Amazon DynamoDB - All You Need To Know About This AWS Service

Minh-Phuc Tran on Nov 30, 2020





DynamoDB is a NoSQL database that delivers single-digit millisecond performance at scale.

"DynamoDB can manage upward of 10 trillion requests daily and can support thresholds of more than 20 million requests per second" - AWS

Having been using it for while, in this article, I'm going to summarize essential knowledge about Amazon DynamoDB that helps you: Quickly understand and know its use cases -> Get started using it -> Know what to consider for best utilization.

It's a 10-minute read, but after reading it, you'll know how DynamoDB works, when you should use it, and how to best utilize it when it's necessary.

Quickly Understand

First of all, it's Amazon DynamoDB, not AWS DynamoDB. (People understand it either way, it's respectful to call a thing correctly though)

What is DynamoDB?

Basically, DynamoDB is a database where you write and read your application data. Application data is the core information of your application like user profiles, user activities, transactions, etc. These data change very frequently and grow enormously large as your application grows. Using a database like DynamoDB, you don't need to work with physical disks, you simply ask the database to write and read data for you, it will make sure things work correctly under the hood.

Technically, DynamoDB is a fully-managed, NoSQL, key-value and document-based database.

  • Fully-managed: AWS fully manages it for you. DynamoDB scales seamlessly, you don't maintain any server. AWS takes full responsibility for running, managing servers, software updates, backing up, cluster scaling, and all other ops. You use it by calling its APIs, just like how you'd use Stripe.

  • NoSQL, key-value: DynamoDB has neither relationship nor schema, which means that you don't have to define what columns or fields an entry of data has. DynamoDB consists of tables. Each table consists of items. Each item has a uniquely identifiable key and other non-constraint attributes. Each attribute is a pair of key and value. A key is a string. A value can be null, string, number, boolean, binary, list, or nested object. Except for the unique key, items can have different attributes from each other.

    // How data in DynamoDB is conceptually structured 
    "Table X": [
         "key": "unique-key-1",
         "a": "some-value",
         "b": 1
         "key": "unique-key-2",
         "c": false,
         "d": {
            "it": "can have nested JSON",
            "too": null
      // other entries
    "Table Y": [
      // ...
    • Document-based: Each entry of data in DynamoDB is a JSON document, a.k.a. JSON object. Every operation happens at the document level, which means that an operation is either a document is retrieved, or a document is created, or a document is replaced, or a document is deleted. There's no operation at attribute level.

Pricing model

DynamoDB offers 2 pricing models:

  • Pay per request: you pay for the number of writes and reads you requested. DynamoDB automatically scales to accommodate traffic from your application. It costs $1.25 per million write request units and $0.25 per million read request units (in Ohio region). This model is a good choice if you can't predict how much traffic your application will receive and don't prioritize cost optimization yet.

  • Provisioned: you specify and pay for a specified number of read units and write units per hour. DynamoDB will be always ready to handle traffic at the provisioned capacity, regardless of how much traffic your application is actually receiving. This helps you set a predictable cost and performance for your application. It's useful when you've already known the pattern. You can change provisioned units every 24 hours.

Additionally, when your application is growing steadily, you can buy reserved units, which can save over half the price if you pay upfront for a year.

Use cases

DynamoDB works best when you're searching for simplicity and scalability.

Because of its serverless trait, DynamoDB helps you have (theoretically) infinitely scalable database infrastructure from day one with zero ops overhead in the future. When your application data grow to terabytes or petabytes, your database infrastructure stays the same, you simply provision and buy more capacity. This can be an advantage to keep your technical team lean.

However, DynamoDB query ability is quite limited compared to most other database services, which makes it difficult and inefficient to query complex models and relationships.

In my experience, DynamoDB is a great choice for microservices, where each service has 2-4 models, and their relationships can be detached and implemented in a NoSQL design.

Netflix uses DynamoDB to build its A/B testing service that serves 125+ million users, while FanFight reported saving 50% of cost and achieving 4x revenue by migrating to DynamoDB.

Get Started Using It

Now, let's explore some essential concepts to get started using it.


DynamoDB slices your table up into smaller chunks of data called partitions, in order to speed up reads.

Each table must have a partition key, which DynamoDB uses to decide which partition to put an item into.

A partition is also logical isolation of a DynamoDB query. A DynamoDB query can only query data in a partition. If you need to query data in 2 different partitions, you'll need 2 queries or a full table scan.

Primary key

A primary key is a way to uniquely identify an item in a table, it is the only attribute that you'll need to define before creating a DynamoDB table and starting to use it.

There're 2 types of a primary key in DynamoDB:

  • Single primary key: the partition key - a single and always-unique attribute, which is usually an ID, e.g. user ID, post ID, etc.

  • Composite primary key: the partition key and a sort key. The first key is used for partitioning, the second key is used to sort items in the same partition. The combined value of these two keys has to be unique across a table.

5 types of read

In DynamoDB, there are 5 types of a read operation:

  • GetItem: read a single item by specifying its primary key.

  • BatchGetItems: send a request that groups up to 100 GetItem requests together. Each request is executed separately and in parallel. It is possible that some reads succeed whereas others fail.

