-
Type:
Bug
-
Resolution: Fixed
-
Priority:
Medium
-
Affects Version/s: 2.7.3, 2.8, 2.8.1, 2.8.2, 2.9, 2.9.1, 2.9.2
-
Component/s: None
-
Environment:
Noticed the problem in a 2 node cluster running 2.7.3
we are working on https://support.atlassian.com/browse/CSP-26648 and after examining the code of ConfluenceVelocityResourceCache found that there is a unchecked recursive loop:
- In getResource() if the holder exists but the resource is null, it will try to remove()
- What is the very first thing remove() does? you guested it
getResource()
I don't understand the logic, and I guess it is an very unlikely scenario, but it is happening (see https://support.atlassian.com/browse/CSP-26648).
Could somebody shed some light?
See code below:
public Resource remove(Object key)
{
runtimeServices.debug("Removing " + key + " from cache" );
try
{
Resource resource = getResource(key);
cache.remove(key);
return resource;
}
catch (Exception e)
{
log.error("Error removing cache key '" + key + "'", e);
}
return null;
}
...
private Resource getResource(Object key)
{
if ("true".equalsIgnoreCase(System.getProperty("atlassian.disable.caches")))
return null;
FakeSerializableResource holder = (FakeSerializableResource) cache.get(key);
if (holder == null)
return null;
Resource resource = holder.get();
if (resource == null)
{
log.warn("A FakeSerializableResource has been serialized");
remove(key);
}
return resource;
}
