kafka 流状态存储保存在哪里？

Question

I have a kafka streams application in which I am using state stores. I have read from a few places that state stores are backed in local rocksdb instance under /tmp directory. But in my application if I stop it and remove the application kafka stream directory from /tmp, I still am able to get my state store restored to previous state. Which means either state is not stored in rocksdb at all or it is stored in both kakfa and local rocksdb. So my question is, where is the state stored and since I am using docker, is there any performance or any other benefit of persisting /tmp/kafka-streams directory to a volume? My test code is below

StoreBuilder<KeyValueStore<String, String>> testStoreBuilder = Stores
            .keyValueStoreBuilder(Stores.persistentKeyValueStore("test-store"), Serdes.String(), Serdes.String())
            .withCachingEnabled();
    builder.addStateStore(testStoreBuilder);

    final KStream<String, String> sensorDataStream = builder.stream("test");

    sensorDataStream.transformValues(() -> new ValueTransformer<String, String>() {
        private KeyValueStore<String, String> testStore;

        @SuppressWarnings("unchecked")

        @Override
        public void init(ProcessorContext context) {
            testStore = (KeyValueStore<String, String>) context.getStateStore("test-store");
        }

        @Override
        public String transform(String value) {
            LocalDateTime start = LocalDateTime.now();
            System.out.println(testStore.get("test"));
            testStore.put("test", value);
            System.out.println("DURATION:: " + Duration.between(start, LocalDateTime.now()).toMillis());
            return null;
        }

        @Override
        public void close() {
        }
    }, "test-store");

Answer 1

When using Apache Kafka dependencies, then Streams are stored in /tmp . If using the Confluent dependencies, it's under /var/lib , I think.

RocksDB is backed by a consumed changelog topic that is stored in Kafka.

If you want clear all the state, use the application reset tool

since I am using docker

That shouldn't matter unless you remove/rebuild the image between application runs since Docker still writes to your host's filesystem