DistributedExecutorMassIndexer starts an IndexWorker for each indexed type and submits them to all the nodes in parallel. Each IndexWorker runs a blocking iteration over the cache, and when the cache has a store, that iteration is very expensive.

Each iteration will load and deserialize all the entries in the store. Most stores don't implement AbstractSegmentedStoreConfiguration, so Infinispan wraps them in a ComposedSegmentedLoadWriteStore, which iterates over each segment in parallel on the persistence/blocking executor. Since the default number of segments is 256 and the persistence/blocking executor has 4*cpu_count/150 max threads, it doesn't take a lot of parallel iterations to fill the blocking executor's threads and prevent the cache from doing other, more urgent, work.