This repository houses the Optimizely Agent service for use with Optimizely Feature Experimentation and Optimizely Full Stack (legacy).

Optimizely Feature Experimentation is an A/B testing and feature management tool for product development teams that enables you to experiment at every step. Using Optimizely Feature Experimentation allows for every feature on your roadmap to be an opportunity to discover hidden insights. Learn more at Optimizely.com, or see the developer documentation.

Optimizely Rollouts is free feature flags for development teams. You can easily roll out and roll back features in any application without code deploys, mitigating risk for every feature on your roadmap.

Refer to the Agent's developer documentation for detailed instructions on getting started with using the SDK.

Optimizely Agent is implemented in Golang. Golang version 1.21.0+ is required for developing and compiling from source. Installers and binary archives for most platforms can be downloaded directly from the Go downloads page.

Once Go is installed, the Optimizely Agent can be started via the following make command:

make setup
make run

This will start the Optimizely Agent with the default configuration in the foreground.

A helper script is available under scripts/build.ps1 to automate compiling Agent in a Windows environment. The script will download and install both Git and Golang and then attempt to compile Agent. Open a Powershell terminal and run

Set-ExecutionPolicy -ExecutionPolicy Unrestricted -Scope CurrentUser

.\scripts\build.ps1

.\bin\optimizely.exe

If you have Docker installed, Optimizely Agent can be started as a container. First pull the Docker image with:

docker pull optimizely/agent

By default this will pull the "latest" tag. You can also specify a specific version of Agent by providing the version as a tag to the docker command:

docker pull optimizely/agent:X.Y.Z

Then run the docker container with:

docker run -p 8080:8080 --env OPTIMIZELY_LOG_PRETTY=true --env OPTIMIZELY_SERVER_HOST=0.0.0.0 --env OPTIMIZELY_SERVER_ALLOWEDHOSTS=127.0.0.1 optimizely/agent

This will start Agent in the foreground and expose the container API port 8080 to the host.

Note that when a new version is released, 2 images are pushed to dockerhub, they are distinguished by their tags:

• :latest (same as :X.Y.Z)
• :alpine (same as :X.Y.Z-alpine)

The difference between latest and alpine is that latest is built FROM scratch while alpine is FROM alpine.

• latest Dockerfile
• alpine Dockerfile

Optimizely Agent configuration can be overridden by a yaml configuration file provided at runtime.

By default the configuration file will be sourced from the current active directory e.g. ./config.yaml. Alternative configuration locations can be specified at runtime via environment variable or command line flag.

OPTIMIZELY_CONFIG_FILENAME=config.yaml make run

The default configuration can be found here.

Below is a comprehensive list of available configuration properties.

More information about configuring Agent can be found in the Advanced Configuration Notes.

The core API is implemented as a REST service configured on it's own HTTP listener port (default 8080). The full API specification is defined in an OpenAPI 3.0 (aka Swagger) spec.

Each request made into the API must include a X-Optimizely-SDK-Key in the request header to identify the context the request should be evaluated. The SDK key maps to a unique Optimizely Project and Environment allowing multiple Environments to be serviced by a single Agent. For a secure environment, this header should include both the SDK key and the datafile access token separated by a colon. For example, if SDK key is my_key and datafile access token is my_token then set header's value to my_key:my_token.

CORS can be enabled for the core API service by setting the the appropriate cors properties.

api:
  cors:
    allowedMethods:
      - "HEAD"
      - "GET"
      - "POST"
      - "OPTIONS"

For more advanced options please refer to the go-chi/cors middleware documentation.

NOTE: To avoid any potential security issues, and reduce risk to your data it's recommended that authentication is enabled alongside CORS.

The webhook listener used to receive inbound Webhook requests from optimizely.com. These webhooks enable PUSH style notifications triggering immediate project configuration updates. The webhook listener is configured on its own port (default: 8085) since it can be configured to select traffic from the internet.

To accept webhook requests Agent must be configured by mapping an Optimizely Project Id to a set of SDK keys along with the associated secret used for validating the inbound request. An example webhook configuration can be found in the the provided config.yaml.

