Best Practices for an Amazon Managed Service for Apache Flink - Amazon Kinesis Data Analytics

Best Practices for an Amazon Managed Service for Apache Flink

This section contains information and recommendations for developing stable, performant Amazon Kinesis Data Analytics applications.

Fault tolerance: checkpoints and savepoints

Use checkpoints and savepoints to implement fault tolerance in your an Amazon Managed Service for Apache Flink application. Keep the following in mind when developing and maintaining your application:

  • We recommend that you leave checkpointing enabled for your application. Checkpointing provides fault tolerance for your application during scheduled maintenance, as well as in case of unexpected failures due to service issues, application dependency failures, and other issues. For information about scheduled maintenance, see an Amazon Managed Service for Apache Flink Maintenance.

  • Set ApplicationSnapshotConfiguration::SnapshotsEnabled to false during application development or troubleshooting. A snapshot is created during every application stop, which may cause issues if the application is in an unhealthy state or isn't performant. Set SnapshotsEnabled to true after the application is in production and is stable.

    Note

    We recommend that your application create a snapshot several times a day to restart properly with correct state data. The correct frequency for your snapshots depends on your application's business logic. Taking frequent snapshots allows you to recover more recent data, but increases cost and requires more system resources.

    For information about monitoring application downtime, see Metrics and Dimensions.

For more information about implementing fault tolerance, see Fault Tolerance.

Performance and parallelism

Your application can scale to meet any throughput level by tuning your application parallelism, and avoiding performance pitfalls. Keep the following in mind when developing and maintaining your application:

  • Verify that all of your application sources and sinks are sufficiently provisioned and are not being throttled. If the sources and sinks are other AWS services, monitor those services using CloudWatch.

  • For applications with very high parallelism, check if the high levels of parallelism are applied to all operators in the application. By default, Apache Flink applies the same application parallelism for all operators in the application graph. This can lead to either provisioning issues on sources or sinks, or bottlenecks in operator data processing. You can change the parallelism of each operator in code with setParallelism.

  • Understand the meaning of the parallelism settings for the operators in your application. If you change the parallelism for an operator, you may not be able to restore the application from a snapshot created when the operator had a parallelism that is incompatible with the current settings. For more information about setting operator parallelism, see Set maximum parallelism for operators explicitly.

For more information about implementing scaling, see Scaling Production readiness.

Logging

You can monitor your application's performance and error conditions using CloudWatch Logs. Keep the following in mind when configuring logging for your application:

  • Enable CloudWatch logging for the application so that any runtime issues can be debugged.

  • Do not create a log entry for every record being processed in the application. This causes severe bottlenecks during processing and might lead to backpressure in processing of data.

  • Create CloudWatch alarms to notify you when your application is not running properly. For more information, see Alarms

For more information about implementing logging, see Logging and Monitoring.

Coding

You can make your application performant and stable by using recommended programming practices. Keep the following in mind when writing application code:

  • Do not use system.exit() in your application code, in either your application's main method or in user-defined functions. If you want to shut down your application from within code, throw an exception derived from Exception or RuntimeException, containing a message about what went wrong with the application.

    Note the following about how the service handles this exception:

    • If the exception is thrown from your application's main method, the service will wrap it in a ProgramInvocationException when the application transitions to the RUNNING status, and the job manager will fail to submit the job.

    • If the exception is thrown from a user-defined function, the job manager will fail the job and restart it, and details of the exception will be written to the exception log.

  • Consider shading your application JAR file and its included dependencies. Shading is recommended when there are potential conflicts in package names between your application and the Apache Flink runtime. If a conflict occurs, your application logs may contain an exception of type java.util.concurrent.ExecutionException. For more information about shading your application JAR file, see Apache Maven Shade Plugin.

Managing credentials

You should not bake any long-term credentials into production (or any other) applications. Long-term credentials are likely checked into a version control system and can easily get lost. Instead, you can associate a role to the Kinesis Data Analytics application and grant privileges to that role. The running Flink application can then pick up temporary credentials with the respective privileges from the environment. In case authentication is needed for a service that is not natively integrated with IAM, e.g., a database that requires a username and password for authentication, you should consider storing secrets in AWS Secrets Manager.

