karapace. Your Apache Kafka® essentials in one tool.

An open-source implementation of Kafka REST and Schema Registry.

Karapace supports the storing of schemas in a central repository, which clients can access to serialize and deserialize messages. The schemas also maintain their own version histories and can be checked for compatibility between their different respective versions.

Karapace rest provides a RESTful interface to your Apache Kafka cluster, allowing you to perform tasks such as producing and consuming messages and perform administrative cluster work, all the while using the language of the WEB.

• Drop-in replacement both on pre-existing Schema Registry / Kafka Rest Proxy client and server-sides
• Moderate memory consumption
• Asynchronous architecture based on aiohttp
• Supports Avro, JSON Schema, and Protobuf
• Leader/Replica architecture for HA and load balancing

Karapace is compatible with Schema Registry 6.1.1 on API level. When a new version of SR is released, the goal is to support it in a reasonable time. Karapace supports all operations in the API.

There are some caveats regarding the schema normalization, and the error messages being the same as in Schema Registry, which cannot be always fully guaranteed.

To get you up and running with the latest build of Karapace, a docker image is available:

# Fetch the latest build from main branch
docker pull ghcr.io/aiven-open/karapace:develop

# Fetch the latest release
docker pull ghcr.io/aiven-open/karapace:latest

Versions 3.7.1 and earlier are available from the ghcr.io/aiven registry:

docker pull ghcr.io/aiven/karapace:3.7.1

An example setup including configuration and Kafka connection is available as compose example:

docker compose -f ./container/compose.yml up -d

Then you should be able to reach two sets of endpoints:

• Karapace schema registry on http://localhost:8081
• Karapace REST on http://localhost:8082

Each configuration key can be overridden with an environment variable prefixed with KARAPACE_, exception being configuration keys that actually start with the karapace string. For example, to override the bootstrap_uri config value, one would use the environment variable KARAPACE_BOOTSTRAP_URI. Here you can find an example configuration file to give you an idea what you need to change.

Alternatively you can do a source install using:

python setup.py install

To register the first version of a schema under the subject "test" using Avro schema:

$ curl -X POST -H "Content-Type: application/vnd.schemaregistry.v1+json" \
  --data '{"schema": "{\"type\": \"record\", \"name\": \"Obj\", \"fields\":[{\"name\": \"age\", \"type\": \"int\"}]}"}' \
  http://localhost:8081/subjects/test-key/versions
{"id":1}

To register a version of a schema using JSON Schema, one needs to use schemaType property:

$ curl -X POST -H "Content-Type: application/vnd.schemaregistry.v1+json" \
  --data '{"schemaType": "JSON", "schema": "{\"type\": \"object\",\"properties\":{\"age\":{\"type\": \"number\"}},\"additionalProperties\":true}"}' \
  http://localhost:8081/subjects/test-key-json-schema/versions
{"id":2}

To list all subjects (including the one created just above):

$ curl -X GET http://localhost:8081/subjects
["test-key"]

To list all the versions of a given schema (including the one just created above):

$ curl -X GET http://localhost:8081/subjects/test-key/versions
[1]

To fetch back the schema whose global id is 1 (i.e. the one registered above):

$ curl -X GET http://localhost:8081/schemas/ids/1
{"schema":"{\"fields\":[{\"name\":\"age\",\"type\":\"int\"}],\"name\":\"Obj\",\"type\":\"record\"}"}

To get the specific version 1 of the schema just registered run:

$ curl -X GET http://localhost:8081/subjects/test-key/versions/1
{"subject":"test-key","version":1,"id":1,"schema":"{\"fields\":[{\"name\":\"age\",\"type\":\"int\"}],\"name\":\"Obj\",\"type\":\"record\"}"}

To get the latest version of the schema under subject test-key run:

$ curl -X GET http://localhost:8081/subjects/test-key/versions/latest
{"subject":"test-key","version":1,"id":1,"schema":"{\"fields\":[{\"name\":\"age\",\"type\":\"int\"}],\"name\":\"Obj\",\"type\":\"record\"}"}

In order to delete version 10 of the schema registered under subject "test-key" (if it exists):

