A consensus register collection is a distributed data structure (DDS), which holds a set of registers and their versions generated during concurrent updates. In the simplest definition, two updates on a single register are concurrent if there is no causal relationship between them (i.e., neither knows about the other). On such cases of concurrent updates, a register internally stores all possible versions of a value.
When taking a dependency on a Fluid Framework library's public APIs, we recommend using a ^
(caret) version range, such as ^1.3.4
.
While Fluid Framework libraries may use different ranges with interdependencies between other Fluid Framework libraries,
library consumers should always prefer ^
.
If using any of Fluid Framework's unstable APIs (for example, its beta
APIs), we recommend using a more constrained version range, such as ~
.
To get started, install the package by running the following command:
npm i @fluidframework/register-collection
This package leverages package.json exports to separate its APIs by support level. For more information on the related support guarantees, see API Support Levels.
To access the public
(SemVer) APIs, import via @fluidframework/register-collection
like normal.
To access the legacy
APIs, import via @fluidframework/register-collection/legacy
.
API documentation for @fluidframework/register-collection is available at https://fluidframework.com/docs/apis/register-collection.
In distributed systems literaure, detecting concurrency requires some form of logical/phsical clock. A popular technique used in replicated databases such as dynamodb is called version vectors where each key stores a collection of [time, value]
tuples. time
is essentially a reference clock used to decide concurrency amongst updates. Each update to a key includes the time
, essentially to indicate how caught up
the replica was during that update.
In Fluid, each operation contains a referenceSequenceNumber (refSeq
), which essenially refers to how caught up the client was (in terms of sequence number) during that update. We can use this property to implement a similar concurreny model. Mathematically, if an update has a refSeq N
, it can overwrite/discard any other prior values with sequenceNumber (seq) <= N
. It is safe to do so because the client must have seen all those updates before posting it's own update. Hence this update is not concurrent with those overwritten updates. However, the update is still concurrent with any other update with seq > N
. Therefore those versions are still kept.
Below are the policies that the DDS implements:
-
Versions: Returns all stored concurrent versions. App is responsible for conflict resolution. Amazon's shopping cart policy based on dynamodb is a popular example of this policy.
-
LWW: The last write to a key always overwrites any prior writes (aka last write win policy). This is exactly same as Fluid's Shared Map policy.
-
Atomic: The policy follows the same semantics of a shared distributed lock. Amongst all concurrent updates, only the first writer wins. In distributed systems literature, the register update policy is called
Atomic
. This behavior requires a linearizable register. A linearizable register behaves as if there is only a single copy of the data, and that every operation appears to take effect atomically at one point in time. This definition implies that operations are executed in an well-defined order. On a concurrent update, we perform acompare-and-set
operation, where we compare a register's storedseq
with the incomingrefSeq
. The earliest operation overwriting the storedseq
wins since every client reaches to an agreement on the value. Hence we can safely return the first value.
These are the platform requirements for the current version of Fluid Framework Client Packages. These requirements err on the side of being too strict since within a major version they can be relaxed over time, but not made stricter. For Long Term Support (LTS) versions this can require supporting these platforms for several years.
It is likely that other configurations will work, but they are not supported: if they stop working, we do not consider that a bug. If you would benefit from support for something not listed here, file an issue and the product team will evaluate your request. When making such a request please include if the configuration already works (and thus the request is just that it becomes officially supported), or if changes are required to get it working.
- NodeJs ^20.10.0 except that we will drop support for it when NodeJs 20 loses its upstream support on 2026-04-30, and will support a newer LTS version of NodeJS (22) at least 1 year before 20 is end-of-life. This same policy applies to NodeJS 22 when it is end of life (2027-04-30).
- Modern browsers supporting the es2022 standard library: in response to asks we can add explicit support for using babel to polyfill to target specific standards or runtimes (meaning we can avoid/remove use of things that don't polyfill robustly, but otherwise target modern standards).
- TypeScript 5.4:
- All
strict
options are supported. strictNullChecks
is required.- Configuration options deprecated in 5.0 are not supported.
exactOptionalPropertyTypes
is currently not fully supported. If used, narrowing members of Fluid Framework types types usingin
,Reflect.has
,Object.hasOwn
orObject.prototype.hasOwnProperty
should be avoided as they may incorrectly excludeundefined
from the possible values in some cases.
- All
- webpack 5
- We are not intending to be prescriptive about what bundler to use. Other bundlers which can handle ES Modules should work, but webpack is the only one we actively test.
Node16
, NodeNext
, or Bundler
resolution should be used with TypeScript compilerOptions to follow the Node.js v12+ ESM Resolution and Loading algorithm.
Node10 resolution is not supported as it does not support Fluid Framework's API structuring pattern that is used to distinguish stable APIs from those that are in development.
-
ES Modules: ES Modules are the preferred way to consume our client packages (including in NodeJs) and consuming our client packages from ES Modules is fully supported.
-
CommonJs: Consuming our client packages as CommonJs is supported only in NodeJS and only for the cases listed below. This is done to accommodate some workflows without good ES Module support. If you have a workflow you would like included in this list, file an issue. Once this list of workflows motivating CommonJS support is empty, we may drop support for CommonJS one year after notice of the change is posted here.
- Testing with Jest (which lacks stable ESM support due to unstable APIs in NodeJs)
There are many ways to contribute to Fluid.
- Participate in Q&A in our GitHub Discussions.
- Submit bugs and help us verify fixes as they are checked in.
- Review the source code changes.
- Contribute bug fixes.
Detailed instructions for working in the repo can be found in the Wiki.
This project has adopted the Microsoft Open Source Code of Conduct. For more information see the Code of Conduct FAQ or contact opencode@microsoft.com with any additional questions or comments.
This project may contain Microsoft trademarks or logos for Microsoft projects, products, or services. Use of these trademarks or logos must follow Microsoft’s Trademark & Brand Guidelines. Use of Microsoft trademarks or logos in modified versions of this project must not cause confusion or imply Microsoft sponsorship.
Not finding what you're looking for in this README? Check out fluidframework.com.
Still not finding what you're looking for? Please file an issue.
Thank you!
This project may contain Microsoft trademarks or logos for Microsoft projects, products, or services.
Use of these trademarks or logos must follow Microsoft's Trademark & Brand Guidelines.
Use of Microsoft trademarks or logos in modified versions of this project must not cause confusion or imply Microsoft sponsorship.