.repository:1.4.88] at org.apache.sling.jcr.base.AbstractSlingRepository2.createServiceSession(AbstractSlingRepository2.java:201) [org.apache.sling.jcr.base:3.0.4] at com.adobe.granite.repository.impl.SlingRepositoryImpl.createServiceSession(SlingRepositoryImpl.java:135) [com.adobe.granite.repository:1.4.88] at org.apache.sling.jcr.base.AbstractSlingRepository2.createServiceSession(AbstractSlingRepository2.java:171) [org.apache.sling.jcr.base:3.0.4] at org.apache.sling.jcr.base.AbstractSlingRepository2.loginService(AbstractSlingRepository2.java:381) [org.apache.sling.jcr.base:3.0.4] at org.apache.sling.jcr.resource.internal.helper.jcr.JcrProviderStateFactory.createProviderState(JcrProviderStateFactory.java:116) [org.apache.sling.jcr.resource:3.0.8] at org.apache.sling.jcr.resource.internal.helper.jcr.JcrResourceProvider.authenticate(JcrResourceProvider.java:304) [org.apache.sling.jcr.resource:3.0.8] at org.apache.sling.jcr.resource.internal.helper.jcr.JcrResourceProvider.authenticate(JcrResourceProvider.java:76) [org.apache.sling.jcr.resource:3.0.8] at org.apache.sling.resourceresolver.impl.providers.stateful.ProviderManager.authenticate(ProviderManager.java:161) [org.apache.sling.resourceresolver:1.5.34] at org.apache.sling.resourceresolver.impl.providers.stateful.ProviderManager.getOrCreateProvider(ProviderManager.java:87) [org.apache.sling.resourceresolver:1.5.34] at org.apache.sling.resourceresolver.impl.providers.stateful.ProviderManager.authenticateAll(ProviderManager.java:129) [org.apache.sling.resourceresolver:1.5.34] at org.apache.sling.resourceresolver.impl.ResourceResolverImpl.createControl(ResourceResolverImpl.java:138) [org.apache.sling.resourceresolver:1.5.34] at org.apache.sling.resourceresolver.impl.ResourceResolverImpl.(ResourceResolverImpl.java:100) [org.apache.sling.resourceresolver:1.5.34] at org.apache.sling.resourceresolver.impl.ResourceResolverImpl.(ResourceResolverImpl.java:94) [org.apache.sling.resourceresolver:1.5.34] at org.apache.sling.resourceresolver.impl.CommonResourceResolverFactoryImpl.getResourceResolverInternal(CommonResourceResolverFactoryImpl.java:263) [org.apache.sling.resourceresolver:1.5.34] at org.apache.sling.resourceresolver.impl.ResourceResolverFactoryImpl.getServiceResourceResolver(ResourceResolverFactoryImpl.java:96) [org.apache.sling.resourceresolver:1.5.34] at com.acme.www.service.impl.MyServiceImpl.activate(MyServiceImpl.java:99) [com.acme.www.web-services:1.23.4.5] |
When we checked the MyServiceImpl
code, we saw this anti-pattern:
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 |
@Reference private ResourceResolverFactory resourceResolverFactory; @Activate protected void activate(final ComponentContext componentContext) { try { //Resource Resolver is created on a component activation this.resourceResolver = resourceResolverFactory.getServiceResourceResolver(resourceResolverParams); ... } catch (Exception e) { ... } } public void perform() { this.resourceResolver.resolve(...); } |
As you can see above, a resource resolver was created once upon component activation and was alive while the component was alive. That caused the idle session issue (see the stack trace above). The long-lived resource resolver was referencing a segment that was altered by online revision cleanup. Therefore, the subsequent calls of the perform()
method of MyServiceImpl
were causing the “Segment not found” error that we noted at the beginning of this post.
As a response, we followed the best practice of creating a resource resolver when it’s required and closing it when it’s not needed anymore. So, we refactored MyServiceImpl
to look like this:
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 |
@Reference private ResourceResolverFactory resourceResolverFactory; @Activate protected void activate(final ComponentContext componentContext) { //Resource Resolver creation was removed from here ... } private void perform() { try (ResourceResolver resourceResolver = resourceResolverFactory.getServiceResourceResolver(...)) { resourceResolver.resolve(...); } catch (Exception e) { ... } } |
For more information about handling the lifecycle of resource resolvers, please see our previous post on Creating a Helper to Close Sling Resource Resolver Instances.
Author: Denis Glushkov
Are you in IT?
Join our team!