Many AWS native services support authentication:

Reading from sources with few shards/partitions

When reading from Apache Kafka or a Kinesis Data Stream, there may be a mismatch between the parallelism of the stream (i.e., the number of partitions for Kafka and the number of shards for Kinesis) and the parallelism of the application. With a naive design, the parallelism of an application cannot scale beyond the parallelism of a stream: Each subtask of a source operator can only read from 1 or more shards/partitions. That means for a stream with only 2 shards and an application with a parallelism of 8, that only two subtasks are actually consuming from the stream and 6 subtasks remain idle. This can substantially limit the throughput of the application, in particular if the deserialization is expensive and carried out by the source (which is the default).

To mitigate this effect, you can either scale the stream. But that may not always be desirable or possible. Alternatively, you can restructure the source so that it does not do any serialization and just passes on the byte[]. You can then rebalance the data to distribute it evenly across all tasks and then deserialize the data there. In this way, you can leverage all subtasks for the deserialization and this potentially expensive operation is no longer bound by the number of shards/partitions of the stream.

Studio notebook refresh interval

If you change the paragraph result refresh interval, set it to a value that is at least 1000 milliseconds.

Studio notebook optimum performance

We tested with the following statement and got the best performance when events-per-second multiplied by number-of-keys was under 25,000,000. This was for events-per-second under 150,000.

SELECT key, sum(value) FROM key-values GROUP BY key

How watermark strategies and idle shards affect time windows

When reading events from Apache Kafka and Kinesis Data Streams, the source can set the event time based on attributes of the stream. In case of Kinesis, the event time equals the approximate arrival time of events. But setting event time at the source for events is not sufficient for a Flink application to use event time. The source must also generate watermarks that propagate information about event time from the source to all other operators. The Flink documentation has a good overview of how that process works.

By default, the timestamp of an event read from Kinesis is set to the approximate arrival time determined by Kinesis. An additional prerequisite for event time to work in the application is a watermark strategy.

WatermarkStrategy<String> s = WatermarkStrategy .<String>forMonotonousTimestamps() .withIdleness(Duration.ofSeconds(...));

The watermark strategy is then applied to a DataStream with the assignTimestampsAndWatermarks method. There are some useful build-in strategies:

  • forMonotonousTimestamps() will just use the event time (approximate arrival time) and periodically emit the maximum value as a watermark (for each specific subtask)

  • forBoundedOutOfOrderness(Duration.ofSeconds(...)) similar to the previous strategy, but will use the event time – duration for watermark generation.

This works, but there are a couple of caveats to be aware of. Watermarks are generated at a subtask level and flow through the operator graph.

From the Flink documentation:

Each parallel subtask of a source function usually generates its watermarks independently. These watermarks define the event time at that particular parallel source.

As the watermarks flow through the streaming program, they advance the event time at the operators where they arrive. Whenever an operator advances its event time, it generates a new watermark downstream for its successor operators.

Some operators consume multiple input streams; a union, for example, or operators following a keyBy(…) or partition(…) function. Such an operator’s current event time is the minimum of its input streams' event times. As its input streams update their event times, so does the operator.

That means, if a source subtask is consuming from an idle shard, downstream operators do not receive new watermarks from that subtask and hence processing stalls for all downstream operators that use time windows. To avoid this, customers can add the withIdleness option to the watermark strategy. With that option, an operator excludes the watermarks from idle upsteam subtasks when computing the event time of the operator. Idle subtask therefor no longer block the advancement of event time in downstream operators.

However, the idleness option with the build-in watermark strategies will not advance the event time if no subtask is reading any event, i.e., there are no events in the stream. This becomes particularly visible for test cases where a finite set of events is read from the stream. As event time does not advance after the last event has been read, the last window (containing the last event) will never close.

Summary

  • the withIdleness setting will not generate new watermarks in case a shard is idle, it will just exclude the last watermark sent by idle subtasks from the min watermark calculation in downstream operators

  • with the build-in watermark strategies the last open window will never close (unless new events that advance the watermark will be sent, but that creates a new window that then remains open)

  • even when the time is set by the Kinesis stream, late arriving events can still happen if one shard is consumed faster than others (eg, during app initialization or when using TRIM_HORIZON where all existing shards are consumed in parallel ignoring their parent/child relationship)

  • the withIdleness settings of the watermark strategy seem to deprecate the the Kinesis source specific settings for idle shards (ConsumerConfigConstants.SHARD_IDLE_INTERVAL_MILLIS

Example

The following application is reading from a stream and creating session windows based on event time.

