This is Lakekeeper: A Rust-native implementation of the Apache Iceberg REST Catalog specification based on apache/iceberg-rust.
If you have questions, feature requests or just want a chat, we are hanging around in Discord!
The catalog is evolving quickly. Especially internal rust APIs are not stable and subject to change. External REST APIs are kept as stable as possible, especially the /catalog API is stable as of today.
- Release 0.5.0: Our biggest release yet is coming early November. It includes Fine Grained Access Control, a first simple UI and much more.
- Release 0.6.0: With Release 0.6.0 we focus on stabilizing some of our most recent features and significantly improve our Docs. Docusaurus is coming!
A Docker Container is available on quay.io.
We have prepared a self-contained docker-compose file to demonstrate the usage of spark with our catalog:
git clone https://github.com/lakekeeper/lakekeeper.git
cd lakekeeper/examples/self-contained
docker compose upThen open your browser and head to localhost:8888.
For more information on deployment, please check the User Guide.
The Iceberg Catalog REST interface has become the standard for catalogs in open Lakehouses. It natively enables multi-table commits, server-side deconflicting and much more. It is figuratively the (TIP) of the Iceberg.
We have started this implementation because we were missing customizability, support for on-premise deployments and other features that are important for us in existing Iceberg Catalogs. Please find following some of our focuses with this implementation:
- Customizable: Our implementation is meant to be extended. We expose the Database implementation, Secrets, Authorization, EventPublishing and ContractValidation as interfaces (Traits). This allows you to tap into any Access management system of your company or stream change events to any system you like - simply by implementing a handful methods. Please find more details in the Customization Guide.
- Change Events: Built-in support to emit change events (CloudEvents), which enables you to react to any change that happen to your tables.
- Change Approval: Changes can also be prohibited by external systems. This can be used to prohibit changes to tables that would invalidate Data Contracts, Quality SLOs etc. Simply integrate with your own change approval via our
ContractVerificationtrait. - Multi-Tenant capable: A single deployment of our catalog can serve multiple projects - all with a single entrypoint. All Iceberg and Warehouse configurations are completely separated between Warehouses.
- Written in Rust: Single 30Mb all-in-one binary - no JVM or Python env required.
- Storage Access Management: Built-in S3-Signing that enables support for self-hosted as well as AWS S3 WITHOUT sharing S3 credentials with clients. We are also working on
vended-credentials! - Well-Tested: Integration-tested with
spark,pyiceberg,trinoandstarrocks. - High Available & Horizontally Scalable: There is no local state - the catalog can be scaled horizontally and updated without downtimes.
- Openid provider integration: Use your own identity provider to secure access to the APIs, just set
LAKEKEEPER__OPENID_PROVIDER_URIand you are good to go. - Fine Grained Access (FGA) (Coming soon): Simple Role-Based access control is not enough for many rapidly evolving Data & Analytics initiatives. We are leveraging OpenFGA based on googles Zanzibar-Paper to implement authorization. If your company already has a different system in place, you can integrate with it by implementing a handful of methods in the
AuthZHandlertrait.
Please find following an overview of currently supported features. Please also check the Issues if you are missing something.
| Operation | Status | Description |
|---|---|---|
| Namespace |  |
All operations implemented |
| Table | |
All operations implemented - additional integration tests in development |
| Views | |
Remove unused files and log entries |
| Metrics |  |
Endpoint is available but doesn't store the metrics |
| Storage | Status | Comment |
|---|---|---|
| S3 - AWS |  |
vended-credentials & remote-signing, assume role missing |
| S3 - Custom | |
vended-credentials & remote-signing, tested against minio |
| Azure ADLS Gen2 | |
|
| Azure Blob | |
|
| Microsoft OneLake | |
|
| Google Cloud Storage | |
Details on how to configure the storage profiles can be found in the Storage Guide.
