I have the following setup:

• 2 nodes in cluster with mode INVALIDATION_SYNC. No-transaction cache.
• Node1 is started
• Called "cache.remove" on some key on node1. At the same time, node2 is starting, which is causing topology change.
• The "cache.remove" call on node1 is throwing OutdatedTopologyException.

I found the cause is that StateTransferInterceptor is not added in InterceptorChain during INVALIDATION mode. It's just available during REPLICATION or DISTRIBUTED modes - https://github.com/infinispan/infinispan/blob/master/core/src/main/java/org/infinispan/factories/InterceptorChainFactory.java#L158

Indeed when I manually added StateTransferInterceptor to my invalidation cache:

invalidationConfigBuilder.customInterceptors()
    .addInterceptor()
    .before(NonTransactionalLockingInterceptor.class)
    .interceptorClass(StateTransferInterceptor.class);

I can see that issue is gone as OutdatedTopologyException is catched and command is retried with new topology.

I am attaching the Java unit test for reproducing issue. On my laptop when I run it, I can almost always simulate the issue.