$ curl -X DELETE http://localhost:8081/subjects/test-key/versions/10
 10

To Delete all versions of the schema registered under subject "test-key":

$ curl -X DELETE http://localhost:8081/subjects/test-key
[1]

Test the compatibility of a schema with the latest schema under subject "test-key":

$ curl -X POST -H "Content-Type: application/vnd.schemaregistry.v1+json" \
  --data '{"schema": "{\"type\": \"int\"}"}' \
  http://localhost:8081/compatibility/subjects/test-key/versions/latest
{"is_compatible":true}

Get current global backwards compatibility setting value:

$ curl -X GET http://localhost:8081/config
{"compatibilityLevel":"BACKWARD"}

Change compatibility requirements for all subjects where it's not specifically defined otherwise:

$ curl -X PUT -H "Content-Type: application/vnd.schemaregistry.v1+json" \
  --data '{"compatibility": "NONE"}' http://localhost:8081/config
{"compatibility":"NONE"}

Change compatibility requirement to FULL for the test-key subject:

$ curl -X PUT -H "Content-Type: application/vnd.schemaregistry.v1+json" \
  --data '{"compatibility": "FULL"}' http://localhost:8081/config/test-key
{"compatibility":"FULL"}

List topics:

$ curl "http://localhost:8082/topics"

Get info for one particular topic:

$ curl "http://localhost:8082/topics/my_topic"

Produce a message backed up by schema registry:

$ curl -H "Content-Type: application/vnd.kafka.avro.v2+json" -X POST -d \
  '{"value_schema": "{\"namespace\": \"example.avro\", \"type\": \"record\", \"name\": \"simple\", \"fields\": \
  [{\"name\": \"name\", \"type\": \"string\"}]}", "records": [{"value": {"name": "name0"}}]}' http://localhost:8082/topics/my_topic

Create a consumer:

$ curl -X POST -H "Content-Type: application/vnd.kafka.v2+json" -H "Accept: application/vnd.kafka.v2+json" \
  --data '{"name": "my_consumer", "format": "avro", "auto.offset.reset": "earliest"}' \
  http://localhost:8082/consumers/avro_consumers

Subscribe to the topic we previously published to:

$ curl -X POST -H "Content-Type: application/vnd.kafka.v2+json" --data '{"topics":["my_topic"]}' \
  http://localhost:8082/consumers/avro_consumers/instances/my_consumer/subscription

Consume previously published message:

$ curl -X GET -H "Accept: application/vnd.kafka.avro.v2+json" \
  http://localhost:8082/consumers/avro_consumers/instances/my_consumer/records?timeout=1000

Commit offsets for a particular topic partition:

$ curl -X POST -H "Content-Type: application/vnd.kafka.v2+json" --data '{}' \
  http://localhost:8082/consumers/avro_consumers/instances/my_consumer/offsets

Delete consumer:

$ curl -X DELETE -H "Accept: application/vnd.kafka.v2+json" \
  http://localhost:8082/consumers/avro_consumers/instances/my_consumer

Karapace natively stores its data in a Kafka topic the name of which you can configure freely but which by default is called _schemas.

Karapace includes a tool to backing up and restoring data. To back up, run:

karapace_schema_backup get --config karapace.config.json --location schemas.log

You can also back up the data by using Kafka's Java console consumer:

./kafka-console-consumer.sh --bootstrap-server brokerhostname:9092 --topic _schemas --from-beginning --property print.key=true --timeout-ms 1000 1> schemas.log

Your backup can be restored with Karapace by running:

karapace_schema_backup restore --config karapace.config.json --location schemas.log

Or Kafka's Java console producer can be used to restore the data to a new Kafka cluster.

