Terraform: provider/aws aws_kinesis_stream recreates stream when shard_count changes

[chaliy]

Terraform Version

Terraform v0.8.6

Affected Resource(s)

aws_kinesis_stream

Debug Output

https://gist.github.com/chaliy/683eb59c1bfc415d8b890aabedb92963

Expected Behavior

Shard count changed on AWS (shards spitted or merged). Items under retention period are safe.

Actual Behavior

Kinesis stream recreated. Effectively all items in stream is wiped out.

Steps to Reproduce

Any change to shard_count.

@stack72

[apparentlymart]

Thanks for this request, @chaliy.

It looks like there's a reasonable "default way" to do this via the UpdateShardCount API function, avoiding the complexities of manual resharding. But it has some constraints and best-practices that I guess we'd need to document to help users avoid apply-time errors or unexpected costs.

[DavidAntaramian]

Also, regarding this issue, by requiring shard_count, the shard_count in the Terraform file has to be updated to reflect changes made by external systems. Because Kinesis is used for real-time data, sharding can occur using automated services such as the Amazon Kinesis Scaling Utils rather than waiting for manual intervention by Ops teams.

If such an external system begins splitting and merging shards, the relevant Terraform file has to be updated to reflect the change. Otherwise, as described above, Terraform will delete the stream and recreate it, causing data loss. Since the shard_count parameter is required, there is no way to avoid this when managing a Kinesis stream in Terraform.

I think the shard_count parameter as it currently exists, should actually be an initial_shard_count. The initial_shard_count parameter would be required, but it would only be used when initially creating the stream. Subsequent management of the stream by automated services would then not cause the service state and the Terraform state to need reconciling.

To allow for explicit shard control through Terraform, the shard_count parameter could then be made optional and be used to trigger the referenced UpdateShardCount API function above when the service state and the Terraform state do not match.

[jessecollier]

This is blocking me as well. Shard_count + manual scaling of shards are not happy when re-applying terraform

[cbroglie]

Note that you can already make shard_count act like the initial_shard_count idea from above by telling Terraform to ignore changes to shard_count:

resource "aws_kinesis_stream" "stream" {
  name = "my-stream"
  shard_count = 4
  lifecycle {
    ignore_changes = ["shard_count"]
  }
}

And when https://github.com/hashicorp/terraform/pull/13562 is available, you can manually resize the stream with the UpdateShardCount API and still access the current number of shards in other parts of the Terraform config.

As @apparentlymart noted, UpdateShardCount has limitations (like only being able to be used twice per 24 hours per stream) which may not be suitable for the default behavior if shard_count changes.

[hashibot[bot]]

I'm going to lock this issue because it has been closed for _30 days_ ⏳. This helps our maintainers find and focus on the active issues.

If you have found a problem that seems similar to this, please open a new issue and complete the issue template so we can capture all the details necessary to investigate further.