Storage

This document describes how to configure Flipt’s storage backend mechanisms.

Relational Database

Flipt supports SQLite, PostgreSQL, CockroachDB and MySQL relational databases.

As of v1.29.0, Flipt also supports libSQL which is a fork of SQLite that can be accessed over a network via sqld.

SQLite is enabled by default for simplicity, however, you should use PostgreSQL, MySQL, or CockroachDB if you intend to run multiple copies of Flipt in a high availability configuration.

The database connection can be configured as follows:

SQLite

The default location of the SQLite database is /var/opt/flipt/flipt.db on Linux and ~/Library/Application Support/flipt/flipt.db on macOS (since v1.26.0).

db:
  # file: informs flipt to use SQLite
  url: file:/var/opt/flipt/flipt.db

LibSQL

See our libSQL Example for a working example of how to use libSQL with Flipt.

Local

db:
  # libsql: informs flipt to use libSQL
  url: libsql://file:/var/opt/flipt/flipt.db

Remote

If using Turso you must use a database auth token to access the database.

db: # http(s): informs flipt to use libSQL over HTTP(s) via sqld/Turso
  url: https://db-[your-github-name].turso.io?authToken=[your-auth-token]

PostgreSQL

db:
  url: postgres://postgres@localhost:5432/flipt?sslmode=disable

CockroachDB

db:
  url: cockroach://root@localhost:26257/flipt?sslmode=disable

MySQL

db:
  url: mysql://mysql@localhost:3306/flipt

Migrations

From time to time the Flipt database must be updated with new schema. To accomplish this, Flipt includes a migrate command that will run any pending database migrations for you.

If Flipt is started and there are pending migrations, you will see the following error in the console:

migrations pending, please backup your database and run `flipt migrate`

If it’s your first run of Flipt, all migrations will automatically be run before starting the Flipt server.

You should backup your database before running flipt migrate to ensure that no data is lost if an error occurs during migration.

If running Flipt via Docker, you can run the migrations in a separate container before starting Flipt by running:

docker run -it -v $HOME/flipt:/var/opt/flipt flipt/flipt:latest /bin/sh -c './flipt migrate'

$HOME/flipt is just used as an example, you can use any directory you would like on the host.

If you don’t use mounted volumes to persist your data, your data will be lost when the migration container exits, having no effect on your Flipt instance!

Declarative

The following backend types are designed to support declarative management of feature flag state via a well-known file format. In particular, they’re designed to support GitOps practices with minimal external dependencies.

The current four declarative backend types include:

• Local
• Git
• Object
• OCI

The local backend has been primarily developed to support a local development experience, whereas, the git, object and oci backends are intended for production use.

Read Only Mode

Once enabled, all declarative backends put the Flipt API and UI into a read-only mode that prevents Flipt from writing to the backend. This is useful for production environments where you want to ensure that flag state is only managed via the configured backend.

You can also put Flipt into read-only mode by setting the FLIPT_STORAGE_READ_ONLY environment variable to true, or setting storage.read_only to true in your configuration.

Local

The purpose of this backend type is to support serving Flipt flag state directly from your local filesystem. You can simply specify a relative or absolute directory in order to start a local Flipt instance and serve flag state. This is particularly useful for local development and validation of flag state changes.

Flipt will periodically rebuild its state from the local disk every 10 seconds.

• Environment Variable
• Configuration Yaml

FLIPT_STORAGE_TYPE="local"
FLIPT_STORAGE_LOCAL_PATH="."

Git

The git type backend is used to configure a target Git repository and Git reference to source feature flag state. The configuration contains fields for addressing the repository, configuring the target reference as well as adding authentication credentials.

Once a target repository and reference are configured, Flipt will poll the source repository on a periodic cadence. This cadence is also configurable and defaults to 30 seconds.

Flipt will follow the configured reference (e.g. branch name) and keep up to date with new changes.

• Environment Variable
• Configuration Yaml

FLIPT_STORAGE_TYPE="git"
FLIPT_STORAGE_GIT_REPOSITORY="https://github.com/predictab.le/config.git"
FLIPT_STORAGE_GIT_REF="main"
FLIPT_STORAGE_GIT_POLL_INTERVAL="30s"
# for private repository access
FLIPT_STORAGE_GIT_AUTHENTICATION_BASIC_USERNAME=...
FLIPT_STORAGE_GIT_AUTHENTICATION_BASIC_PASSWORD=...

Authentication

Authentication enables the ability to leverage private Git repositories as flag state backends. The git type backend supports both basic, token and ssh based authentication schemes.

GitHub

When using GitHub and their PATs (Personal Access Tokens), basic authentication should be used. GitHub expects you to supply a valid username and provide your PAT as the password parameter.

storage:
  type: git
  git:
    repository: "https://github.com/predictab.le/config.git"
    ref: "main"
    poll_interval: "30s"
    authentication:
      basic:
        username: < username >
        password: < github-personal-access-token >

SSH

In order to configure Flipt with SSH, you will need to generate an SSH key-pair and configure your repository provider with the public key. GitHub has some excellent documentation regarding how to generate and install you credentials here.

