Skip to main content

Unleash Edge

Generated content

This document was generated from the README in the Unleash Edge GitHub repository.

Unleash Edge Documentation MIT licensed Dependency Status CI Coverage Status downloads

Warning: Unleash Edge requires Unleash v4.15.0 or higher

Unleash Edge is the successor to the Unleash Proxy.

Unleash Edge sits between the Unleash API and your SDKs and provides a cached read-replica of your Unleash instance. This means you can scale up your Unleash instance to thousands of connected SDKs without increasing the number of requests you make to your Unleash instance.

Unleash Edge offers two important features:

  • Performance: Unleash Edge caches in memory and can run close to your end-users. A single instance can handle tens to hundreds of thousands of requests per second.
  • Resilience: Unleash Edge is designed to survive restarts and operate properly even if you lose connection to your Unleash server.

Unleash Edge is built to help you scale Unleash.

  • If you're looking for the easiest way to connect your client SDKs you can check out our Frontend API.
  • If you're looking to learn how to scale your own feature flag system why not check out our recommendations for building and scaling feature flags

Migrating to Edge from the Proxy

For more info on migrating, check out the migration guide that details the differences between Edge and the Proxy and how to achieve similar behavior in Edge.

Running Unleash Edge

Edge provides a range of powerful ways in which you can run it. For a standard production configuration we recommend the following:

  • Run in Edge mode: Edge mode connects to your upstream Unleash and syncs feature flags and tokens. This should be the default mode that you choose for most Edge configurations in production.
  • Start Edge with initialization tokens: Edge mode allows you to specify a set of tokens on startup that Edge will use to hydrate data ahead of time. This means that Edge will have the data it requires to respond to frontend API requests and client API requests will not need to hydrate data on demand. If you are running Edge behind a load balancer and making use of the frontend API, setting startup tokens is necessary for predictable responses from Edge.
  • Choose a appropriate scope for your initialization tokens: We recommend using one wildcard token per environment. This gives more predictability over the resources that Edge will use at runtime. If Edge needs to run in a sensitive context, starting Edge with tokens that are scoped to all the projects that downstream SDKs are expected to consume is okay.

Unleash Edge is compiled to a single binary. You can configure it by passing in arguments or setting environment variables.

Usage: unleash-edge [OPTIONS] <COMMAND>

edge Run in edge mode
offline Run in offline mode
help Print this message or the help of the given subcommand(s)

-p, --port <PORT>
Which port should this server listen for HTTP traffic on [env: PORT=] [default: 3063]
-i, --interface <INTERFACE>
Which interfaces should this server listen for HTTP traffic on [env: INTERFACE=] [default:]
-b, --base-path <BASE_PATH>
Which base path should this server listen for HTTP traffic on [env: BASE_PATH=] [default: ]
-w, --workers <WORKERS>
How many workers should be started to handle requests. Defaults to number of physical cpus [env: WORKERS=] [default: number of physical cpus]
Should we bind TLS [env: TLS_ENABLE=]
--tls-server-key <TLS_SERVER_KEY>
Server key to use for TLS [env: TLS_SERVER_KEY=] (Needs to be a path to a file)
--tls-server-cert <TLS_SERVER_CERT>
Server Cert to use for TLS [env: TLS_SERVER_CERT=] (Needs to be a path to a file)
--tls-server-port <TLS_SERVER_PORT>
Port to listen for https connection on (will use the interfaces already defined) [env: TLS_SERVER_PORT=] [default: 3043]
--instance-id <INSTANCE_ID>
Instance id. Used for metrics reporting [env: INSTANCE_ID=] [default: Ulid::new()]
-a, --app-name <APP_NAME>
App name. Used for metrics reporting [env: APP_NAME=] [default: unleash-edge]
-h, --help
Print help
-l, --log-format <LOG_FORMAT>
Which log format should Edge use
[env: LOG_FORMAT=]
[default: `plain`]
Possible values: `plain`, `json`, `pretty`
--token-header <TOKEN_HEADER>
token header to use for edge authorization [env: TOKEN_HEADER=] [default: Authorization]

Built-in Health check

There is now (from 5.1.0) a subcommand named health which will ping your health endpoint and exit with status 0 provided the health endpoint returns 200 OK.


