[IGNITE-5041] NPE during deadlock detection because cache isn't started on node

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 1.7
• Fix Version/s: 2.0
• Component/s: cache
• Assignee: Andrey N. Gura
• Reporter: Andrey N. Gura
• Created: 20/Apr/17 12:46
• Updated: 26/Apr/17 16:15
• Resolved: 26/Apr/17 16:15

Description

NPE is possible during deadlock detection process in case when one of participating cache isn't started on node that initiates transaction:

Exception in thread "sys-#66%transactions.TxDeadlockNpeClientTest1%" java.lang.NullPointerException
	at org.apache.ignite.internal.processors.cache.transactions.IgniteTxKey.finishUnmarshal(IgniteTxKey.java:92)
	at org.apache.ignite.internal.processors.cache.transactions.TxLocksResponse.finishUnmarshal(TxLocksResponse.java:190)
	at org.apache.ignite.internal.processors.cache.transactions.IgniteTxManager$DeadlockDetectionListener.unmarshall(IgniteTxManager.java:2588)
	at org.apache.ignite.internal.processors.cache.transactions.IgniteTxManager$DeadlockDetectionListener.onMessage(IgniteTxManager.java:2478)
	at org.apache.ignite.internal.managers.communication.GridIoManager.invokeListener(GridIoManager.java:1082)
	at org.apache.ignite.internal.managers.communication.GridIoManager.processRegularMessage0(GridIoManager.java:710)
	at org.apache.ignite.internal.managers.communication.GridIoManager.access$1700(GridIoManager.java:102)
	at org.apache.ignite.internal.managers.communication.GridIoManager$5.run(GridIoManager.java:673)
	at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1145)
	at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:615)
	at java.lang.Thread.run(Thread.java:745)

Steps to reproduce:

1. Start server node with two caches.
2. Start two transactions on server node that involve both caches and lead to deadlock.
3. Start client node and create only one cache
4. Start transaction on client node that will try to update key that involved into deadlock

As result transaction on client node will hang until deadlock detection timeout happens.

The same result could be achived using node filters for caches.

Workaround:

Start all caches on node if possible.

Another way is deadlock detection disabling using JVM parameter -DIGNITE_TX_DEADLOCK_DETECTION_MAX_ITERS=0

Possible solution:

Try to create CacheObjectContext without starting of the cache.