Once you have your private key credentials you will need to configure Flipt to use them. This can be done via the storage.git.authentication.ssh configuration section:

storage:
  type: git
  git:
    repository: [e-mail]:flipt-io/some-private-repo.git
    authentication:
      ssh:
        password: flipt
        private_key_path: private-key.pem
        # private_key_bytes: <raw-key-bytes> # alternatively pass the raw bytes inline
        insecure_ignore_host_key: true

insecure_ignore_host_key is not encouraged for production use, and is false by default. Instead, you are advised to put the key fingerprint in the known hosts file where you are running Flipt. For example, for GitHub you can do ssh-keyscan github.com >> ~/.ssh/known_hosts on the Flipt host.

See our GitOps Guide for an example of how to set up a GitHub repository as a flag state backend.

Object

The object storage type supports using a hosted object storage service as the source of truth for Flipt state configuration.

Currently, Flipt supports the following object store providers:

• AWS S3
• Azure Blob Storage
• Google Cloud Storage

Contents

The contents of a target object storage bucket must contain Flipt state configuration files. As with the git and local backend types, the same rules apply with regard to how Flipt will locate feature flag state in your target bucket.

See the section below on Flag State Configuration for how Flipt decides which files in a target are considered for serving flag state.

With the object storage backend, Flipt will respect a file at the root of the target with the name .flipt.yml to serve as an index for locating flag state configuration in the bucket.

It will also use the same default strategy when the index isn’t supplied (e.g. file name features.yml or *.features.yml).

Amazon S3

The AWS S3 backend can be configured to serve state from a single bucket from a target S3-compatible API. This means that both AWS S3 and open-source alternatives such as Minio can be used.

The following is an example of how to configure Flipt to leverage this backend type:

• Environment Variable
• Configuration Yaml

FLIPT_STORAGE_TYPE="object"
FLIPT_STORAGE_OBJECT_TYPE="s3"
FLIPT_STORAGE_OBJECT_S3_REGION="us-east-1"
FLIPT_STORAGE_OBJECT_S3_BUCKET="flipt_feature_flags"
FLIPT_STORAGE_OBJECT_S3_POLL_INTERVAL="1m"
# optional: bucket prefix for locating flag state files
FLIPT_STORAGE_OBJECT_S3_PREFIX="production"
# optional: for non-AWS hosted S3
FLIPT_STORAGE_OBJECT_S3_ENDPOINT=http://localhost:9009

In addition to these Flipt configuration parameters, valid credentials will also be required for Flipt to authenticate with the target object store.

These should be provided as environment variables to the Flipt server process:

AWS_ACCESS_KEY_ID=...
AWS_SECRET_ACCESS_KEY=...

Azure Blob Storage

The Azure Blob Storage backend can be configured to serve state from a single container from a target Azure Blob Storage account.

The following is an example of how to configure Flipt to leverage this backend type:

• Environment Variable
• Configuration Yaml

FLIPT_STORAGE_TYPE="object"
FLIPT_STORAGE_OBJECT_TYPE="azblob"
FLIPT_STORAGE_OBJECT_AZBLOB_CONTAINER="flipt-feature-flags"
FLIPT_STORAGE_OBJECT_AZBLOB_POLL_INTERVAL="1m"
# optional: for customizing Azure blob storage endpoint
FLIPT_STORAGE_OBJECT_AZBLOB_ENDPOINT=http://localhost:10000

In addition to these Flipt configuration parameters, valid credentials will also be required for Flipt to authenticate with the target object store.

These should be provided as environment variables to the Flipt server process. There are 2 options supported for authentication:

• Using Azure Credentials

AZURE_CLIENT_ID=...	# application ID of an Azure service principal
AZURE_TENANT_ID=...	# ID of the application's Microsoft Entra tenant
AZURE_CLIENT_SECRET=...	# password of the Azure service principal

• Using Azure Blob Storage Account Keys

AZURE_STORAGE_ACCOUNT=...
AZURE_STORAGE_KEY=...

Google Cloud Storage

The Google Cloud Storage backend can be configured to serve state from a single bucket from a target Google Cloud Storage account.

The following is an example of how to configure Flipt to leverage this backend type:

• Environment Variable
• Configuration Yaml

FLIPT_STORAGE_TYPE="object"
FLIPT_STORAGE_OBJECT_TYPE="googlecloud"
FLIPT_STORAGE_OBJECT_GOOGLECLOUD_BUCKET="flipt-feature-flags"
FLIPT_STORAGE_OBJECT_GOOGLECLOUD_POLL_INTERVAL="1m"
# optional: bucket prefix for locating flag state files
FLIPT_STORAGE_OBJECT_GOOGLECLOUD_PREFIX="production"

In addition to these Flipt configuration parameters, valid credentials will also be required for Flipt to authenticate with the target object store.

If running in a Google Cloud environment, you can use Application Default Credentials to authenticate with Google Cloud Storage.

Alternatively, you can use a Service Account to authenticate with Google Cloud Storage and provide the service account key file to Flipt.

