DynamoDB Deep Dive Advence Design
from 2004 amazon.com outage
實際使用案例
- amazon.com
- disney+
- netflix
- dropbox
- pokemon company
Under the hood
- partition 10GB做一次切割
data model
- different concept from Relational DB
- NoSQL Workbench
features
- Dynamodb stream(item-level change log) for events
- TTL
- transcation
- Point-in-time recovery (PITR)
Detail Description
- partion key best practice
- Secondary Index
- 事前就想到 => LSI
- 事後才想加Index => GSI
- Fine-Grain Access Control
- IAM
- horizontal
- vertical
- Memory Cache
- DAX
Scenario
- building and maintaing an S3 metadata
- users buy products in online retail shops
- aggration for raw records
Amazon Dynamo DB NoSQL Designs and Best Praction
- 針對Analytics(OLAP)場景較弱
- 客戶體驗(OLTP) vs 後端統計(OLAP)(ex. 過去一年/一季/一個月的訂單量)
- SQL is good for OLAP; NoSQL for OLTP for scale
- 可以用DynomaDB Stream強化OLAP應用
- Archive
- TTL + DynomaDB Stream
SQL vs NoSQL
- Data Modeling
- Data Access Pattern
- Data Query
partion key
- AWS Blog: Choosing the right dynamodb partition key 高鑑別度的資料，會影響實體partion的存放
operation
建議Batch Write Operation for cost saving
QA
- prefix use #? => 不成文規定
-
PROFILE#alexdebrie vs ORDER#abcsdf => #PROFILE ORDER 沒有特別意思
- update ok status 代表 已經同步至3個AZ完成，包含GSI?
- partition有可能只有一個item的情況嗎? => 有可能，看AWS後面hash function的設計
- partition key是具有唯一性的資料, ex. uuid, 代表一定會在不同partition嗎? => 不一定，看AWS後面hash function的設計或是該item是否是非常熱門
- primary key(partion key+sort key)一定具有唯一性?