When running Agent in High Availability (HA) mode, it's important to ensure that all nodes are updated promptly when a webhook event (datafile updated) is received. By default, only one Agent node or instance will receive the webhook notification. A pub-sub system can be used to ensure this.

Redis, a powerful in-memory data structure store, can be used as a relay to propagate the datafile webhook event to all other nodes in the HA setup. This ensures that all nodes are notified about the event and can update their datafiles accordingly.

To set up Redis as a relay, you need to enable the datafile synchronization in your Optimizely Agent configuration. The PubSub feature of Redis is used to publish the webhook notifications to all subscribed Agent nodes.

Here's an example of how you can enable the datafile synchronization with Redis:

## synchronization should be enabled when features for multiple nodes like notification streaming are deployed
synchronization:
    pubsub:
        redis:
            host: "localhost:6379"
            password: ""
            database: 0
    ## if datafile synchronization is enabled, then for each webhook API call
    ## the datafile will be sent to all available replicas to achieve better eventual consistency
    datafile:
        enable: true
        default: "redis"

The Admin API provides system information about the running process. This can be used to check the availability of the service, runtime information and operational metrics. By default the admin listener is configured on port 8088.

The /info endpoint provides basic information about the Optimizely Agent instance.

Example Request:

curl localhost:8088/info

Example Response:

{
  "version": "v0.10.0",
  "author": "Optimizely Inc.",
  "app_name": "optimizely"
}

The /health endpoint is used to determine service availability.

Example Request:

curl localhost:8088/health

Example Response:

{
  "status": "ok"
}

Agent will return a HTTP 200 - OK response if and only if all configured listeners are open and all external dependent services can be reached. A non-healthy service will return a HTTP 503 - Unavailable response with a descriptive message to help diagnose the issue.

This endpoint can used when placing Agent behind a load balancer to indicate whether a particular instance can receive inbound requests.

The /metrics endpoint exposes telemetry data of the running Optimizely Agent.

Currently, Agent exposes two type of metrics data (expvar or prometheus) based on user's input. By default, expvar metrics will be used. To configure this, update config.yaml or set the value of the environment variable OPTIMIZELY_ADMIN_METRICSTYPE. Supported values are expvar (default) & promethues.

##
## admin service configuration
##
admin:
    ## http listener port
    port: "8088"
    ## metrics package to use
    ## supported packages are expvar and prometheus
    ## default is expvar
    metricsType: ""
    ## metricsType: "promethues" ## for prometheus metrics

The core runtime metrics are exposed via the Go expvar package. Documentation for the various statistics can be found as part of the mstats package.

Example Request:

curl localhost:8088/metrics

Example Response:

{
    "cmdline": [
        "bin/optimizely"
    ],
    "memstats": {
        "Alloc": 924136,
        "TotalAlloc": 924136,
        "Sys": 71893240,
        "Lookups": 0,
        "Mallocs": 4726,
        "HeapAlloc": 924136,
        ...
        "Frees": 172
    },
    ...
}

Custom metrics are also provided for the individual service endpoints and follow the pattern of:

"timers.<metric-name>.counts": 0,
"timers.<metric-name>.responseTime": 0,
"timers.<metric-name>.responseTimeHist.p50": 0,
"timers.<metric-name>.responseTimeHist.p90": 0,
"timers.<metric-name>.responseTimeHist.p95": 0,
"timers.<metric-name>.responseTimeHist.p99": 0,

Optimizely Agent also supports Prometheus metrics. Prometheus is an open-source toolkit for monitoring and alerting. You can use it to collect and visualize metrics in a time-series database.

To access the Prometheus metrics, you can use the /metrics endpoint with a Prometheus server. The metrics are exposed in a format that Prometheus can scrape and aggregate.

Example Request:

curl localhost:8088/metrics

This will return a plain text response in the Prometheus Exposition Format, which includes all the metrics that Prometheus is currently tracking.

Please note that you need to configure your Prometheus server to scrape metrics from this endpoint.

For more information on how to use Prometheus for monitoring, you can refer to the official Prometheus documentation.

