Why so many shardIds when I've only configured 3 in my Kinesis Stream?

Question

I have Kinesis consumer code that does a DescribeStream and then spins up a new Java thread per shardId to consume off each shard.

I get 8 shardIds when I've only configured 3 in my Stream. Why is that?  I don't want to have 5 extra threads consuming constantly and getting zero records. Below, you can see I'm logging the total # of records processed on each shard.

2020-11-19 08:59:49 INFO  GetRecords:109 - # Kinesis consumers: 8
    2020-11-19 08:59:49 INFO  GetRecords:112 - Kinesis - ShardId: 'shardId-000000000000', Total Records: 0
    2020-11-19 08:59:49 INFO  GetRecords:112 - Kinesis - ShardId: 'shardId-000000000001', Total Records: 0
    2020-11-19 08:59:49 INFO  GetRecords:112 - Kinesis - ShardId: 'shardId-000000000002', Total Records: 0
    2020-11-19 08:59:49 INFO  GetRecords:112 - Kinesis - ShardId: 'shardId-000000000003', Total Records: 19110
    2020-11-19 08:59:49 INFO  GetRecords:112 - Kinesis - ShardId: 'shardId-000000000004', Total Records: 0
    2020-11-19 08:59:49 INFO  GetRecords:112 - Kinesis - ShardId: 'shardId-000000000005', Total Records: 0
    2020-11-19 08:59:49 INFO  GetRecords:112 - Kinesis - ShardId: 'shardId-000000000006', Total Records: 18981
    2020-11-19 08:59:49 INFO  GetRecords:112 - Kinesis - ShardId: 'shardId-000000000007', Total Records: 16195

**Background:** I started with 1, then configured 2, then, 3. Does this have something to do with the other shardIds that have 0 records? If so, what is the recommended code/practice to ignore a certain type of shard?

Accepted Answer

When you change the number of shards, presumably with the update-shard-count api, Kinesis is taking care of merging and splitting individual shards to get to your desired number of shards. Records in kinesis are immutable, outside of aging off the stream, and need to be able to be read in order. So you end up with a lineage of shards. A parent shard that contained all records within a given hash key range, and then two child shards that split that range and take new records going forward. When those child shards are created the parent shard doesn't go away, the records stay in place in that parent shard and the child shards start taking new records. This also works in reverse when you merge two shards.

When you do the describeStream call each shard in the shards list will list a ParentShardId if it has one. This allows you to contruct a map of the lineage to start reading in the correct place. This is also some of the complexity that is handled for you if you're using the KCL or a Lambda consumer

When you do the describeStream call each shard in the shards list will list a ParentShardId if it has one. This allows you to contruct a map of the lineage to start reading in the correct place. This is also some of the complexity that is handled for you if you're using the [KCL][1] or a [Lambda consumer][2]

[1]: https://docs.aws.amazon.com/streams/latest/dev/shared-throughput-kcl-consumers.html
  [2]: https://docs.aws.amazon.com/streams/latest/dev/lambda-consumer.html

Why so many shardIds when I've only configured 3 in my Kinesis Stream?

Relevant content