Kinesis Streams、Firehose、リアルタイム分析

Kinesis ファミリーの概要

Kinesis Data Streams のアーキテクチャ

# Create a Kinesis Data Stream with 4 shards
aws kinesis create-stream \
  --stream-name clickstream \
  --shard-count 4

# Put a record into the stream
aws kinesis put-record \
  --stream-name clickstream \
  --partition-key 'user-123' \
  --data 'eyJldmVudCI6ICJjbGljayJ9'

シャード、スループット、スケーリング

# Split shard to increase throughput
aws kinesis split-shard \
  --stream-name clickstream \
  --shard-to-split shardId-000000000001 \
  --new-starting-hash-key 170141183460469231731687303715884105728

# Register an enhanced fan-out consumer
aws kinesis register-stream-consumer \
  --stream-arn arn:aws:kinesis:us-east-1:123456789012:stream/clickstream \
  --consumer-name analytics-app

Kinesis Data Firehose：マネージド配信

# Create a Firehose delivery stream to S3
aws firehose create-delivery-stream \
  --delivery-stream-name clickstream-to-s3 \
  --s3-destination-configuration '{
    "RoleARN": "arn:aws:iam::123456789012:role/FirehoseRole",
    "BucketARN": "arn:aws:s3:::my-data-lake-123",
    "Prefix": "landing/clickstream/year=!{timestamp:yyyy}/month=!{timestamp:MM}/",
    "BufferingHints": {"SizeInMBs": 64, "IntervalInSeconds": 300},
    "CompressionFormat": "GZIP"
  }'

Lambda による Firehose データ変換

# Lambda transform function signature for Firehose
def lambda_handler(event, context):
    output = []
    for record in event['records']:
        import base64, json
        payload = json.loads(base64.b64decode(record['data']))
        # Drop events with no user_id
        if not payload.get('user_id'):
            output.append({'recordId': record['recordId'], 'result': 'Dropped', 'data': record['data']})
        else:
            output.append({'recordId': record['recordId'], 'result': 'Ok', 'data': record['data']})
    return {'records': output}

Managed Service for Apache Flink

# Flink SQL-style tumbling window (conceptual)
# Count page views per URL every 5 minutes
CREATE TABLE clickstream (
  url STRING,
  event_time TIMESTAMP(3),
  WATERMARK FOR event_time AS event_time - INTERVAL '5' SECOND
) WITH ('connector' = 'kinesis', 'stream' = 'clickstream', ...);

SELECT
  url,
  COUNT(*) AS views,
  TUMBLE_START(event_time, INTERVAL '5' MINUTE) AS window_start
FROM clickstream
GROUP BY url, TUMBLE(event_time, INTERVAL '5' MINUTE);

Streams と Firehose の選択

Kinesis と SQS：試験で頻出する選択

Kinesis プロデューサー：SDK と KPL

# Basic KPL usage (Java pseudocode, no backticks)
KinesisProducer producer = new KinesisProducer();
byte[] data = 'hello world'.getBytes();
ByteBuffer buf = ByteBuffer.wrap(data);
// addUserRecord handles aggregation and retry internally
ListenableFuture future = producer.addUserRecord('clickstream', 'partitionKey', buf);
Futures.addCallback(future, new FutureCallback() { ... });

Firehose から Redshift へのパターン

# Firehose Redshift destination (CLI snippet)
--redshift-destination-configuration '{
  "RoleARN": "arn:aws:iam::123456789012:role/FirehoseRole",
  "ClusterJDBCURL": "jdbc:redshift://cluster.xyz.us-east-1.redshift.amazonaws.com:5439/sales",
  "CopyCommand": {
    "DataTableName": "clickevents",
    "CopyOptions": "JSON 'auto'"
  },
  "Username": "firehose_user",
  "Password": "{{resolve:secretsmanager:redshift-pw}}",
  "S3Configuration": {
    "RoleARN": "...",
    "BucketARN": "arn:aws:s3:::firehose-staging"
  }
}'

CloudWatch による Kinesis の監視

# CloudWatch alarm on high consumer lag
aws cloudwatch put-metric-alarm \
  --alarm-name kinesis-high-lag \
  --metric-name GetRecords.IteratorAgeMilliseconds \
  --namespace AWS/Kinesis \
  --dimensions Name=StreamName,Value=clickstream \
  --statistic Maximum \
  --period 60 \
  --threshold 60000 \
  --comparison-operator GreaterThanThreshold \
  --evaluation-periods 3 \
  --alarm-actions arn:aws:sns:us-east-1:123456789012:ops-alerts

クイックチェック

レッスンのまとめ