| Backend | Status | Comment |
|---|---|---|
| Postgres | |
|
| MongoDB | |
| Backend | Status | Comment |
|---|---|---|
| Postgres | |
|
| kv2 (hcp-vault) | |
userpass auth |
| Backend | Status | Comment |
|---|---|---|
| Nats | |
|
| Kafka | |
| Operation | Status | Description |
|---|---|---|
| Warehouse Management | |
Create / Update / Delete a Warehouse |
| AuthZ | |
Manage access to warehouses, namespaces and tables |
| More to come! | |
| Operation | Status | Description |
|---|---|---|
| OIDC (AuthN) | |
Secure access to the catalog via OIDC |
| Custom (AuthZ) | |
If you are willing to implement a single rust Trait, the AuthZHandler can be implement to connect to your system |
| OpenFGA (AuthZ) | |
Internal Authorization management |
Lakekeeper can host multiple independent projects that each host multiple warehouses. The overall structure looks like this:
<project-1-uuid>/
├─ foo-warehouse
├─ bar-warehouse
<project-2-uuid>/
├─ foo-warehouse
├─ bas-warehouse
Each Project in Lakekeeper corresponds to a distinct Application Object within your IdP, enabling genuine Multi-Tenancy. A single instance of Lakekeeper can host multiple Projects, ensuring separation at all levels, including Authentication. Our Multi-Tenancy definition is broader than most catalogs. While other catalogs may consider hosting multiple warehouses under a single IdP Application as Multi-Tenancy, a single Lakekeeper Project can achieve that and more.
As this additional layer isn't necessary for most deployments, the LAKEKEEPER__ENABLE_DEFAULT_PROJECT configuration option is set to true by default. This setting effectively disables multi-tenancy by defaulting to a project with the NIL UUID ("00000000-0000-0000-0000-000000000000") as the project-id.
To enable multi-project setups, simply set LAKEKEEPER__ENABLE_DEFAULT_PROJECT to false. Multi-project setups require additional configuration to include project-ids in HTTP headers and separate authentication audiences. This documentation does not cover these steps yet. Of course, hosting multiple warehouses within a single Lakekeeper Project is also possible.
When a table or view is dropped, it is not immediately deleted from the catalog. Instead, it is marked as dropped and a job for its cleanup is scheduled. The table, including its data if purgeRequested=True, is then deleted after the configured LAKEKEEPER__DEFAULT_TABULAR_EXPIRATION_DELAY_SECONDS (default: 7 days) have passed. This will allow for a recovery of tables that have been dropped by accident.
Undrop of a table is only possible if soft-deletes are enabled for a Warehouse.
The basic setup of the Catalog is configured via environment variables. As this catalog supports a multi-tenant setup, each catalog ("warehouse") also comes with its own configuration options including its Storage Configuration. The documentation of the Management-API for warehouses is hosted at the unprotected /swagger-ui endpoint.
Following options are global and apply to all warehouses:
Previous to Lakekeeper Version 0.5.0 please prefix all environment variables with ICEBERG_REST__ instead of LAKEKEEPER__.
You may be running Lakekeeper in your own environment which uses self-signed certificates for e.g. minio. Lakekeeper is built with reqwest's rustls-tls-native-roots feature activated, this means SSL_CERT_FILE and SSL_CERT_DIR are respected. If both are not set, the system's default CA store is used. If you want to use a custom CA store, set SSL_CERT_FILE to the path of the CA file or SSL_CERT_DIR to the path of the CA directory. The certificate used by the server cannot be a CA. It needs to be an end entity certificate, else you may run into CaUsedAsEndEntity errors.
Currently, the catalog uses two task queues, one to ultimately delete soft-deleted tabulars and another to purge tabulars which have been deleted with the purgeRequested=True query parameter. The task queues are configured as follows:
| Variable | Example | Description |
|---|---|---|
LAKEKEEPER__QUEUE_CONFIG__MAX_RETRIES |
5 | Number of retries before a task is considered failed Default: 5 |
LAKEKEEPER__QUEUE_CONFIG__MAX_AGE |
3600 | Amount of seconds before a task is considered stale and could be picked up by another worker. Default: 3600 |
LAKEKEEPER__QUEUE_CONFIG__POLL_INTERVAL |
10 | Amount of seconds between polling for new tasks. Default: 10 |
The queues are currently implemented using the sqlx Postgres backend. If you want to use a different backend, you need to implement the TaskQueue trait.
