Skip to content

conduktor/kafka-security-manager

Repository files navigation

Actions Status

An open-source project by Conduktor.io

This project is sponsored by conduktor.io. Conduktor provides a platform to help central teams defining global security and governance controls and developers have access to a central Kafka console and get a self-serve experience. It also helps visualize your ACLs (attached to Service Accounts) in your Apache Kafka cluster!

Kafka Security Manager Diagram

Kafka Security Manager

Kafka Security Manager (KSM) allows you to manage your Kafka ACLs at scale by leveraging an external source as the source of truth. Zookeeper just contains a copy of the ACLs instead of being the source.

Kafka Security Manager Diagram

There are several advantages to this:

  • Kafka administration is done outside of Kafka: anyone with access to the external ACL source can manage Kafka Security
  • Prevents intruders: if someone were to add ACLs to Kafka using the CLI, they would be reverted by KSM within 10 seconds.
  • Full auditability: KSM provides the guarantee that ACLs in Kafka are those in the external source. Additionally, if for example your external source is GitHub, then PRs, PR approvals and commit history will provide Audit the full log of who did what to the ACLs and when
  • Notifications: KSM can notify external channels (such as Slack) in order to give feedback to admins when ACLs are changed. This is particularly useful to ensure that 1) ACL changes are correctly applied 2) ACL are not changed in Kafka directly.

Your role is to ensure that Kafka Security Manager is never down, as it is now a custodian of your ACL.

Parsers

CSV

The csv parser is the default parser and also the fallback one in case no other parser is matched.

This is a sample CSV acl file:

KafkaPrincipal,ResourceType,PatternType,ResourceName,Operation,PermissionType,Host
User:alice,Topic,LITERAL,foo,Read,Allow,*
User:bob,Group,PREFIXED,bar,Write,Deny,12.34.56.78
User:peter,Cluster,LITERAL,kafka-cluster,Create,Allow,*

Important Note: As of KSM 0.4, a new column PatternType has been added to match the changes that happened in Kafka 2.0. This enables KSM to manage LITERAL and PREFIXED ACLs. See #28

YAML

The yaml parser will load ACLs from yaml instead, to activate the parser just provide files with yml or yaml extension.

An example YAML permission file might be:

users:
  alice:
    topics:
      foo:
        - Read
      bar*:
        - Produce
  bob:
    groups:
      bar:
        - Write,Deny,12.34.56.78
      bob*:
        - All
    transactional_ids:
      bar-*:
        - All
  peter:
    clusters:
      kafka-cluster:
        - Create

The YAML parser will handle automatically prefix patterns by simply appending a star to your resource name.

It also supports some helpers to simplify setup:

  • Consume (Read, Describe)
  • Produce (Write, Describe, Create, Cluster Create)

Sources

Current sources shipping with KSM include:

  • File
  • GitHub
  • GitLab (using Personal Auth Tokens)
  • BitBucket
  • Amazon S3
  • Build your own (and contribute back!)

Building

sbt clean test
sbt universal:stage

Fat JAR:

sbt clean assembly

This is a Scala app and therefore should run on the JVM like any other application

Artifacts

By using the JAR dependency, you can create your own SourceAcl.

RELEASES artifacts are deployed to Maven Central:

build.sbt (see Maven Central for the latest version)

libraryDependencies += "io.conduktor" %% "kafka-security-manager" % "version"

Configuration

Security configuration - Zookeeper client

Make sure the app is using a property file and launch options similar to your broker so that it can

  1. Authenticate to Zookeeper using secure credentials (usually done with JAAS)
  2. Apply Zookeeper ACL if enabled

Kafka Security Manager does not connect to Kafka.

Sample run for a typical SASL Setup:

target/universal/stage/bin/kafka-security-manager -Djava.security.auth.login.config=conf/jaas.conf

Where conf/jaas.conf contains something like:

Client {
    com.sun.security.auth.module.Krb5LoginModule required
    useKeyTab=true
    storeKey=true
    keyTab="/etc/kafka/secrets/zkclient1.keytab"
    principal="zkclient/[email protected]";
};

Security configuration - Admin client

When configured authorizer class is io.conduktor.ksm.compat.AdminClientAuthorizer, kafka-security-manager will use kafka admin client instead of direct zookeeper connection. Configuration example would be

KafkaClient {
  org.apache.kafka.common.security.plain.PlainLoginModule required
  username="admin"
  password="admin-secret";
};

Configuration file

For a list of configuration see application.conf. You can customise them using environment variables or create your own application.conf file and pass it at runtime doing:

target/universal/stage/bin/kafka-security-manager -Dconfig.file=path/to/config-file.conf

Overall we use the lightbend config library to configure this project.

