タダです.

re:Invent 中に S3 Vectors が GA されました.東京リージョンにもきており,Terraform aws provider v6.24.0 で S3 Vectors のサポートが入っているため試してみます.

リリース概要

v6.24.0 では次の S3 Vectors のリソースがサポートされました.

aws_s3vectors_vector_bucket: ベクトルデータを格納するバケット
aws_s3vectors_vector_bucket_policy: 上記ベクトルバケットに対するアクセスポリシー管理
aws_s3vectors_index: ベクトル検索用のインデックスリソース

これらのリソースを東京リージョンで定義して使ってみましょう.

Terraform で S3 Vectors を定義する

以前書いた記事をベースにリソースを作ります.

resource "aws_s3vectors_vector_bucket" "blog" {
  vector_bucket_name = "blog-s3-vectors"
}

resource "aws_s3vectors_index" "blog" {
  index_name         = "blog-s3-vectors-index"
  vector_bucket_name = aws_s3vectors_vector_bucket.blog.vector_bucket_name
  data_type          = "float32"
  dimension          = 1024
  distance_metric    = "cosine"
}

この定義ですと次のような実行計画ができるので作成します.

Terraform will perform the following actions:

  # aws_s3vectors_index.blog will be created
  + resource "aws_s3vectors_index" "blog" {
      + creation_time            = (known after apply)
      + data_type                = "float32"
      + dimension                = 1024
      + distance_metric          = "cosine"
      + encryption_configuration = (known after apply)
      + index_arn                = (known after apply)
      + index_name               = "blog-s3-vectors-index"
      + region                   = "ap-northeast-1"
      + tags_all                 = {}
      + vector_bucket_name       = "blog-s3-vectors"
    }

  # aws_s3vectors_vector_bucket.blog will be created
  + resource "aws_s3vectors_vector_bucket" "blog" {
      + creation_time            = (known after apply)
      + encryption_configuration = (known after apply)
      + force_destroy            = false
      + region                   = "ap-northeast-1"
      + tags_all                 = {}
      + vector_bucket_arn        = (known after apply)
      + vector_bucket_name       = "blog-s3-vectors"
    }

Plan: 2 to add, 0 to change, 0 to destroy.

Do you want to perform these actions?
  Terraform will perform the actions described above.
  Only 'yes' will be accepted to approve.

  Enter a value: yes

aws_s3vectors_vector_bucket.blog: Creating...
aws_s3vectors_vector_bucket.blog: Creation complete after 0s
aws_s3vectors_index.blog: Creating...
aws_s3vectors_index.blog: Creation complete after 0s

Apply complete! Resources: 2 added, 0 changed, 0 destroyed.

作成されたリソースが作成できたかを AWS CLI を使って確認したところ意図通り東京リージョンにリソースが存在してました.

# バケットの確認
❯ aws s3vectors list-vector-buckets --region ap-northeast-1
{
    "vectorBuckets": [
        {
            "vectorBucketName": "blog-s3-vectors",
            "vectorBucketArn": "arn:aws:s3vectors:ap-northeast-1:1234567891011:bucket/blog-s3-vectors",
            "creationTime": "2025-12-14T12:34:42+09:00"
        }
    ]
}
# インデックスの確認
❯ aws s3vectors get-index --vector-bucket-name blog-s3-vectors --index-name blog-s3-vectors-index --region ap-northeast-1
{
    "index": {
        "vectorBucketName": "blog-s3-vectors",
        "indexName": "blog-s3-vectors-index",
        "indexArn": "arn:aws:s3vectors:ap-northeast-1:1234567891011:bucket/blog-s3-vectors/index/blog-s3-vectors-index",
        "creationTime": "2025-12-14T12:34:43+09:00",
        "dataType": "float32",
        "dimension": 1024,
        "distanceMetric": "cosine"
    }
}

S3 Vectors にテストデータを投入する

テストデータを投入し,検索してみます.テストデータは以下のように生成して S3 Vectors にアップロードします.

❯ cat input.json
{
 "inputText": "This is test vector data."
}
❯ aws bedrock-runtime invoke-model --model-id "amazon.titan-embed-text-v2:0" --body fileb://input.json --content-type application/json --accept application/json --region ap-northeast-1 output.json
{
    "contentType": "application/json"
}
❯ jq -c '.embedding' output.json > vector_data.json
❯ cat <<EOF > vector_upload.json
∙ [
  {
    "key": "doc1",
    "data": {
      "float32": $(cat vector_data.json)
    },
    "metadata": {
      "source_text": "This is test vector data.",
      "category": "test"
    }
  }
]
∙ EOF
❯ aws s3vectors put-vectors --vector-bucket-name blog-s3-vectors --index-name blog-s3-vectors-index --vectors file://vector_upload.json --region ap-northeast-1

投入したデータを検索します.意図通りに検索できました.

❯ jq '{float32: .embedding}' output.json > embedding.json
❯ aws s3vectors query-vectors --vector-bucket-name blog-s3-vectors --index-name blog-s3-vectors-index --top-k 3 --query-vector file://embedding.json --return-metadata --region ap-northeast-1
{
    "vectors": [
        {
            "key": "doc1",
            "metadata": {
                "category": "test",
                "source_text": "This is test vector data."
            }
        }
    ],
    "distanceMetric": "cosine"
}

まとめ

GA された S3 Vectors を Terraform でリソース管理を試してメモをまとめました.