You can restore the data from the previous step by running:

./kafka-console-producer.sh --broker-list brokerhostname:9092 --topic _schemas --property parse.key=true < schemas.log

• 50 concurrent connections, 50.000 requests

• 15 concurrent connections, 50.000 requests

• 4 concurrent connections, 50.000 requests

Also, it appears there is quite a bit of variation on subsequent runs, especially for the lower numbers, so once more exact measurements are required, it's advised we increase the total req count to something like 500K

We'll focus on Avro serialization only after this round, as it's the more expensive one, plus it tests the entire stack

A basic push pull test , with 12 connections on the publisher process and 3 connections on the subscriber process, with a 10 minute duration. The publisher has the 100 ms timeout and 100 max_bytes parameters set on each request so both processes have work to do Heap size limit is set to 256M on Rest proxy

Ram consumption, different consumer count, over 300s

Once installed, the karapace program should be in your path. It is the main daemon process that should be run under a service manager such as systemd to serve clients.

Keys to take special care are the ones needed to configure Kafka and advertised_hostname.

To enable HTTP Basic Authentication and user authorization the authorization configuration file is set in the main configuration key registry_authfile of the Karapace.

Karapace Schema Registry authorization file is an optional JSON configuration, which contains a list of authorized users in users and a list of access control rules in permissions.

Each user entry contains following attributes:

Password hashing can be done using karapace_mkpasswd tool, if installed, or by invoking directly with python -m karapace.auth. The tool generates JSON entry with these fields.

$ karapace_mkpasswd -u user -a sha512 secret
{
    "username": "user",
    "algorithm": "sha512",
    "salt": "iuLouaExTeg9ypqTxqP-dw",
    "password_hash": "R6ghYSXdLGsq6hkQcg8wT4xkD4QToxBhlp7NerTnyB077M+mD2qiN7ZxXCDb4aE+5lExu5P11UpMPYAcVYxSQA=="
}

Each access control rule contains following attributes:

Supported resource authorization:

{
    "users": [
        {
            "username": "admin",
            "algorithm": "scrypt",
            "salt": "<put salt for randomized hashing here>",
            "password_hash": "<put hashed password here>"
        },
        {
            "username": "plainuser",
            "algorithm": "sha256",
            "salt": "<put salt for randomized hashing here>",
            "password_hash": "<put hashed password here>"
        }
    ],
    "permissions": [
        {
            "username": "admin",
            "operation": "Write",
            "resource": ".*"
        },
        {
            "username": "plainuser",
            "operation": "Read",
            "resource": "Subject:general.*"
        },
        {
            "username": "plainuser",
            "operation": "Read",
            "resource": "Config:"
        }
    ]
}

The principal used by the Karapace Schema Registry has to have adequate access to the schemas topic (see the topic_name configuration option above). In addition to what is required to access the topic, as described in the Confluent Schema Registry documentation, the unique, single-member consumer group used by consumers in the schema registry needs Describe and Read permissions on the group. These unique (per instance of the schema registry) consumer group names are prefixed by karapace-autogenerated-, followed by a random string.

The Karapace REST proxy supports passing OAuth2 credentials to the underlying Kafka service (defined in the sasl_bootstrap_uri configuration parameter). The JSON Web Token (JWT) is extracted from the Authorization HTTP header if the authorization scheme is Bearer, eg. Authorization: Bearer $JWT. If a Bearer token is present, the Kafka clients managed by Karapace will be created to use the SASL OAUTHBEARER mechanism and the JWT will be passed along. The Karapace REST proxy does not verify the token, that is done by the underlying Kafka service itself, if it's configured accordingly.

Authorization is also done by Kafka itself, typically using the sub claim (although it's configurable) from the JWT as the username, checked against the configured ACLs.

OAuth2 and Bearer token usage is dependent on the rest_authorization configuration parameter being true.

The REST proxy process manages a set of producer and consumer clients, which are identified by the OAuth2 JWT token. These are periodically cleaned up if they are idle, as well as before the JWT token expires (the clean up currently runs every 5 minutes).

Before a client refreshes its OAuth2 JWT token, it is expected to remove currently running consumers (eg. after committing their offsets) and producers using the current token.

If specified as a rest parameter for the POST /subjects/{subject}/versions?normalize=true endpoint and the POST subjects/{subject}?normalize=true endpoint, Karapace uses a schema normalization algorithm to ensure that the schema is stored in a canonical form.

This normalization process is done so that schemas semantically equivalent are stored in the same way and should be considered equal.