Environment variables

The default configurations can be overwritten using the following environment variables:

  • KSM_READONLY=false: enables KSM to synchronize from an External ACL source. The default value is true, which prevents KSM from altering ACLs in Zookeeper

  • KSM_EXTRACT_ENABLE=true: enable extract mode (get all the ACLs from Kafka formatted as a CSV or YAML)

  • KSM_EXTRACT_FORMAT=csv: selects which format to extract the ACLs with (defaults to csv, supports also yaml)

  • KSM_REFRESH_FREQUENCY_MS=10000: how often to check for changes in ACLs in Kafka and in the Source. 10000 ms by default. If it's set to 0 or negative value, for example -1, then KMS executes ACL synchronization just once and exits

  • KSM_NUM_FAILED_REFRESHES_BEFORE_NOTIFICATION=1: how many times that the refresh of a Source needs to fail (e.g. HTTP timeouts) before a notification is sent. Any value less than or equal to 1 here will notify on every failure to refresh.

  • AUTHORIZER_CLASS: authorizer class for ACL operations. Default is SimpleAclAuthorizer, configured with

    • AUTHORIZER_ZOOKEEPER_CONNECT: zookeeper connection string
    • AUTHORIZER_ZOOKEEPER_SET_ACL=true (default false): set to true if you want your ACLs in Zookeeper to be secure (you probably do want them to be secure) - when in doubt set as the same as your Kafka brokers.

    No-zookeeper authorizer class on top of Kafka Admin Client is bundled with KSM as io.conduktor.ksm.compat.AdminClientAuthorizer, configured with options for org.apache.kafka.clients.admin.AdminClientConfig:

    • ADMIN_CLIENT_ID - client.id, an id to pass to the server when making requests, for tracing/audit purposes, default kafka-security-manager Properties below are not provided to client unless environment variable is set:
    • ADMIN_CLIENT_BOOTSTRAP_SERVERS - bootstrap.servers
    • ADMIN_CLIENT_SECURITY_PROTOCOL - security.protocol
    • ADMIN_CLIENT_SASL_JAAS_CONFIG -sasl.jaas.config - alternative to system jaas configuration
    • ADMIN_CLIENT_SASL_MECHANISM - sasl.mechanism
    • ADMIN_CLIENT_SSL_KEY_PASSWORD - ssl.key.password
    • ADMIN_CLIENT_SSL_KEYSTORE_LOCATION - ssl.keystore.location
    • ADMIN_CLIENT_SSL_KEYSTORE_PASSWORD - ssl.keystore.password
    • ADMIN_CLIENT_SSL_TRUSTSTORE_LOCATION - ssl.truststore.location
    • ADMIN_CLIENT_SSL_TRUSTSTORE_PASSWORD - ssl.truststore.password
  • SOURCE_CLASS: Source class. Valid values include

    • io.conduktor.ksm.source.NoSourceAcl (default): No source for the ACLs. Only use with KSM_READONLY=true
    • io.conduktor.ksm.source.FileSourceAcl: get the ACL source from a file on disk. Good for POC
    • io.conduktor.ksm.source.GitHubSourceAcl: get the ACL from GitHub. Great to get started quickly and store the ACL securely under version control.
    • io.conduktor.ksm.source.GitLabSourceAcl: get the ACL from GitLab using personal access tokens. Great to get started quickly and store the ACL securely under version control.
      • SOURCE_GITLAB_REPOID GitLab project id
      • SOURCE_GITLAB_FILEPATH Path to the ACL file in GitLab project
      • SOURCE_GITLAB_BRANCH Git Branch name
      • SOURCE_GITLAB_HOSTNAME GitLab Hostname
      • SOURCE_GITLAB_ACCESSTOKEN GitLab Personal Access Token. See Personal access tokens to authenticate with the GitLab API.
    • io.conduktor.ksm.source.S3SourceAcl: get the ACL from S3. Good for when you have a S3 bucket managed by Terraform or Cloudformation. This requires region, bucketname and objectkey. See Access credentials for credentials management.
      • SOURCE_S3_REGION AWS S3 Region
      • SOURCE_S3_BUCKETNAME AWS S3 Bucket name
      • SOURCE_S3_OBJECTKEY The Object containing the ACL CSV in S3
    • io.conduktor.ksm.source.BitbucketServerSourceAcl: get the ACL from Bitbucket Server using the v1 REST API. Great if you have private repos in Bitbucket.
    • io.conduktor.ksm.source.BitbucketCloudSourceAcl: get the ACL from Bitbucket Cloud using the Bitbucket Cloud REST API v2.
    • io.conduktor.ksm.source.HttpSourceAcl: get the ACL from an HTTP endpoint. You can enable Google OAuth OIDC Token Authentication.
      • SOURCE_HTTP_URL HTTP endpoint to retrieve ACL data.
      • SOURCE_HTTP_METHOD HTTP Method. Default is GET.
      • SOURCE_HTTP_AUTH_TYPE To enable Http Authentication. googleiam for Google IAM. Default is NONE.
      • SOURCE_HTTP_AUTH_GOOGLEIAM_SERVICE_ACCOUNT Google Service Account name.
      • SOURCE_HTTP_AUTH_GOOGLEIAM_SERVICE_ACCOUNT_KEY Google Service Account Key in JSON string encoded. If not the key isn't configured, it'll try to get the token from environment.
      • SOURCE_HTTP_AUTH_GOOGLEIAM_TARGET_AUDIENCE Google Target Audience for token authentication.
  • NOTIFICATION_CLASS: Class for notification in case of ACL changes in Kafka.