./unleash-edge health

will check an Edge process running on http://localhost:3063. If you're using base-path or the port variable you should use the -e --edge-url CLI arg (or the EDGE_URL environment variable) to tell the health checker where edge is running.

If you're hosting Edge with a self-signed certificate using the tls cli arguments, you should use the --ca-certificate-file <file_containing_your_ca_and_key_in_pem_format> flag (or the CA_CERTIFICATE_FILE environment variable) to allow the health checker to trust the self signed certificate.

Built-in Ready check

There is now (from 12.0.0) a subcommand named ready which will ping your ready endpoint and exit with status 0 provided the ready endpoint returns 200 OK and { status: "READY" }. Otherwise it will return status 1 and an error message to signal that Edge is not ready (it has not spoken to upstream or recovered from a persisted backup).


  • Edge not running:
$ ./unleash-edge ready
Error: Failed to connect to ready endpoint at http://localhost:3063/internal-backstage/ready. Failed with status None
$ echo $?
  • Edge running but not populated its feature cache yet (not spoken to upstream or restored from backup)
$ ./unleash-edge ready
Error: Ready check returned a different status than READY. It returned EdgeStatus { status: NotReady }
$ echo $?
  • Edge running and synchronized. I.e. READY
$ ./unleash-edge ready
$ echo $?

If you're hosting Edge with a self-signed certificate using the tls cli arguments, you should use the --ca-certificate-file <file_containing_your_ca_and_key_in_pem_format> flag (or the CA_CERTIFICATE_FILE environment variable) to allow the health checker to trust the self signed certificate.

Getting Unleash Edge

Unleash Edge is distributed as a binary and as a docker image.


  • The binary is downloadable from our Releases page.
  • We're currently building for linux x86_64, windows x86_64, darwin (OS X) x86_64 and darwin (OS X) aarch64 (M1/M2 macs)


  • The docker image gets uploaded to dockerhub and Github Package registry.
  • For dockerhub use the coordinates unleashorg/unleash-edge:<version>.
  • For Github package registry use the coordinates<version>
  • If you'd like to live on the edge (sic) you can use the tag edge. This is built from HEAD on each commit
  • When running the docker image, the same CLI arguments that's available when running the binary is available to your docker run command. To start successfully you will need to decide which mode you're running in.
    • If running in edge mode your command should be
      • docker run -p 3063:3063 -e UPSTREAM_URL=<YOUR_UNLEASH_INSTANCE> unleashorg/unleash-edge:<version> edge
    • If running in offline mode you will need to provide a volume containing your feature toggles file. An example is available inside the examples folder. To use this, you can use the command
      • docker run -v ./examples:/edge/data -p 3063:3063 -e BOOTSTRAP_FILE=/edge/data/features.json -e TOKENS='my-secret-123,another-secret-789' unleashorg/unleash-edge:<version> offline


If you have the Rust toolchain installed you can build a binary for the platform you're running by cloning this repo and running cargo build --release. This will give you an unleash-edge binary in ./target/release



Edge currently supports 2 different modes:

  • Edge - Connection to upstream node (Unleash instance or another Edge). Supports dynamic tokens, metrics and other advanced features;
  • Offline - No connection to upstream node. Full control of data and tokens;


Edge mode is the "standard" mode for Unleash Edge and the one you should default to in most cases. It connects to an upstream node, such as your Unleash instance, and uses that as the source of truth for feature toggles.

Other than connecting Edge directly to your Unleash instance, it's also possible to connect to another Edge instance ( daisy chaining). You can have as many Edge nodes as you'd like between the Edge node your clients are accessing and the Unleash server, and it's also possible for multiple nodes to connect to a single upstream one. Depending on your architecture and requirements this can be a powerful feature, offering you flexibility and scalability when planning your implementation.

This means that, in order to start up, Edge mode needs to know where the upstream node is. This is done by passing the --upstream-url command line argument or setting the UPSTREAM_URL environment variable.