Example Response:

...
# HELP promhttp_metric_handler_requests_total Total number of scrapes by HTTP status code.
# TYPE promhttp_metric_handler_requests_total counter
promhttp_metric_handler_requests_total{code="200"} 1
promhttp_metric_handler_requests_total{code="500"} 0
promhttp_metric_handler_requests_total{code="503"} 0
# HELP timer_decide_hits 
# TYPE timer_decide_hits counter
timer_decide_hits 1
# HELP timer_decide_response_time 
# TYPE timer_decide_response_time counter
timer_decide_response_time 658.109
# HELP timer_decide_response_time_hist 
# TYPE timer_decide_response_time_hist histogram
timer_decide_response_time_hist_bucket{le="0.005"} 0
timer_decide_response_time_hist_bucket{le="0.01"} 0
timer_decide_response_time_hist_bucket{le="0.025"} 0
timer_decide_response_time_hist_bucket{le="0.05"} 0
timer_decide_response_time_hist_bucket{le="0.1"} 0
timer_decide_response_time_hist_bucket{le="0.25"} 0
timer_decide_response_time_hist_bucket{le="0.5"} 0
timer_decide_response_time_hist_bucket{le="1"} 0
timer_decide_response_time_hist_bucket{le="2.5"} 0
timer_decide_response_time_hist_bucket{le="5"} 0
timer_decide_response_time_hist_bucket{le="10"} 0
timer_decide_response_time_hist_bucket{le="+Inf"} 1
timer_decide_response_time_hist_sum 658.109
timer_decide_response_time_hist_count 1
# HELP timer_track_event_hits 
# TYPE timer_track_event_hits counter
timer_track_event_hits 1
# HELP timer_track_event_response_time 
# TYPE timer_track_event_response_time counter
timer_track_event_response_time 0.356334
# HELP timer_track_event_response_time_hist 
# TYPE timer_track_event_response_time_hist histogram
timer_track_event_response_time_hist_bucket{le="0.005"} 0
timer_track_event_response_time_hist_bucket{le="0.01"} 0
timer_track_event_response_time_hist_bucket{le="0.025"} 0
timer_track_event_response_time_hist_bucket{le="0.05"} 0
timer_track_event_response_time_hist_bucket{le="0.1"} 0
timer_track_event_response_time_hist_bucket{le="0.25"} 0
timer_track_event_response_time_hist_bucket{le="0.5"} 1
timer_track_event_response_time_hist_bucket{le="1"} 1
timer_track_event_response_time_hist_bucket{le="2.5"} 1
timer_track_event_response_time_hist_bucket{le="5"} 1
timer_track_event_response_time_hist_bucket{le="10"} 1
timer_track_event_response_time_hist_bucket{le="+Inf"} 1
timer_track_event_response_time_hist_sum 0.356334
timer_track_event_response_time_hist_count 1
...

Agent exposes the runtime profiling data in the format expected by the pprof visualization tool.

You can use the pprof tool to look at the heap profile:

go tool pprof http://localhost:6060/debug/pprof/heap

Or to look at a 5-second CPU profile: (higher durations require configuring the server.writeTimeout)

go tool pprof http://localhost:6060/debug/pprof/profile?seconds=5

Or to look at the goroutine blocking profile, after setting runtime.blockProfileRate in the configuration:

go tool pprof http://localhost:8088/debug/pprof/block

Or to collect a 5-second execution trace:

wget "http://localhost:8088/debug/pprof/trace?seconds=5"

Or to look at the holders of contended mutexes, after setting runtime.mutexProfileFraction in your configuration:

go tool pprof http://localhost:6060/debug/pprof/mutex

To view all available profiles can be found at http://localhost:8088/debug/pprof/ in your browser.

Optimizely Agent can be extended through the use of plugins. Plugins are distinct from the standard Agent packages to provide a namespaced environment for custom logic. Plugins must be compiled as part of the Agent distribution and are enabled through configuration.

• httplog - Adds HTTP request logging based on go-chi/httplog.

• UserProfileService - Adds UserProfileService.

• ODPCache - Adds ODP Cache.

