Queryable State Beta

    In a nutshell, this feature exposes Flink’s managed keyed (partitioned) state(see Working with State) to the outside world and allows the user to query a job’s state from outside Flink. For some scenarios, queryable state eliminates the need for distributed operations/transactions with external systems such as key-value stores which are often the bottleneck in practice. In addition, this feature may be particularly useful for debugging purposes.

    Attention: When querying a state object, that object is accessed from a concurrent thread without any synchronization or copying. This is a design choice, as any of the above would lead to increased job latency, which we wanted to avoid. Since any state backend using Java heap space, e.g. or FsStateBackend, does not work with copies when retrieving values but instead directly references the stored values, read-modify-write patterns are unsafe and may cause the queryable state server to fail due to concurrent modifications. The RocksDBStateBackend is safe from these issues.

    Before showing how to use the Queryable State, it is useful to briefly describe the entities that compose it.The Queryable State feature consists of three main entities:

    • the QueryableStateClient, which (potentially) runs outside the Flink cluster and submits the user queries,
    • the QueryableStateClientProxy, which runs on each TaskManager (i.e. inside the Flink cluster) and is responsible for receiving the client’s queries, fetching the requested state from the responsible Task Manager on his behalf, and returning it to the client, and
    • the QueryableStateServer which runs on each TaskManager and is responsible for serving the locally stored state.The client connects to one of the proxies and sends a request for the state associated with a specific key, k. As stated in , keyed state is organized in Key Groups, and each TaskManager is assigned a number of these key groups. To discover which TaskManager is responsible for the key group holding k, the proxy will ask the JobManager. Based on the answer, the proxy will then query the QueryableStateServer running on that TaskManager for the state associated with k, and forward theresponse back to the client.

    Activating Queryable State

    To enable queryable state on your Flink cluster, you need to do the following:

    • copy the flink-queryable-state-runtime_2.11-1.9.0.jar from the opt/ folder of your , to the lib/ folder.
    • set the property queryable-state.enable to true. See the Configuration documentation for details and additional parameters.To verify that your cluster is running with queryable state enabled, check the logs of any task manager for the line: "Started the Queryable State Proxy Server @ …".

    Now that you have activated queryable state on your cluster, it is time to see how to use it. In order for a state to be visible to the outside world, it needs to be explicitly made queryable by using:

    • either a QueryableStateStream, a convenience object which acts as a sink and offers its incoming values as queryablestate, or
    • the stateDescriptor.setQueryable(String queryableStateName) method, which makes the keyed state represented by the state descriptor, queryable.The following sections explain the use of these two approaches.

    Calling .asQueryableState(stateName, stateDescriptor) on a KeyedStream returns a QueryableStateStream which offersits values as queryable state. Depending on the type of state, there are the following variants of the asQueryableState()method:

    Note: There is no queryable ListState sink as it would result in an ever-growing list which may not be cleaned up and thus will eventually consume too much memory.

    1. stream.keyBy(0).asQueryableState("query-name")

    This acts like the Scala API’s flatMapWithState.

    Managed keyed state of an operator(see )can be made queryable by making the appropriate state descriptor queryable viaStateDescriptor.setQueryable(String queryableStateName), as in the example below:

    Note: The queryableStateName parameter may be chosen arbitrarily and is only used for queries. It does not have to be identical to the state's own name.

    This variant has no limitations as to which type of state can be made queryable. This means that this can be used for any ValueState, , ListState, MapState, AggregatingState, and the currently deprecated FoldingState.

    Querying State

    So far, you have set up your cluster to run with queryable state and you have declared (some of) your state asqueryable. Now it is time to see how to query this state.

    For this you can use the QueryableStateClient helper class. This is available in the flink-queryable-state-client jar which must be explicitly included as a dependency in the pom.xml of your project along with flink-core, as shown below:

    1. <dependency>
    2.   <artifactId>flink-core</artifactId>
    3.   <version>1.9.0</version>
    4. </dependency>
    5. <dependency>
    6.   <groupId>org.apache.flink</groupId>
    7.   <artifactId>flink-queryable-state-client-java</artifactId>
    8.   <version>1.9.0</version>
    9. </dependency>

    For more on this, you can check how to .