Normalization is currently only supported for Protobuf schemas. Karapace does not support all normalization features implemented by Confluent Schema Registry. Currently the normalization process is done only for the ordering of the optional fields in the schema. Use the feature with the assumption that it will be extended in the future and so two schemas that are semantically equivalent could be considered different by the normalization process in different future versions of Karapace. The safe choice, when using a normalization process, is always to consider as different two schemas that are semantically equivalent while the problem is when two semantically different schemas are considered equivalent. In that view the future extension of the normalization process isn't considered a breaking change but rather an extension of the normalization process.

To unistall Karapace from the system you can follow the instructions described below. We would love to hear your reasons for uninstalling though. Please file an issue if you experience any problems or email us with feedback

If you installed Karapace via Docker, you would need to first stop and remove the images like described:

First obtain the container IDs related to Karapace, you should have one for the registry itself and another one for the rest interface:

docker ps | grep karapace

After this, you can stop each of the containers with:

docker stop <CONTAINER_ID>

If you don't need or want to have the Karapace images around you can now proceed to delete them using:

docker rm <CONTAINER_ID>

If you installed Karapace from the sources via python setup.py install, it can be uninstalled with the following pip command:

pip uninstall karapace

Execute make (GNU, usually gmake on BSD and Mac) to set up a venv and install the required software for development. Use make unit-tests and make integration-tests to execute the respective test suite, or simply make test to execute both. You can set PYTEST_ARGS to customize the execution (e.g. PYTEST_ARGS=--maxfail=1 make test).

By default pyenv is expected to be installed and in PATH. This ensures on all platforms that arbitrary Python versions can be used for development. It is possible to overwrite this by setting PYENV to something else (e.g. PYENV=python3 make venv to simply use the global Python executable). The default Python version is defined in .python-version.

Karapace currently depends on various system software to be installed. The installation of these is automated for some operation systems, but not all. At the time of writing Java, the Protobuf Compiler, and the Snappy shared library are required to work with Karapace. You need to install them manually if your operating system is not supported by the automatic installation scripts. Note that the scripts are going to ask before installing any of these on your system.

Note that Karapace requires a Protobuf Compiler older than 3.20.0, because 3.20.0 introduces various breaking changes. The tests are going to fail if the Protobuf Compiler is newer than that. However, you can work around this locally by running pip install --upgrade protobuf in your venv. We are going to fix this soon.

Note that the integration tests are currently not working on Mac. You can use Docker, just be sure to set VENV_DIR to a directory outside the working directory so that the container is not overwriting files from the host (e.g. docker run --env VENV_DIR=/tmp/venv ...).

Note that the runtime directory MUST exist and that Karapace is going to fail if it does not. The runtime directory is also not cleaned between test runs, and left over data might result in failing tests. Use the make test targets that correctly clean the runtime directory without deleting it, but keep this in mind whenever you are not using make (e.g. running tests from your IDE).

Note that the pre-commit checks are currently not working with the default Python version. This is because isort dropped Python 3.7 support. You have to use at least Python 3.8 for the pre-commit checks. Use pipx or brew or … to install pre-commit and use the global installation, there is also no dependency on it.

Karapace is licensed under the Apache license, version 2.0. Full license text is available in the LICENSE file.

Please note that the project explicitly does not require a CLA (Contributor License Agreement) from its contributors.

Bug reports and patches are very welcome, please post them as GitHub issues and pull requests at https://github.com/aiven/karapace . Any possible vulnerabilities or other serious issues should be reported directly to the maintainers <[email protected]>.

Apache Kafka is either a registered trademark or trademark of the Apache Software Foundation in the United States and/or other countries. Kafka Rest and Schema Registry are trademarks and property of their respective owners. All product and service names used in this page are for identification purposes only and do not imply endorsement.

Karapace was created by, and is maintained by, Aiven cloud data hub developers.

The schema storing part of Karapace loans heavily from the ideas of the earlier Schema Registry implementation by Confluent and thanks are in order to them for pioneering the concept.

Recent contributors are listed on the GitHub project page, https://github.com/aiven/karapace/graphs/contributors

Copyright ⓒ 2021 Aiven Ltd.