Properties consumerConfig = new Properties(); consumerConfig.put(AWSConfigConstants.AWS_REGION, "eu-west-1"); consumerConfig.put(ConsumerConfigConstants.STREAM_INITIAL_POSITION, "TRIM_HORIZON"); FlinkKinesisConsumer<String> consumer = new FlinkKinesisConsumer<>("...", new SimpleStringSchema(), consumerConfig); WatermarkStrategy<String> s = WatermarkStrategy .<String>forMonotonousTimestamps() .withIdleness(Duration.ofSeconds(15)); env.addSource(consumer) .assignTimestampsAndWatermarks(s) .map(new MapFunction<String, Long>() { @Override public Long map(String s) throws Exception { return Long.parseLong(s); } }) .keyBy(l -> 0l) .window(EventTimeSessionWindows.withGap(Time.seconds(10))) .process(new ProcessWindowFunction<Long, Object, Long, TimeWindow>() { @Override public void process(Long aLong, ProcessWindowFunction<Long, Object, Long, TimeWindow>.Context context, Iterable<Long>iterable, Collector<Object> collector) throws Exception { long count = StreamSupport.stream(iterable.spliterator(), false).count(); long timestamp = context.currentWatermark(); System.out.print("XXXXXXXXXXXXXX Window with " + count + " events"); System.out.println("; Watermark: " + timestamp + ", " + Instant.ofEpochMilli(timestamp)); for (Long l : iterable) { System.out.println(l); } } });

In the following example, 8 events are written to a 16 shard stream (the first 2 and the last event happen to land in the same shard).

$ aws kinesis put-record --stream-name hp-16 --partition-key 1 --data MQ== $ aws kinesis put-record --stream-name hp-16 --partition-key 2 --data Mg== $ aws kinesis put-record --stream-name hp-16 --partition-key 3 --data Mw== $ date { "ShardId": "shardId-000000000012", "SequenceNumber": "49627894338614655560500811028721934184977530127978070210" } { "ShardId": "shardId-000000000012", "SequenceNumber": "49627894338614655560500811028795678659974022576354623682" } { "ShardId": "shardId-000000000014", "SequenceNumber": "49627894338659257050897872275134360684221592378842022114" } Wed Mar 23 11:19:57 CET 2022 $ sleep 10 $ aws kinesis put-record --stream-name hp-16 --partition-key 4 --data NA== $ aws kinesis put-record --stream-name hp-16 --partition-key 5 --data NQ== $ date { "ShardId": "shardId-000000000010", "SequenceNumber": "49627894338570054070103749783042116732419934393936642210" } { "ShardId": "shardId-000000000014", "SequenceNumber": "49627894338659257050897872275659034489934342334017700066" } Wed Mar 23 11:20:10 CET 2022 $ sleep 10 $ aws kinesis put-record --stream-name hp-16 --partition-key 6 --data Ng== $ date { "ShardId": "shardId-000000000001", "SequenceNumber": "49627894338369347363316974173886988345467035365375213586" } Wed Mar 23 11:20:22 CET 2022 $ sleep 10 $ aws kinesis put-record --stream-name hp-16 --partition-key 7 --data Nw== $ date { "ShardId": "shardId-000000000008", "SequenceNumber": "49627894338525452579706688535878947299195189349725503618" } Wed Mar 23 11:20:34 CET 2022 $ sleep 60 $ aws kinesis put-record --stream-name hp-16 --partition-key 8 --data OA== $ date { "ShardId": "shardId-000000000012", "SequenceNumber": "49627894338614655560500811029600823255837371928900796610" } Wed Mar 23 11:21:27 CET 2022

This input should result in 5 session windows: event 1,2,3; event 4,5; event 6; event 7; event 8. However, the program only yields the first 4 windows.

