I don't understand why I save it on a streaming server? such as Amazon Kinesis or Apache Stream before saving it to DB.

If a large number of data sources are accessing the datastore directly,

• Very difficult to make changes to the datastore
• Performance depends on the data source and cannot be managed
• Processing a small amount of data many times is too much of an overhead
• Data source side processing is bloated
• Data store side also takes action when data from multiple sources must be collected

There are many problems such as and so on.If there is something between the data source and the datastore that buffers, these problems can be easily resolved.

• The buffer side knows where the datastore is, so you just have to change it
• Output from buffers allows you to manage performance
• Each datastore has a low overhead data entry method
• Data source side can only specialize in sending data to buffer
• Intermediate processing such as data deformation and aggregation

Performance is key, but not all.

Learn what Amazon Kinesis is from examples

The article in the link above summarizes the advantages of Amazon Kinesis in a compact way, so it would be good if you could refer to it.
This article is a little over a year ago, so the information may be a little out of date.

What comes to mind about the reason why I'm

If asynchronous processing does not equal the throughput of input data and the processing power of the output destination, I think it is common to temporarily save it somewhere (queueing, caching, etc.)

To ensure data order and allow batch processing of large amounts of input data in parallel, I think it is temporarily stored.

The ability to retrieve the same data over and over again may also be difficult with a proxy-like data transfer mechanism.

Considering that the resources are finite, I think that sometimes output processing cannot keep up.There is also a limitation to keeping data put to Kinesis up to 24 hours, so if you have never taken it out before and the output side has not processed it, you will miss it.

Also, there is no limit to Kinesis throughput, but it scales only to 10 shards by default.If the throughput is insufficient, the application should mitigate it.

Please refer to FAQ
https://aws.amazon.com/jp/kinesis/streams/faqs/

If it is the role of streaming processing, I think it is to process a large amount of input data in real time and output it to a data warehouse.The Apache Spark Stream site's materials were easy to understand as an overview.

http://spark.apache.org/talks/
http://spark.apache.org/talks/strata_spark_streaming.pdf

According to this document, there are five requirements for the Streaming Framework:
The role of streaming processing is to do this.

• Scalable to large clusters
• Achieves second-scale latencies
• Has simple programming model
• Integrates with batch&interactive workloads
• Ensures efficient fault-tolerance in stateful computations