  • TransactGetItems: all-or-nothing BatchGetItems. All operations must either succeed or fail together. This costs twice as much as BatchGetItems.

  • Query: read multiple items in the same partition. You must specify the partition key, and optionally a filter condition on the sort key.

  • Scan: read all items in a table.

Read consistency

DynamoDB has 2 types of read consistency, which can be useful for different applications:

  • Eventually read consistency: you get your data the fastest, but not always the latest. You'll usually receive the data 1 second ago.

  • Strongly read consistency: you always get the latest data, but the latency may be higher and the reading cost is double.

Secondary indexes

Every read operation except for a Scan must start with the partition key, to query on other attributes without scanning through the whole table, you'll need another tool - a secondary index.

A secondary index is essentially an (automatic) copy of a table but has a different partition key and sort key so that you can query on these different keys.

There're two types of a secondary index in DynamoDB:

  • Global secondary index: is a complete copy of a table, has its own partition key and (optionally) a sort key. A partition key in a global secondary index doesn't have to be unique. Global secondary index only supports eventually consistent read.

  • Local secondary index: is a local copy of a partition sorted by another key. A local secondary index shares the same partition key as the base table but has a different sort key. Local secondary index supports both strongly and eventually consistent read.

After creating an index, you do all read operations on it as if you do on the base table.

Creating a secondary index will cost you more for data storage and write operations. DynamoDB automatically updates the indexes when you update the base table and it will charge for all these updates as typical write operations.

3 types of write

In DynamoDB, there are 3 types of a write operation:

  • PutItem, UpdateItem, and DeleteItem: create, update and delete a single item by specifying its primary key.

  • BatchWriteItems: send a request that groups up to 25 PutItem and DeleteItem requests together. Each request is executed separately and in parallel. It is possible that some writes succeed whereas others fail.

  • TransactWriteItems: all-or-nothing BatchWriteItems. All operations must either succeed or fail together. This costs twice as much as BatchWriteItems.

Advice for best utilization

There're at most 3 related models

Related models mean that these models have a direct relationship to each other. For example, a post belongs to a user, a post has multiple comments, a user can comment on multiple posts.

When your app has more than 3 related models, it's highly a chance that SQL is a better choice because you have out-of-the-box extremely complex query features in SQL. Modeling relationships in SQL is much simpler. On the contrary, trying to model relationships in NoSQL is like trying to workaround its disadvantages.

Can I break my app into smaller apps?

DynamoDB is best for microservices when your app is a combination of multiple smaller apps.

Let's take the posts-comments-users example, can you break it into smaller apps? The answer is yes, you can make it 3 smaller apps:

  • User directory app: stores user profiles, credentials, etc, and authenticates users.

  • Blogging app: stores and renders posts.

  • Commenting app: stores and shows comments.

DynamoDB is the perfect database for these smaller apps because they have very simple and discrete data.

Pro-tip #1: There's a high possibility that there're external services implemented for one of your smaller apps, e.g. Firebase for authentication, Disqus for commenting.

Pro-tip #2: Because of this, DynamoDB is a great choice if you want to build micro-SaaS, which is on-trend in the current market .

1 write = 40 reads

Use this to decide whether you should create an index, duplicate, or cache some values to optimize for reads.

When you create an index, duplicate, or cache a value, you need at least one more write which costs as much as 40 eventually consistent reads based on the DynamoDB pricing model.

So, if a write operation that can reduce more than 40 reads, it's worthing implementing, otherwise, it isn't.

Avoid Scan and FilterExpression

Scan reads all items every time you call it, which is both costly and time-consuming. You are usually able to create a secondary index to avoid scanning. If you can't avoid a scan, you may think about whether you should use DynamoDB.

DynamoDB has a FilterExpression in Query and Scan for you to filter out results before returning. However, it isn't a filter condition on query operations. The filter is applied only after the query is executed, which only has value in saving network bandwidth.

Keep less than 5 secondary indexes

If you create more than 5 secondary indexes, it means your application models and queries are becoming more complex than what DynamoDB may be best for. In this case, consider migrating to SQL or breaking your apps into smaller ones.

Key takeaways

Before summarizing, hey ! Thank you for reading this far and I hope you learn how DynamoDB should work for you in the future 😉.

  • DynamoDB is a NoSQL document database, there's no relationship and schema, all operations happen at the document level.

  • DynamoDB is fully-managed by AWS. You have a scalable database infrastructure from day one with zero ops overhead.

  • You can pay per request at the early stage. Later, you can buy provisioned and reserved units to save and set predictable cost + performance for your app.

  • A primary key in DynamoDB is a partition key or a partition key and a sort key. DynamoDB uses these keys to logically identify items and physically isolate them for better read performance at scale.

  • All read operations except for Scan operates on a single partition. To query on a different attribute, you'll need a secondary index.

  • DynamoDB works best when you're searching for simplicity and scalability. If your application has more than 3 related models or 5 secondary indexes, consider breaking it into smaller apps or migrating to a SQL database.

Subscribe to the newsletter

Get emails from me about software development, SaaS, and early access to new articles.