11:59:21,529 INFO org.apache.flink.streaming.connectors.kinesis.FlinkKinesisConsumer [] - Subtask 5 will be seeded with initial shard StreamShardHandle{streamName='hp-16', shard='{ShardId: shardId-000000000006,HashKeyRange: {StartingHashKey: 127605887595351923798765477786913079296,EndingHashKey: 148873535527910577765226390751398592511},SequenceNumberRange: {StartingSequenceNumber: 49627894338480851089309627289524549239292625588395704418,}}'}, starting state set as sequence number EARLIEST_SEQUENCE_NUM 11:59:21,530 INFO org.apache.flink.streaming.connectors.kinesis.internals.KinesisDataFetcher [] - Subtask 5 will start consuming seeded shard StreamShardHandle{streamName='hp-16', shard='{ShardId: shardId-000000000006,HashKeyRange: {StartingHashKey: 127605887595351923798765477786913079296,EndingHashKey: 148873535527910577765226390751398592511},SequenceNumberRange: {StartingSequenceNumber: 49627894338480851089309627289524549239292625588395704418,}}'} from sequence number EARLIEST_SEQUENCE_NUM with ShardConsumer 0 11:59:21,530 INFO org.apache.flink.streaming.connectors.kinesis.FlinkKinesisConsumer [] - Subtask 6 will be seeded with initial shard StreamShardHandle{streamName='hp-16', shard='{ShardId: shardId-000000000007,HashKeyRange: {StartingHashKey: 148873535527910577765226390751398592512,EndingHashKey: 170141183460469231731687303715884105727},SequenceNumberRange: {StartingSequenceNumber: 49627894338503151834508157912666084957565273949901684850,}}'}, starting state set as sequence number EARLIEST_SEQUENCE_NUM 11:59:21,530 INFO org.apache.flink.streaming.connectors.kinesis.FlinkKinesisConsumer [] - Subtask 6 will be seeded with initial shard StreamShardHandle{streamName='hp-16', shard='{ShardId: shardId-000000000010,HashKeyRange: {StartingHashKey: 212676479325586539664609129644855132160,EndingHashKey: 233944127258145193631070042609340645375},SequenceNumberRange: {StartingSequenceNumber: 49627894338570054070103749782090692112383219034419626146,}}'}, starting state set as sequence number EARLIEST_SEQUENCE_NUM 11:59:21,530 INFO org.apache.flink.streaming.connectors.kinesis.internals.KinesisDataFetcher [] - Subtask 6 will start consuming seeded shard StreamShardHandle{streamName='hp-16', shard='{ShardId: shardId-000000000007,HashKeyRange: {StartingHashKey: 148873535527910577765226390751398592512,EndingHashKey: 170141183460469231731687303715884105727},SequenceNumberRange: {StartingSequenceNumber: 49627894338503151834508157912666084957565273949901684850,}}'} from sequence number EARLIEST_SEQUENCE_NUM with ShardConsumer 0 11:59:21,531 INFO org.apache.flink.streaming.connectors.kinesis.FlinkKinesisConsumer [] - Subtask 4 will be seeded with initial shard StreamShardHandle{streamName='hp-16', shard='{ShardId: shardId-000000000005,HashKeyRange: {StartingHashKey: 106338239662793269832304564822427566080,EndingHashKey: 127605887595351923798765477786913079295},SequenceNumberRange: {StartingSequenceNumber: 49627894338458550344111096666383013521019977226889723986,}}'}, starting state set as sequence number EARLIEST_SEQUENCE_NUM 11:59:21,532 INFO org.apache.flink.streaming.connectors.kinesis.internals.KinesisDataFetcher [] - Subtask 4 will start consuming seeded shard StreamShardHandle{streamName='hp-16', shard='{ShardId: shardId-000000000005,HashKeyRange: {StartingHashKey: 106338239662793269832304564822427566080,EndingHashKey: 127605887595351923798765477786913079295},SequenceNumberRange: {StartingSequenceNumber: 49627894338458550344111096666383013521019977226889723986,}}'} from sequence number EARLIEST_SEQUENCE_NUM with ShardConsumer 0 11:59:21,532 INFO org.apache.flink.streaming.connectors.kinesis.FlinkKinesisConsumer [] - Subtask 3 will