By default, Edge mode uses an in-memory cache to store the features it fetches from the upstream node. However, you may want to use a more persistent storage solution. For this purpose, Edge supports either Redis or a backup file, which you can configure by passing in either the --redis-url or --backup_folder command line argument, respectively. On start-up, Edge checks whether the persistent backup option is specified, in which case it uses it to populate its internal caches. This can be useful when your Unleash server is unreachable.

Persistent storage is useful for high availability and resilience. If an Edge instance using persistent storage is restarted, it will not have to synchronize with upstream before it is ready. The persistent storage is only read on startup and will only be used until Edge is able to synchronize with upstream. Because of this, there is no need to purge the cache. After Edge has synchronized with an upstream, it will periodically save its in-memory state to the persistent storage.

Multiple Edge nodes can share the same Redis instance or backup folder. Failure to read from persistent storage will not prevent Edge from starting up. In this case an Edge instance will be ready only after it is able to connect with upstream.

Edge mode also supports dynamic tokens, meaning that Edge doesn't need a token to be provided when starting up. Once we make a request to the /api/client/features endpoint using a client token Edge will validate upstream and fetch its respective features. After that, it gets added to the list of known tokens that gets periodically synced, making sure it is a valid token and its features are up-to-date.

Even though Edge supports dynamic tokens, you still have the option of providing a token through the command line argument or environment variable. This way, since Edge already knows about your token at start up, it will sync your features for that token and should be ready for your requests right away (warm up / hot start).

Front-end tokens

Front-end tokens can also be used with /api/frontend and /api/proxy endpoints, however they are not allowed to fetch features upstream. In order to use these tokens correctly and make sure they return the correct information, it's important that the features they are allowed to access are already present in that Edge node's features cache. The easiest way to ensure this is by passing in at least one client token as one of the command line arguments, ensuring it has access to the same features as the front-end token you'll be using. If you're using a frontend token that doesn't have data in the node's feature cache, you will receive an HTTP Status code: 511 Network Authentication Required along with a body of which project and environment you will need to add a client token for.

Starting in edge mode

To see parameters available when running in this mode, run:

$ unleash-edge edge -h
Run in edge mode

Usage: unleash-edge edge [OPTIONS] --upstream-url <UPSTREAM_URL>

-u, --upstream-url <UPSTREAM_URL>
Where is your upstream URL. Remember, this is the URL to your instance, without any trailing /api suffix [env: UPSTREAM_URL=]
-r, --redis-url <REDIS_URL>
A URL pointing to a running Redis instance. Edge will use this instance to persist feature and token data and read this back after restart. Mutually exclusive with the --backup-folder option [env: REDIS_URL=]
-b, --backup-folder <BACKUP_FOLDER>
A path to a local folder. Edge will write feature and token data to disk in this folder and read this back after restart. Mutually exclusive with the --redis-url option [env: BACKUP_FOLDER=]
-m, --metrics-interval-seconds <METRICS_INTERVAL_SECONDS>
How often should we post metrics upstream? [env: METRICS_INTERVAL_SECONDS=] [default: 60]
-f, --features-refresh-interval-seconds <FEATURES_REFRESH_INTERVAL_SECONDS>
How long between each refresh for a token [env: FEATURES_REFRESH_INTERVAL_SECONDS=] [default: 10]
--token-revalidation-interval-seconds <TOKEN_REVALIDATION_INTERVAL_SECONDS>
How long between each revalidation of a token [env: TOKEN_REVALIDATION_INTERVAL_SECONDS=] [default: 3600]
-t, --tokens <TOKENS>
Get data for these client tokens at startup. Accepts comma-separated list of tokens. Hot starts your feature cache [env: TOKENS=]
-H, --custom-client-headers <CUSTOM_CLIENT_HEADERS>
Expects curl header format (-H <HEADERNAME>: <HEADERVALUE>) for instance `-H X-Api-Key: mysecretapikey` [env: CUSTOM_CLIENT_HEADERS=]
-s, --skip-ssl-verification
If set to true, we will skip SSL verification when connecting to the upstream Unleash server [env: SKIP_SSL_VERIFICATION=]
--pkcs8-client-certificate-file <PKCS8_CLIENT_CERTIFICATE_FILE>
Client certificate chain in PEM encoded X509 format with the leaf certificate first. The certificate chain should contain any intermediate certificates that should be sent to clients to allow them to build a chain to a trusted root [env: PKCS8_CLIENT_CERTIFICATE_FILE=]
--pkcs8-client-key-file <PKCS8_CLIENT_KEY_FILE>
Client key is a PEM encoded PKCS#8 formatted private key for the leaf certificate [env: PKCS8_CLIENT_KEY_FILE=]
--pkcs12-identity-file <PKCS12_IDENTITY_FILE>
Identity file in pkcs12 format. Typically this file has a pfx extension [env: PKCS12_IDENTITY_FILE=]
--pkcs12-passphrase <PKCS12_PASSPHRASE>
Passphrase used to unlock the pkcs12 file [env: PKCS12_PASSPHRASE=]
--upstream-certificate-file <UPSTREAM_CERTIFICATE_FILE>
Extra certificate passed to the client for building its trust chain. Needs to be in PEM format (crt or pem extensions usually are) [env: UPSTREAM_CERTIFICATE_FILE=]