Configuration parameters if Postgres is used as a backend, you may either provide connection strings or use the PG_* environment variables, connection strings take precedence:
| Variable | Example | Description |
|---|---|---|
LAKEKEEPER__PG_DATABASE_URL_READ |
postgres://postgres:password@localhost:5432/iceberg |
Postgres Database connection string used for reading |
LAKEKEEPER__PG_DATABASE_URL_WRITE |
postgres://postgres:password@localhost:5432/iceberg |
Postgres Database connection string used for writing. |
LAKEKEEPER__PG_ENCRYPTION_KEY |
<This is unsafe, please set a proper key> |
If LAKEKEEPER__SECRET_BACKEND=postgres, this key is used to encrypt secrets. It is required to change this for production deployments. |
LAKEKEEPER__PG_READ_POOL_CONNECTIONS |
10 |
Number of connections in the read pool |
LAKEKEEPER__PG_WRITE_POOL_CONNECTIONS |
5 |
Number of connections in the write pool |
LAKEKEEPER__PG_HOST_R |
localhost |
Hostname for read operations |
LAKEKEEPER__PG_HOST_W |
localhost |
Hostname for write operations |
LAKEKEEPER__PG_PORT |
5432 |
Port number |
LAKEKEEPER__PG_USER |
postgres |
Username for authentication |
LAKEKEEPER__PG_PASSWORD |
password |
Password for authentication |
LAKEKEEPER__PG_DATABASE |
iceberg |
Database name |
LAKEKEEPER__PG_SSL_MODE |
require |
SSL mode (disable, allow, prefer, require) |
LAKEKEEPER__PG_SSL_ROOT_CERT |
/path/to/root/cert |
Path to SSL root certificate |
LAKEKEEPER__PG_ENABLE_STATEMENT_LOGGING |
true |
Enable SQL statement logging |
LAKEKEEPER__PG_TEST_BEFORE_ACQUIRE |
true |
Test connections before acquiring from the pool |
LAKEKEEPER__PG_CONNECTION_MAX_LIFETIME |
1800 |
Maximum lifetime of connections in seconds |
Configuration parameters if a KV2 compatible storage is used as a backend. Currently, we only support the userpass authentication method. You may provide the envs as single values like LAKEKEEPER__KV2__URL=http://vault.local etc. or as a compound value like:
LAKEKEEPER__KV2='{url="http://localhost:1234", user="test", password="test", secret_mount="secret"}'
| Variable | Example | Description |
|---|---|---|
LAKEKEEPER__KV2__URL |
https://vault.local |
URL of the KV2 backend |
LAKEKEEPER__KV2__USER |
admin |
Username to authenticate against the KV2 backend |
LAKEKEEPER__KV2__PASSWORD |
password |
Password to authenticate against the KV2 backend |
LAKEKEEPER__KV2__SECRET_MOUNT |
kv/data/iceberg |
Path to the secret mount in the KV2 backend |
If you want the server to publish events to a NATS server, set the following environment variables:
| Variable | Example | Description |
|---|---|---|
LAKEKEEPER__NATS_ADDRESS |
nats://localhost:4222 |
The URL of the NATS server to connect to |
LAKEKEEPER__NATS_TOPIC |
iceberg |
The subject to publish events to |
LAKEKEEPER__NATS_USER |
test-user |
User to authenticate against nats, needs LAKEKEEPER__NATS_PASSWORD |
LAKEKEEPER__NATS_PASSWORD |
test-password |
Password to authenticate against nats, needs LAKEKEEPER__NATS_USER |
LAKEKEEPER__NATS_CREDS_FILE |
/path/to/file.creds |
Path to a file containing nats credentials |
LAKEKEEPER__NATS_TOKEN |
xyz |
Nats token to authenticate against server |
If you want to limit ac
cess to the API, set LAKEKEEPER__OPENID_PROVIDER_URI to the URI of your OpenID Connect Provider. The catalog will then verify access tokens against this provider. The provider must have the .well-known/openid-configuration endpoint under ${LAKEKEEPER__OPENID_PROVIDER_URI}/.well-known/openid-configuration and the openid-configuration needs to have the jwks_uri and issuer defined.
If LAKEKEEPER__OPENID_PROVIDER_URI is set, every request needs have an authorization header, e.g.
curl {your-catalog-url}/catalog/v1/transactions/commit -X POST -H "authorization: Bearer {your-token-here}" -H "content-type: application/json" -d ...| Variable | Example | Description |
|---|---|---|
LAKEKEEPER__OPENID_PROVIDER_URI |
https://keycloak.local/realms/{your-realm} |
OpenID Provider URL, with keycloak this is the url pointing to your realm, for Azure App Registration it would be something like https://login.microsoftonline.com/{your-tenant-id-here}/v2.0/. If this variable is not set, endpoints are not secured |
Licensed under the Apache License, Version 2.0