be seeded with initial shard StreamShardHandle{streamName='hp-16', shard='{ShardId: shardId-000000000004,HashKeyRange: {StartingHashKey: 85070591730234615865843651857942052864,EndingHashKey: 106338239662793269832304564822427566079},SequenceNumberRange: {StartingSequenceNumber: 49627894338436249598912566043241477802747328865383743554,}}'}, starting state set as sequence number EARLIEST_SEQUENCE_NUM 11:59:21,532 INFO org.apache.flink.streaming.connectors.kinesis.FlinkKinesisConsumer [] - Subtask 2 will be seeded with initial shard StreamShardHandle{streamName='hp-16', shard='{ShardId: shardId-000000000003,HashKeyRange: {StartingHashKey: 63802943797675961899382738893456539648,EndingHashKey: 85070591730234615865843651857942052863},SequenceNumberRange: {StartingSequenceNumber: 49627894338413948853714035420099942084474680503877763122,}}'}, starting state set as sequence number EARLIEST_SEQUENCE_NUM 11:59:21,532 INFO org.apache.flink.streaming.connectors.kinesis.FlinkKinesisConsumer [] - Subtask 3 will be seeded with initial shard StreamShardHandle{streamName='hp-16', shard='{ShardId: shardId-000000000015,HashKeyRange: {StartingHashKey: 319014718988379809496913694467282698240,EndingHashKey: 340282366920938463463374607431768211455},SequenceNumberRange: {StartingSequenceNumber: 49627894338681557796096402897798370703746460841949528306,}}'}, starting state set as sequence number EARLIEST_SEQUENCE_NUM 11:59:21,532 INFO org.apache.flink.streaming.connectors.kinesis.FlinkKinesisConsumer [] - Subtask 2 will be seeded with initial shard StreamShardHandle{streamName='hp-16', shard='{ShardId: shardId-000000000014,HashKeyRange: {StartingHashKey: 297747071055821155530452781502797185024,EndingHashKey: 319014718988379809496913694467282698239},SequenceNumberRange: {StartingSequenceNumber: 49627894338659257050897872274656834985473812480443547874,}}'}, starting state set as sequence number EARLIEST_SEQUENCE_NUM 11:59:21,532 INFO org.apache.flink.streaming.connectors.kinesis.internals.KinesisDataFetcher [] - Subtask 3 will start consuming seeded shard StreamShardHandle{streamName='hp-16', shard='{ShardId: shardId-000000000004,HashKeyRange: {StartingHashKey: 85070591730234615865843651857942052864,EndingHashKey: 106338239662793269832304564822427566079},SequenceNumberRange: {StartingSequenceNumber: 49627894338436249598912566043241477802747328865383743554,}}'} from sequence number EARLIEST_SEQUENCE_NUM with ShardConsumer 0 11:59:21,532 INFO org.apache.flink.streaming.connectors.kinesis.internals.KinesisDataFetcher [] - Subtask 2 will start consuming seeded shard StreamShardHandle{streamName='hp-16', shard='{ShardId: shardId-000000000003,HashKeyRange: {StartingHashKey: 63802943797675961899382738893456539648,EndingHashKey: 85070591730234615865843651857942052863},SequenceNumberRange: {StartingSequenceNumber: 49627894338413948853714035420099942084474680503877763122,}}'} from sequence number EARLIEST_SEQUENCE_NUM with ShardConsumer 0 11:59:21,532 INFO org.apache.flink.streaming.connectors.kinesis.FlinkKinesisConsumer [] - Subtask 0 will be seeded with initial shard StreamShardHandle{streamName='hp-16', shard='{ShardId: shardId-000000000001,HashKeyRange: {StartingHashKey: 21267647932558653966460912964485513216,EndingHashKey: 42535295865117307932921825928971026431},SequenceNumberRange: {StartingSequenceNumber: 49627894338369347363316974173816870647929383780865802258,}}'}, starting state set as sequence number EARLIEST_SEQUENCE_NUM 11:59:21,532 INFO org.apache.flink.streaming.connectors.kinesis.FlinkKinesisConsumer [] - Subtask 0 will be seeded with initial shard StreamShardHandle{streamName='hp-16', shard='{ShardId: shardId-000000000009,HashKeyRange: {StartingHashKey: 191408831393027885698148216680369618944,EndingHashKey: 212676479325586539664609129644855132159},SequenceNumberRange: {StartingSequenceNumber: 49627894338547753324905219158949156394110570672913645714,}}'}, starting state set as sequence number EARLIEST_SEQUENCE_NUM 11:59:21,532 INFO org.apache.flink.streaming.connectors.kinesis.FlinkKinesisConsumer [] - Subtask 7 will be seeded with initial shard StreamShardHandle{streamName='hp-16', shard='{ShardId: shardId-000000000000,HashKeyRange: {StartingHashKey: 0,EndingHashKey: 21267647932558653966460912964485513215},SequenceNumberRange: {StartingSequenceNumber: 49627894338347046618118443550675334929656735419359821826,}}'}, starting state set as sequence number EARLIEST_SEQUENCE_NUM 11:59:21,533 INFO org.apache.flink.streaming.connectors.kinesis.FlinkKinesisConsumer [] - Subtask 0 will be seeded with initial shard StreamShardHandle{streamName='hp-16', shard='{ShardId: shardId-000000000012,HashKeyRange: {StartingHashKey: 255211775190703847597530955573826158592,EndingHashKey: 276479423123262501563991868538311671807},SequenceNumberRange: {StartingSequenceNumber: 49627894338614655560500811028373763548928515757431587010,}}'}, starting state set as sequence number EARLIEST_SEQUENCE_NUM 11:59:21,533 INFO org.apache.flink.streaming.connectors.kinesis.FlinkKinesisConsumer [] - Subtask 7 will be seeded with initial shard StreamShardHandle{streamName='hp-16', shard='{ShardId: shardId-000000000008,HashKeyRange: {StartingHashKey: 170141183460469231731687303715884105728,EndingHashKey: 191408831393027885698148216680369618943},SequenceNumberRange: {StartingSequenceNumber: 49627894338525452579706688535807620675837922311407665282,}}'}, starting state set as sequence number EARLIEST_SEQUENCE_NUM 11:59:21,533 INFO org.apache.flink.streaming.connectors.kinesis.internals.KinesisDataFetcher [] - Subtask 0 will start consuming seeded shard StreamShardHandle{streamName='hp-16', shard='{ShardId: shardId-000000000001,HashKeyRange: {StartingHashKey: 21267647932558653966460912964485513216,EndingHashKey: 42535295865117307932921825928971026431},SequenceNumberRange: {StartingSequenceNumber: 49627894338369347363316974173816870647929383780865802258,}}'} from sequence number EARLIEST_SEQUENCE_NUM with ShardConsumer 0 11:59:21,533 INFO org.apache.flink.streaming.connectors.kinesis.FlinkKinesisConsumer [] - Subtask 7 will be seeded with initial shard StreamShardHandle{streamName='hp-16', shard='{ShardId: shardId-000000000011,HashKeyRange: {StartingHashKey: 233944127258145193631070042609340645376,EndingHashKey: 255211775190703847597530955573826158591},SequenceNumberRange: {StartingSequenceNumber: 49627894338592354815302280405232227830655867395925606578,}}'}, starting state set as sequence number EARLIEST_SEQUENCE_NUM 11:59:21,533 INFO org.apache.flink.streaming.connectors.kinesis.internals.KinesisDataFetcher [] - Subtask 7 will start consuming seeded shard StreamShardHandle{streamName='hp-16', shard='{ShardId: shardId-000000000000,HashKeyRange: {StartingHashKey: 0,EndingHashKey: 21267647932558653966460912964485513215},SequenceNumberRange: {StartingSequenceNumber: 49627894338347046618118443550675334929656735419359821826,}}'} from sequence number EARLIEST_SEQUENCE_NUM with ShardConsumer 0 11:59:21,568 INFO org.apache.flink.streaming.connectors.kinesis.FlinkKinesisConsumer [] - Subtask 1 will be seeded with initial shard StreamShardHandle{streamName='hp-16', shard='{ShardId: shardId-000000000002,HashKeyRange: {StartingHashKey: 42535295865117307932921825928971026432,EndingHashKey: 63802943797675961899382738893456539647},SequenceNumberRange: {StartingSequenceNumber: 49627894338391648108515504796958406366202032142371782690,}}'}, starting