Optimizely Agent supports authorization workflows based on OAuth and JWT standards, allowing you to protect access to its API and Admin interfaces. For details, see Authorization Guide.

Just as you can use Notification Listeners to subscribe to events of interest with Optimizely SDKs, you can use the Notifications endpoint to subscribe to events in Agent. For more information, see the Notifications Guide.

When the Agent is operating in High Availability (HA) mode, you need to enable notification synchronization to get notifications from all nodes in an HA setup. A PubSub system (Redis) is now used to ensure consistent retrieval of notification events across all nodes in an HA setup. Here's an example of how you can enable the notification synchronization with Redis:

## synchronization should be enabled when features for multiple nodes like notification streaming are deployed
synchronization:
    pubsub:
        redis:
            host: "localhost:6379"
            password: ""
            database: 0
    ## if notification synchronization is enabled, then the active notification event-stream API
    ## will get the notifications from available replicas
    notification:
        enable: true
        default: "redis"

Following best practice for go project layout as defined here

• api - OpenAPI/Swagger specs, JSON schema files, protocol definition files.
• bin - Compiled application binaries.
• cmd - Main applications for this project.
• config - Application configuration.
• docs - User documentation files.
• pkg - Library code that can be used by other applications.
• plugins - Plugin libraries for extending Agent functionality.
• scripts - Scripts to perform various build, install, analysis, etc operations.

The following make targets can be used to build and run the application:

• build - builds optimizely and installs binary in bin/optimizely
• clean - runs go clean and removes the bin/ dir
• cover - runs test suite with coverage profiling
• cover-html - generates test coverage html report
• setup - installs all dev and ci dependencies, but does not install golang
• lint - runs golangci-lint linters defined in .golangci.yml file
• run - builds and executes the optimizely binary
• test - recursively tests all .go files

This software is used with additional code that is separately downloaded by you. These components are subject to their own license terms which you should review carefully.

Gohistogram (c) 2013 VividCortex License (MIT): https://github.com/VividCortex/gohistogram

Chi (c) 2015-present Peter Kieltyka (https://github.com/pkieltyka), Google Inc. License (MIT): https://github.com/go-chi/chi

chi-render (c) 2016-Present https://github.com/go-chi ‑ authors License (MIT): https://github.com/go-chi/render

hostrouter (c) 2016-Present https://github.com/go-chi - authors License (MIT): https://github.com/go-chi/hostrouter

go-chi/cors (c) 2014 Olivier Poitrey License (MIT): https://github.com/go-chi/cors

go-kit (c) 2015 Peter Bourgon License (MIT): https://github.com/go-kit/kit

guuid (c) 2009,2014 Google Inc. All rights reserved. License (BSD 3-Clause): https://github.com/google/uuid

optimizely go sdk (c) 2016-2017, Optimizely, Inc. and contributors License (Apache 2): https://github.com/optimizely/go-sdk

concurrent-map (c) 2014 streamrail License (MIT): https://github.com/orcaman/concurrent-map

zerolog (c) 2017 Olivier Poitrey License (MIT): https://github.com/rs/zerolog

viper (c) 2014 Steve Francia License (MIT): https://github.com/spf13/viper

testify (c) 2012-2018 Mat Ryer and Tyler Bunnell License (MIT): https://github.com/stretchr/testify

net (c) 2009 The Go Authors License (BSD 3-Clause): https://github.com/golang/net

sync (c) 2009 The Go Authors License (BSD 3-Clause): https://github.com/golang/sync

statik (c) 2014 rakyll License (Apache 2): https://github.com/rakyll/statik v0.1.7

sys (c) 2009 The Go Authors License (BSD 3-Clause): https://github.com/golang/sys

opentelemetry-go Copyright The OpenTelemetry Authors License (Apache-2.0): https://github.com/open-telemetry/opentelemetry-go

prometheus client_golang Copyright 2015 The Prometheus Authors License (Apache-2.0): https://github.com/prometheus/client_golang

Copyright 2019-present, Optimizely, Inc. and contributors

Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at http://www.apache.org/licenses/LICENSE-2.0 Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.