    With the client ready, to query a state of type V, associated with a key of type K, you can use the method:

    1. CompletableFuture<S> getKvState(
    2.     JobID jobId,
    3.     String queryableStateName,
    4.     K key,
    5.     TypeInformation<K> keyTypeInfo,
    6.     StateDescriptor<S, V> stateDescriptor)

    The above returns a CompletableFuture eventually holding the state value for the queryable state instance identified by queryableStateName of the job with ID jobID. The key is the key whose state you are interested in and the keyTypeInfo will tell Flink how to serialize/deserialize it. Finally, the stateDescriptor contains the necessary information about the requested state, namely its type (Value, Reduce, etc) and the necessary information on how to serialize/deserialize it.

    The careful reader will notice that the returned future contains a value of type S, i.e. a State object containingthe actual value. This can be any of the state types supported by Flink: , ReduceState, ListState, MapState,AggregatingState, and the currently deprecated FoldingState.

    Note: These state objects do not allow modifications to the contained state. You can use them to get the actual value of the state, e.g. using valueState.get(), or iterate over the contained <K, V> entries, e.g. using the mapState.entries(), but you cannot modify them. As an example, calling the add() method on a returned list state will throw an UnsupportedOperationException.

    Note: The client is asynchronous and can be shared by multiple threads. It needs to be shutdown via QueryableStateClient.shutdown() when unused in order to free resources.

    The following example extends the CountWindowAverage example(see Using Managed Keyed State)by making it queryable and shows how to query this value:

    Once used in a job, you can retrieve the job ID and then query any key’s current state from this operator:

    1. QueryableStateClient client = new QueryableStateClient(tmHostname, proxyPort);
    3. // the state descriptor of the state to be fetched.
    4. ValueStateDescriptor<Tuple2<Long, Long>> descriptor =
    5.         new ValueStateDescriptor<>(
    6.           "average",
    8. CompletableFuture<ValueState<Tuple2<Long, Long>>> resultFuture =
    9.         client.getKvState(jobId, "query-name", key, BasicTypeInfo.LONG_TYPE_INFO, descriptor);
    11. // now handle the returned value
    12. resultFuture.thenAccept(response -> {
    13.         try {
    14.             Tuple2<Long, Long> res = response.get();
    15.         } catch (Exception e) {
    16.             e.printStackTrace();
    17.         }
    18. });
    • queryable-state.server.ports: the server port range of the queryable state server. This is useful to avoid port clashes if more than 1 task managers run on the same machine. The specified range can be: a port: “9123”, a range of ports: “50100-50200”, or a list of ranges and or points: “50100-50200,50300-50400,51234”. The default port is 9067.
    • queryable-state.server.network-threads: number of network (event loop) threads receiving incoming requests for the state server (0 => #slots)
    • queryable-state.server.query-threads: number of threads handling/serving incoming requests for the state server (0 => #slots).
    • queryable-state.proxy.ports: the server port range of the queryable state proxy. This is useful to avoid port clashes if more than 1 task managers run on the same machine. The specified range can be: a port: “9123”, a range of ports: “50100-50200”,or a list of ranges and or points: “50100-50200,50300-50400,51234”. The default port is 9069.
    • queryable-state.proxy.network-threads: number of network (event loop) threads receiving incoming requests for the client proxy (0 => #slots)
    • queryable-state.proxy.query-threads: number of threads handling/serving incoming requests for the client proxy (0 => #slots).

    Limitations

    • The queryable state life-cycle is bound to the life-cycle of the job, e.g. tasks registerqueryable state on startup and unregister it on disposal. In future versions, it is desirable todecouple this in order to allow queries after a task finishes, and to speed up recovery via statereplication.
    • Notifications about available KvState happen via a simple tell. In the future this should be improved to bemore robust with asks and acknowledgements.