state set as sequence number EARLIEST_SEQUENCE_NUM 11:59:21,568 INFO org.apache.flink.streaming.connectors.kinesis.FlinkKinesisConsumer [] - Subtask 1 will be seeded with initial shard StreamShardHandle{streamName='hp-16', shard='{ShardId: shardId-000000000013,HashKeyRange: {StartingHashKey: 276479423123262501563991868538311671808,EndingHashKey: 297747071055821155530452781502797185023},SequenceNumberRange: {StartingSequenceNumber: 49627894338636956305699341651515299267201164118937567442,}}'}, starting state set as sequence number EARLIEST_SEQUENCE_NUM 11:59:21,568 INFO org.apache.flink.streaming.connectors.kinesis.internals.KinesisDataFetcher [] - Subtask 1 will start consuming seeded shard StreamShardHandle{streamName='hp-16', shard='{ShardId: shardId-000000000002,HashKeyRange: {StartingHashKey: 42535295865117307932921825928971026432,EndingHashKey: 63802943797675961899382738893456539647},SequenceNumberRange: {StartingSequenceNumber: 49627894338391648108515504796958406366202032142371782690,}}'} from sequence number EARLIEST_SEQUENCE_NUM with ShardConsumer 0 11:59:23,209 INFO org.apache.flink.streaming.connectors.kinesis.internals.KinesisDataFetcher [] - Subtask 0 will start consuming seeded shard StreamShardHandle{streamName='hp-16', shard='{ShardId: shardId-000000000009,HashKeyRange: {StartingHashKey: 191408831393027885698148216680369618944,EndingHashKey: 212676479325586539664609129644855132159},SequenceNumberRange: {StartingSequenceNumber: 49627894338547753324905219158949156394110570672913645714,}}'} from sequence number EARLIEST_SEQUENCE_NUM with ShardConsumer 1 11:59:23,244 INFO org.apache.flink.streaming.connectors.kinesis.internals.KinesisDataFetcher [] - Subtask 6 will start consuming seeded shard StreamShardHandle{streamName='hp-16', shard='{ShardId: shardId-000000000010,HashKeyRange: {StartingHashKey: 212676479325586539664609129644855132160,EndingHashKey: 233944127258145193631070042609340645375},SequenceNumberRange: {StartingSequenceNumber: 49627894338570054070103749782090692112383219034419626146,}}'} from sequence number EARLIEST_SEQUENCE_NUM with ShardConsumer 1 event: 6; timestamp: 1648030822428, 2022-03-23T10:20:22.428Z 11:59:23,377 INFO org.apache.flink.streaming.connectors.kinesis.internals.KinesisDataFetcher [] - Subtask 3 will start consuming seeded shard StreamShardHandle{streamName='hp-16', shard='{ShardId: shardId-000000000015,HashKeyRange: {StartingHashKey: 319014718988379809496913694467282698240,EndingHashKey: 340282366920938463463374607431768211455},SequenceNumberRange: {StartingSequenceNumber: 49627894338681557796096402897798370703746460841949528306,}}'} from sequence number EARLIEST_SEQUENCE_NUM with ShardConsumer 1 11:59:23,405 INFO org.apache.flink.streaming.connectors.kinesis.internals.KinesisDataFetcher [] - Subtask 2 will start consuming seeded shard StreamShardHandle{streamName='hp-16', shard='{ShardId: shardId-000000000014,HashKeyRange: {StartingHashKey: 297747071055821155530452781502797185024,EndingHashKey: 319014718988379809496913694467282698239},SequenceNumberRange: {StartingSequenceNumber: 49627894338659257050897872274656834985473812480443547874,}}'} from sequence number EARLIEST_SEQUENCE_NUM with ShardConsumer 1 11:59:23,581 INFO org.apache.flink.streaming.connectors.kinesis.internals.KinesisDataFetcher [] - Subtask 7 will start consuming seeded shard StreamShardHandle{streamName='hp-16', shard='{ShardId: shardId-000000000008,HashKeyRange: {StartingHashKey: 170141183460469231731687303715884105728,EndingHashKey: 191408831393027885698148216680369618943},SequenceNumberRange: {StartingSequenceNumber: 49627894338525452579706688535807620675837922311407665282,}}'} from