[OPENJPA-2422] ClassMapping and it's strategy will never get initialized due to some synchronizatin/wrong initializaion order problems

• Type: Bug
• Status: Open
• Priority: Blocker
• Resolution: Unresolved
• Affects Version/s: 2.2.0, 2.2.1, 2.2.2
• Component/s: jpa, kernel
• Labels: ClassCastException ManyToOne OneToMany synchronization
• Environment: Apache ServiceMix 4.5.2
• Assignee: Unassigned
• Reporter: Piotr Klimczak
• Created: 19/Aug/13 16:30
• Updated: 14/Jan/14 20:48

Description

After I have switched my software to ServiceMix 4.5.2 I have faced very strange blocking issue.
I have bidirection relation: OneToMany and ManyToOne that was working fine before switching to ServiceMix 4.5.2.
After switching to ServiceMix 4.5.2 i am getting this Exception:

Caused by: java.lang.ClassCastException: org.apache.openjpa.util.LongId cannot be cast to XXXXXXXXXXXXXXX.PARENT
        at XXXXXXXXXXXXXXX.CHILD.pcReplaceField(CHILD.java)[227:XXXXXXXXXXXXXXX:4.0.1.SNAPSHOT]
        at org.apache.openjpa.kernel.StateManagerImpl.replaceField(StateManagerImpl.java:3213)
        at org.apache.openjpa.kernel.StateManagerImpl.storeObjectField(StateManagerImpl.java:2643)
        at org.apache.openjpa.kernel.StateManagerImpl.storeObject(StateManagerImpl.java:2633)
        at org.apache.openjpa.jdbc.kernel.JDBCStoreManager.setMappedBy(JDBCStoreManager.java:478)
        at org.apache.openjpa.jdbc.kernel.JDBCStoreManager.initializeState(JDBCStoreManager.java:404)
        at org.apache.openjpa.jdbc.kernel.JDBCStoreManager.initialize(JDBCStoreManager.java:306)
        at org.apache.openjpa.kernel.DelegatingStoreManager.initialize(DelegatingStoreManager.java:112)
        at org.apache.openjpa.kernel.ROPStoreManager.initialize(ROPStoreManager.java:57)
        at org.apache.openjpa.kernel.BrokerImpl.initialize(BrokerImpl.java:1046)
        at org.apache.openjpa.kernel.BrokerImpl.find(BrokerImpl.java:1004)
        ... 133 more

As I have done some debugging, it looks like as some synchronization problems in OpenJPA.
I'm getting this exception, because ClassMappings are constructed and putted into MODE_META state, before MetaDataRepository that owns them reaches MODE_MAPPING_INIT. So then when MetaDataRepository reaches MODE_MAPPING_INIT state, all ClassMappings are already in MODE_META state and will never be putted into MODE_MAPPING_INIT state as implemented in resolve method in MetaDataRepository:

    /**
     * Resolve the given metadata if needed. There are three goals:
     * <ol>
     * <li>Make sure no unresolved metadata gets back to the client.</li>
     * <li>Avoid infinite reentrant calls for mutually-dependent metadatas by allowing unresolved
     * metadata to be returned to other metadatas.</li>
     * <li>Always make sure the superclass metadata is resolved before the subclass metadata so that
     * the subclass can access the super's list of fields.</li>
     * </ol>
     * Note that the code calling this method is synchronized, so this method doesn't have to be.
     */
    private void resolve(ClassMetaData meta) {
        // return anything that has its metadata resolved, because that means
        // it is either fully resolved or must at least be in the process of
        // resolving mapping, etc since we do that right after meta resolve
        if (meta == null || _resMode == MODE_NONE || (meta.getResolve() & MODE_META) != 0)
            return;

So it makes ClassMapping's strategy impossible to initialize, once it reaches MODE_META state.

Due to above problem for example the RelationFieldStrategy will never get initialized, so then the FieldMetadata has _intermediate filed set to false which causes OpenJPA to set a LongId value to the field of type PARENT.

In my case, this is a blocker.
Unfortunately i am not able to write a testcase for this bug due to both: do not have a time for this and scenerio is way too complex- multithreaded app.

In my case problem occurs in aprox 95% app starts.

Tested on versions 2.2.2, 2.2.0, so same problems probably occurs on 2.2.1.
All test has been made with ServiceMix 4.5.2.