Awesome

Streamstone is a tiny embeddable library targeted at building scalable event-sourced applications on top of Azure Table Storage. It has simple, functional style API, heavily inspired by Greg Young's Event Store.

Features

Fully ACID compliant
Optimistic concurrency support
Duplicate event detection (based on identity)
Automatic continuation for both writes and reads (over WATS limits)
Custom stream and event properties you can query on
Synchronous (inline) projections and snapshots
Change tracking support for inline projections
Friendly for multi-tenant designs
Sharding support (jump consistent hashing)
Compatible with .NET Standard 2.0 and .NET Framework 4.6

Installing from NuGet

To install Streamstone via NuGet, run this command in NuGet package manager console:

PM> Install-Package Streamstone

Building from source

To build Streamstone binaries on Windows you will need to have Visual Studio 17 Update 3 or higher and .NET Core SDK 2.0 or higher. To build binaries on Linux use dotnet cli tooling (ie dotnet build).

Running unit tests

Windows/Linux/MacOs

Use Azurite npm package to run tests and example app against an emulated table storage service.

WARNING: Azurite doesn't fully emulate Azure Table Storage functionality so you may have some of the tests failing.

NOTE: Alternatively, you could run against real Azure by setting storage account connection string to Streamstone-Test-Storage user-level environment variable.

Design

Streamstone is just a thin layer (library, not a server) on top of Windows Azure Table Storage. It implements low-level mechanics for dealing with event streams, and all heavy-weight lifting is done by underlying provider.

The api is stateless and all exposed objects are immutable, once fully constructed. Streamstone doesn't dictate payload serialization protocol, so you are free to choose any protocol you want.

Optimistic concurrency is implemented by always including stream header entity with every write, making it impossible to append to a stream without first having a latest Etag. Duplicate event detection is done by automatically creating additional entity for every event, with RowKey value set to a unique identifier of a source event (consistent secondary index).

Schema

Usage

Essentials

Provisioning stream [see]
Opening stream [see]
Writing to stream [see]
Reading from stream [see]
Additional entity includes [see]
Optimistic concurrency [see]
Handling duplicate events [see]
Custom stream metadata [see]
Virtual partitions [see]

Application

Implementing stream directory [see]
Using snapshots [see]
Creating projections [see]
Querying events [see]

Demo

Classic Greg Young's CQRS demo using Streamstone [repo]
Using Streamstone in stateful applications. Event-sourced actors for Project Orleans [see]

Limitations

While Streamstone allows you to pass any number of events to Stream.Write, the max batch size limit imposed by Azure Table Storage is 100 entities, therefore:

The batch will be automatically flushed for every 99 events (100 - 1 header entity)
The batch will be automatically flushed for every 49 events with id being set (100/2 - 1 header entity)
You will get back InvalidOperationException when trying to write an event which together with its includes is over max batch size limit
The actual size in bytes of event payload is not taken into account, so all limitations outlined below still apply

Other limitations of the underlying Azure Table Storage API:

Maximum size of batch is 4MB
Maximum size of entity is 1 MB
Maximum size of property is 64Kb
Maximum length of property name is 255 chars
An entity can have up to 255 custom properties

WATS limitations on MSDN

Community

License

MIT License