sequence number EARLIEST_SEQUENCE_NUM with ShardConsumer 1 11:59:23,586 INFO org.apache.flink.streaming.connectors.kinesis.internals.KinesisDataFetcher [] - Subtask 1 will start consuming seeded shard StreamShardHandle{streamName='hp-16', shard='{ShardId: shardId-000000000013,HashKeyRange: {StartingHashKey: 276479423123262501563991868538311671808,EndingHashKey: 297747071055821155530452781502797185023},SequenceNumberRange: {StartingSequenceNumber: 49627894338636956305699341651515299267201164118937567442,}}'} from sequence number EARLIEST_SEQUENCE_NUM with ShardConsumer 1 11:59:24,790 INFO org.apache.flink.streaming.connectors.kinesis.internals.KinesisDataFetcher [] - Subtask 0 will start consuming seeded shard StreamShardHandle{streamName='hp-16', shard='{ShardId: shardId-000000000012,HashKeyRange: {StartingHashKey: 255211775190703847597530955573826158592,EndingHashKey: 276479423123262501563991868538311671807},SequenceNumberRange: {StartingSequenceNumber: 49627894338614655560500811028373763548928515757431587010,}}'} from sequence number EARLIEST_SEQUENCE_NUM with ShardConsumer 2 event: 4; timestamp: 1648030809282, 2022-03-23T10:20:09.282Z event: 3; timestamp: 1648030797697, 2022-03-23T10:19:57.697Z event: 5; timestamp: 1648030810871, 2022-03-23T10:20:10.871Z 11:59:24,907 INFO org.apache.flink.streaming.connectors.kinesis.internals.KinesisDataFetcher [] - Subtask 7 will start consuming seeded shard StreamShardHandle{streamName='hp-16', shard='{ShardId: shardId-000000000011,HashKeyRange: {StartingHashKey: 233944127258145193631070042609340645376,EndingHashKey: 255211775190703847597530955573826158591},SequenceNumberRange: {StartingSequenceNumber: 49627894338592354815302280405232227830655867395925606578,}}'} from sequence number EARLIEST_SEQUENCE_NUM with ShardConsumer 2 event: 7; timestamp: 1648030834105, 2022-03-23T10:20:34.105Z event: 1; timestamp: 1648030794441, 2022-03-23T10:19:54.441Z event: 2; timestamp: 1648030796122, 2022-03-23T10:19:56.122Z event: 8; timestamp: 1648030887171, 2022-03-23T10:21:27.171Z XXXXXXXXXXXXXX Window with 3 events; Watermark: 1648030809281, 2022-03-23T10:20:09.281Z 3 1 2 XXXXXXXXXXXXXX Window with 2 events; Watermark: 1648030834104, 2022-03-23T10:20:34.104Z 4 5 XXXXXXXXXXXXXX Window with 1 events; Watermark: 1648030834104, 2022-03-23T10:20:34.104Z 6 XXXXXXXXXXXXXX Window with 1 events; Watermark: 1648030887170, 2022-03-23T10:21:27.170Z 7

The output is only showing 4 windows (missing the last window containing event 8). This is due to event time and the watermark strategy. The last window cannot close because the with the per-built watermark strategies the time never advances beyond the time of the last event that has been read from the stream. But for the window to close, time needs to advance more than 10 seconds after the last event. In this case the last watermark is 2022-03-23T10:21:27.170Z but in order for the session window to close, a watermark 10s and 1ms later is required.

If the withIdleness option is removed from the watermark strategy, no session window will ever close, because the the “global watermark” of the window operator cannot advance.

Note that when the Flink application starts (or if there is data skew), some shards may be consumed faster than others. This can cause some watermarks to be emitted too early from a subtask (the subtask may emit the watermark based on the content of one shard without having consumed from the other shards it’s subscribed to). Ways to mitigate are a different watermarking strategies that add a safety buffer (forBoundedOutOfOrderness(Duration.ofSeconds(30)) or explicitly allowing late arriving events (allowedLateness(Time.minutes(5)).