-h, --help
Print help


Offline mode is useful when there is no connection to an upstream node, such as your Unleash instance or another Edge instance, or as a tool to make working with Unleash easier during development.

To use offline mode, you'll need a features file. The easiest way to get one is to download a JSON dump of a result from a query against an Unleash server on the /api/client/features endpoint. You can also use a hand rolled, human readable JSON version of the features file. Edge will automatically convert it to the API format when it starts up. Here's an example:

"featureOne": {
"enabled": true,
"variant": "variantOne"
"featureTwo": {
"enabled": false,
"variant": "variantTwo"
"featureThree": {
"enabled": true

The simplified JSON format should be an object with a key for each feature. You can force the result of is_enabled in your SDK by setting the enabled property, likewise can also force the result of get_variant by specifying the name of the variant you want. This format is primarily for development.

When using offline mode you must specify one or more tokens at startup. These tokens will let your SDKs access Edge. Tokens following the Unleash API format [project]:[environment].<somesecret> allow Edge to recognize the project and environment specified in the token, returning only the relevant features to the calling SDK. On the other hand, for tokens not adhering to this format, Edge will return all features if there is an exact match with any of the startup tokens.

To make local development easier, you can specify a reload interval in seconds (Since Unleash-Edge 10.0.x); this will cause Edge to reload the features file from disk every X seconds. This can be useful for local development.

Since offline mode does not connect to an upstream node, it does not support metrics or dynamic tokens.

To launch in this mode, run:

$ ./unleash-edge offline --help
Usage: unleash-edge offline [OPTIONS]

-b, --bootstrap-file <BOOTSTRAP_FILE> [env: BOOTSTRAP_FILE=]
-t, --tokens <TOKENS> [env: TOKENS=]
-r, --reload-interval <RELOAD_INTERVAL> [env: RELOAD_INTERVAL=]

Environments in offline mode

Currently, Edge does not support multiple environments in offline mode. All tokens added at startup will receive the same list of features passed in as the bootstrap argument. However, tokens in <project>:<environment>.<secret> format will still filter by project.


❗ Note: For Edge to correctly register SDK usage metrics, your Unleash instance must be v5.9.0. ! Note: If you're daisy chaining you will need at least Edge 17.0.0 upstream of any Edge 19.0.0 to preserve metrics.

Since Edge is designed to avoid overloading its upstream, Edge gathers and accumulates usage metrics from SDKs for a set interval (METRICS_INTERVAL_SECONDS) before posting a batch upstream. This reduces load on Unleash instances down to a single call every interval, instead of every single client posting to Unleash for updating metrics. Unleash instances running on versions older than 4.22 are not able to handle the batch format posted by Edge, which means you won't see any metrics from clients connected to an Edge instance until you're able to update to 4.22 or newer.


Unleash Edge adheres to Semantic Versioning (SemVer) on the API and CLI layers. If you're using Unleash Edge as a library in your projects, be cautious, internal codebase changes, which might occur in any version release (including minor and patch versions), could potentially break your implementation.