This should be provided as an environment variable to the Flipt server process:

GOOGLE_APPLICATION_CREDENTIALS=... # path to a service account key file

OCI

Since v1.31.0, Flipt supports using any OCI compatible registry as a declarative backend source. Flipt has its own custom OCI manifest format (we call them bundles), which can be built and managed using the Flipt CLI.

• Environment Variable
• Configuration Yaml

FLIPT_STORAGE_TYPE="oci"
FLIPT_STORAGE_OCI_REPOSITORY="some.oci.registry/repository/image:tag"
FLIPT_STORAGE_OCI_POLL_INTERVAL="30s"
# authentication credentials
FLIPT_STORAGE_OCI_AUTHENTICATION_USERNAME="username"
FLIPT_STORAGE_OCI_AUTHENTICATION_PASSWORD="password"
# location used for storing local bundles
FLIPT_STORAGE_OCI_BUNDLES_DIRECTORY="<user_config_dir>/flipt/bundles"

Flag State Configuration

Each of Flipt’s filesystem backends expects you to represent your feature flag configuration via a set of YAML files. These files declaratively define what flags, segments, variants, etc. exist and in what configuration.

Locating Flag State

Flipt’s filesystem backends allow you to define feature flags alongside other configurations in a shared directory, repository, or object storage bucket.

Flipt uses a naming scheme to index which files are flag state files. By default, Flipt will look for the following filename patterns to attempt to parse as Flipt state:

• **/features.yaml
• **/features.yml
• **/*.features.yaml
• **/*.features.yml

Any file named features.yaml, features.yml, or with either extension .features.yaml or .features.yml is considered recursively from the root of your target.

If this naming convention doesn’t work for you, it can be overridden by creating a file named .flipt.yml in the root of your target directory tree. This file will be used to instruct Flipt on how to index your directory tree and find flag state files:

version: "1.0"
include:
  - "**/features.yaml"
  - "**/features.yml"
  - "**/*.features.yaml"
  - "**/*.features.yml"
exclude: []

The index file contains two lists include and exclude. These can contain specific paths or glob-matching patterns. The indexing process first matches the include section and then filters that are set by the exclude section.

Defining Flag State

Flipt flag state file format has been taken directly from Flipt’s existing import and export flag state format.

You can run flipt export on your existing Flipt instance, and then add/commit the result to a directory, object storage, or Git repository to get started.

This can be used to migrate from a relational database-backed instance of Flipt to a filesystem-backed deployment with ease.

features.yaml

namespace: backend
flags:
  - key: awesomeNewFeature
    name: Awesome New Feature
    enabled: true
    variants:
      - key: enabled
        name: Enabled
      - key: disabled
        name: Disabled
    rules:
      - segment: internal-users
        distributions:
          - variant: enabled
            rollout: 100
      - segment: all-users
        distributions:
          - variant: enabled
            rollout: 20
          - variant: disabled
            rollout: 80

segments:
  - key: internal-users
    name: Internal Users
    constraints:
      - type: STRING_COMPARISON_TYPE
        property: organization
        operator: eq
        value: internal
    match_type: ALL_MATCH_TYPE
  - key: all-users
    name: All Users
    match_type: ALL_MATCH_TYPE

Each file identified for use by Flipt represents the contents of a single namespace. Multiple namespaces can be defined across multiple files. You can organize these files however you like in your target directory.

By defining different namespaces in different directories, you can leverage features such as GitHub’s Codeowners. This gives you authorization mechanisms for managing contributions to Flipt state.

The file format currently consists of four top-level keys:

version: "1.0" # a version for this file format
namespace: default # string identifying the resources collective namespace
flags: [] # [Flag] list of Flag definitions
segments: [] # [Segment] list of Segment definitions

Caching

Flipt supports both in-memory cache as well as Redis to enable faster reads and evaluations. Enabling caching has been shown to speed up read performance by several orders of magnitude.

Enabling in-memory caching when running more than one instance of Flipt isn’t advised as it may lead to unpredictable results. It’s recommended to use Redis instead if you are running more than one instance of Flipt.

Caching works as follows:

• All flag reads and evaluation requests go through the cache
• Flag cache entries are purged whenever a write to a flag or its variants occur or the TTL expires
• Evaluation cache entries are purged after the TTL expires only
• A cache miss will fetch the item from the database and add the item to the cache for the next read
• A cache hit will simply return the item from the cache, not interacting with the database

See the Cache section for how to configure caching.

Expiration/Eviction

You can also configure an optional duration at which items in the cache are marked as expired.

For example, if you set the cache TTL to 5m, items that have been in the cache for longer than 5 minutes will be marked as expired, meaning the next read for that item will hit the database.

Setting an eviction interval (in-memory cache only) will automatically remove expired items from your cache at a defined period.

The combination of cache expiration and eviction can help lessen the amount of memory your cache uses, as infrequently accessed items will be removed over time.

To tune the expiration and eviction interval of the cache set the following in your configuration:

cache:
  enabled: true
  backend: memory
  ttl: 5m # items older than 5 minutes will be marked as expired
  memory:
    eviction_interval: 2m # expired items will be evicted from the cache every 2 minutes