    • io.conduktor.ksm.notification.ConsoleNotification (default): Print changes to the console. Useful for logging
    • io.conduktor.ksm.notification.SlackNotification: Send notifications to a Slack channel (useful for devops / admin team)
  • ACL_PARSER_CSV_DELIMITER: Change the delimiter character for the CSV Parser (useful when you have SSL)

Running on Docker

Building the image

./build-docker.sh

Docker Hub

Alternatively, you can get the automatically built Docker images on Docker Hub

Running

(read above for configuration details)

Then apply to the docker run using for example (in EXTRACT mode):

docker run -it -e AUTHORIZER_ZOOKEEPER_CONNECT="zookeeper-url:2181" -e KSM_EXTRACT_ENABLE=true \
            conduktor/kafka-security-manager:latest

Any of the environment variables described above can be used by the docker run command with the -e options.

Example

docker-compose up -d
docker-compose logs kafka-security-manager
# view the logs, have fun changing example/acls.csv
docker-compose down

For full usage of the docker-compose file see kafka-security-manager

Extracting ACLs

You can initially extract all your existing ACL in Kafka by running the program with the config ksm.extract.enable=true or export KSM_EXTRACT_ENABLE=true

Output should look like:

[2018-03-06 21:49:44,704] INFO Running ACL Extraction mode (ExtractAcl)
[2018-03-06 21:49:44,704] INFO Getting ACLs from Kafka (ExtractAcl)
[2018-03-06 21:49:44,704] INFO Closing Authorizer (ExtractAcl)

KafkaPrincipal,ResourceType,PatternType,ResourceName,Operation,PermissionType,Host
User:bob,Group,PREFIXED,bar,Write,Deny,12.34.56.78
User:alice,Topic,LITERAL,foo,Read,Allow,*
User:peter,Cluster,LITERAL,kafka-cluster,Create,Allow,*

You can then use place this CSV anywhere and use it as your source of truth.

Compatibility

KSM Version Kafka Version Notes
1.1.0-SNAPSHOT 2.8.x updated log4j dependency
1.0.1 2.8.x updated log4j dependency
0.11.0 2.5.x renamed packages to io.conduktor. Breaking change on extract config name
0.10.0 2.5.x YAML support
Add configurable num failed refreshes before notification
0.9 2.5.x Upgrade to Kafka 2.5.x
0.8 2.3.1 Add a "run once" mode
0.7 2.1.1 Kafka Based ACL refresher available (no zookeeper dependency)
0.6 2.0.0 important stability fixes - please update
0.5 2.0.0
0.4 2.0.0 important change: added column 'PatternType' in CSV
0.3 1.1.x
0.2 1.1.x upgrade to 0.3 recommended
0.1 1.0.x might work for earlier versions

Contributing

You can break the API / configs as long as we haven't reached 1.0. Each API break would introduce a new version number.

PRs are welcome, especially with the following:

  • Code refactoring / cleanup / renaming
  • External Sources for ACLs (JDBC, Microsoft AD, etc...)
  • Notification Channels (Email, etc...)

Please open an issue before opening a PR.

Release process

  • update version in [build.sbt] (make sure to use format vX.Y.Z)
  • update [README.md] and [CHANGELOG.md]
  • push the tag (eg: v0.10.0)
  • update version in [build.sbt] to the next snapshot version

That's it !