Unleash Edge will scale linearly with CPU. There are k6 benchmarks in the benchmark folder. We've already got some initial numbers from hey.

Do note that the number of requests Edge can handle does depend on the total size of your toggle response. That is, Edge is faster if you only have 10 toggles with 1 strategy each, than it will be with 1000 toggles with multiple strategies on each. Benchmarks here were run with data fetched from the Unleash demo instance (roughly 100kB (350 features / 200 strategies)) as well as against a small dataset of 5 features with one strategy on each.

Edge was started using docker run --cpus="<cpu>" --memory=128M -p 3063:3063 -e UPSTREAM_URL=<upstream> -e TOKENS="<client token>" unleashorg/unleash-edge:edge -w <number of cpus rounded up to closest integer> edge

Then we run hey against the proxy endpoint, evaluating toggles

Large Dataset (350 features (100kB))

$ hey -z 10s -H "Authorization: <frontend token>" http://localhost:3063/api/frontend`
CPUMemoryRPSEndpointp95Data transferred
0.16.7 Mi600/api/frontend103ms76Mi
16.7 Mi6900/api/frontend7.4ms866Mi

and against our client features endpoint.

$ hey -z 10s -H "Authorization: <client token>" http://localhost:3063/api/client/features
CPUMemory observedRPSEndpointp95Data transferred
0.111 Mi309/api/client/features199ms300 Mi
111 Mi3236/api/client/features16ms3 Gi
411 Mi12815/api/client/features4.5ms14 Gi
817 Mi23207/api/client/features2.7ms26 Gi

Small Dataset (5 features (2kB))

$ hey -z 10s -H "Authorization: <frontend token>" http://localhost:3063/api/frontend`
CPUMemoryRPSEndpointp95Data transferred
0.14.3 Mi3673/api/frontend93ms9Mi
16.7 Mi39000/api/frontend1.6ms80Mi
46.9 Mi100020/api/frontend600μs252Mi
812.5 Mi141090/api/frontend600μs324Mi

and against our client features endpoint.

$ hey -z 10s -H "Authorization: <client token>" http://localhost:3063/api/client/features
CPUMemory observedRPSEndpointp95Data transferred
0.14 Mi3298/api/client/features92ms64 Mi
14 Mi32360/api/client/features2ms527Mi
411 Mi95838/api/client/features600μs2.13 Gi
817 Mi129381/api/client/features490μs2.87 Gi

Why choose Unleash Edge over the Unleash Proxy?

Edge offers a superset of the same feature set as the Unleash Proxy and we've made sure it offers the same security and privacy features.

However, there are a few notable differences between the Unleash Proxy and Unleash Edge:

  • Unleash Edge is built to be light and fast, it handles an order of magnitude more requests per second than the Unleash Proxy can, while using two orders of magnitude less memory.
  • All your Unleash environments can be handled by a single instance, no more running multiple instances of the Unleash Proxy to handle both your development and production environments.
  • Backend SDKs can connect to Unleash Edge without turning on experimental feature flags.
  • Unleash Edge is smart enough to dynamically resolve the tokens you use to connect to it against the upstream Unleash instance. This means you don't have to worry about knowing in advance what tokens your SDKs use - if you want to swap out the Unleash token your SDK uses, this can be done without ever restarting or worrying about Unleash Edge. Unleash Edge will only collect and cache data for the environments and projects you use.


You can adjust the RUST_LOG environment variable to see more verbose log output. For example, in order to get logs originating directly from Edge but not its dependencies, you can raise the default log level from error to warning and set Edge to debug, like this:

RUST_LOG="warn,unleash_edge=debug" ./unleash-edge #<command>

See more about available logging and log levels at


Missing metrics in upstream server

Possible reasons

  • Old edge version. In order to guarantee metrics on newer Unleash versions, you will need to be using Edge v17.0.0 or newer
  • Old SDK clients. We've seen some clients, particularly early Python (1.x branch) as well as earlier .NET SDKs (we recommend you use 4.1.5 or newer) struggle to post metrics with the strict headers Edge requires.


See our Contributors guide as well as our development-guide

This content was generated on