Gateway resources - Conduktor documentation

These are high-level Gateway object specifications for Conduktor CLI.

Deploy Interceptor

Deploys an Interceptor on Gateway.

Managed with: API, CLI, UI, TF

---
apiVersion: gateway/v2
kind: Interceptor
metadata:
  name: enforce-partition-limit
  # scope:
  #   vCluster: aaa
  #   group: bbb
  #   username: ccc
spec:
  pluginClass: "io.conduktor.gateway.interceptor.safeguard.CreateTopicPolicyPlugin"
  priority: 100
  config:
    topic: "myprefix-.*"
    numPartition:
      min: 5
      max: 5
      action: "INFO"

Interceptor checks:

metadata.scope is optional (default empty).
metadata.scope.[vCluster | group | username] combine with each other to define the targeting
- Check the dedicated Interceptor targeting section
spec.pluginClass is mandatory. Has to be a valid Interceptor class name.
spec.priority is mandatory
spec.config is a valid config for the pluginClass

Interceptor targeting

You can activate your Interceptor only in specific scenarios. Use the table below to configure Targeting settings.

Use case	`metadata.scope.vcluster`	`metadata.scope.group`	`metadata.scope.username`
Global Interceptor (Including Virtual Clusters)	Set to `null`	Set to `null`	Set to `null`
Global Interceptor (Excluding Virtual Clusters)	Empty	Empty	Empty
Username targeting	Empty	Empty	Set
Group targeting	Empty	Set	Empty
Virtual Cluster targeting	Set	Empty	Empty
Virtual Cluster + Username targeting	Set	Empty	Set
Virtual Cluster + Group targeting	Set	Set	Empty

You can deploy multiple Interceptors with the same name using a different targeting scope. This will effectively override the configuration for the scope.

The order of precedence from highest (overrides all others) to lowest (most easily overridden) is:

ServiceAccount
Group
VirtualCluster
Global

Examples

  ---
  # This Interceptor targets everyone (Including Virtual Clusters)
  apiVersion: gateway/v2
  kind: Interceptor
  metadata:
    name: enforce-partition-limit
    scope:
      vCluster: null
      group: null
      username: null
  spec:

  ---
  # This Interceptor targets everyone (Excluding Virtual Clusters)
  apiVersion: gateway/v2
  kind: Interceptor
  metadata:
    name: enforce-partition-limit
  spec:

  ---
  # This Interceptor targets only `admin` service account
  apiVersion: gateway/v2
  kind: Interceptor
  metadata:
    name: enforce-partition-limit
    scope:
      username: admin
  spec:

  ---
  # This interceptor targets only `read-only` virtual cluster
  apiVersion: gateway/v2
  kind: Interceptor
  metadata:
    name: enforce-partition-limit
    scope:
      vCluster: read-only
  spec:

GatewayServiceAccount

When using Oauth, mTLS or delegated backing Kafka authentication, GatewayServiceAccount is generally optional.

Managed with: API, CLI, TF

GatewayServiceAccount resource is enabled/disabled depending on your Gateway configuration. This is to prevent you from declaring a resource that’s incompatible with your current configuration:

GATEWAY_SECURITY	LOCAL GatewayServiceAccount	EXTERNAL GatewayServiceAccount
PLAINTEXT	🚫	🚫
SSL	🚫	only if mTls
SASL_PLAINTEXT	✅	only if OAuth configured
SASL_SSL	✅	only if OAuth configured
DELEGATED_SASL_PLAINTEXT	🚫	✅
DELEGATED_SASL_SSL	🚫	✅

Here are a few cases where you have to declare GatewayServiceAccount objects:

creating local service accounts
renaming service accounts for easier clarity when using Interceptors
attaching service accounts to Virtual Clusters

---
# External user renamed
apiVersion: gateway/v2
kind: GatewayServiceAccount
metadata:
  name: application1
spec:
  type: EXTERNAL
  externalNames:
  - 00u9vme99nxudvxZA0h7
---
# Local User on Virtual Cluster vc-B
apiVersion: gateway/v2
kind: GatewayServiceAccount
metadata:
  vCluster: vc-B
  name: admin
spec:
  type: LOCAL

GatewayServiceAccount checks:

When spec.type is EXTERNAL:
- spec.externalNames must be a non-empty list of external names. Each name must be unique across all declared GatewayServiceAccount.
- we currently only support a list of one element.

GatewayServiceAccount side effects:

When spec.type is EXTERNAL:
- During Client connection, the authenticated user will be checked against the list of externalNames to decide which GatewayServiceAccount it is.
When spec.type is LOCAL:
- Access to /gateway/v2/tokens endpoint to generate a password for this service account
- Switching a GatewayServiceAccount spec.type from LOCAL to EXTERNAL does not invalidate previously emitted tokens. They will keep on working for their TTL.

GatewayGroup

Gateway group lets you add multiple users in the same GatewayGroup for easier Interceptor targeting capabilities.

---
# Users added to the group manually
apiVersion: gateway/v2
kind: GatewayGroup
metadata:
  name: group-a
spec:
  members:
    - name: admin
    - vCluster: vc-B
      name: "0000-AAAA-BBBB-CCCC"

GatewayGroup checks:

spec.members[].name is mandatory.
- Currently, the username needs to refer to an existing GatewayServiceAccount otherwise it will fail. This is a known issue that we’ll address in a further release.
spec.members[].vCluster is optional. It has to refer to an existing Virtual Cluster. When not using Virtual Clusters, don’t set this attribute.

GatewayGroup side effects:

All members of the group will be affected by Interceptors deployed with this group’s scope.

ConcentrationRule

Concentration Rules allow you to define patterns where topic creation won’t generate a physical topic, but will instead use our topic concentration feature.

---
apiVersion: gateway/v2
kind: ConcentrationRule
metadata:
  # vCluster: vc-B
  name: toutdanstiti
spec:
  pattern: titi-.*
  physicalTopics:
    delete: titi-delete
    compact: titi-compact
    deleteCompact: titi-cd
  autoManaged: false
  offsetCorrectness: false

ConcentrationRule checks:

metadata.vCluster is optional. Must refer to an existing Virtual Cluster. When not using Virtual Clusters, don’t set this attribute.
spec.physicalTopics.delete is mandatory. Has to be a valid topic name with a cleanup.policy set to delete.
spec.physicalTopics.compact is optional. Has ti be a valid topic name with a cleanup.policy set to compact.
spec.physicalTopics.deleteCompact is optional. Has to be a valid topic name with a cleanup.policy set to delete,compact.
spec.autoManaged is optional, default is false.
spec.offsetCorrectness is optional, default is false.

ConcentrationRule side effects:

Once the Concentration Rule is deployed, topics created with a name matching the spec.pattern will not be created as real Kafka topics but as concentrated topics instead.
Depending on the topic’s cleanup.policy, the topic’s data will be stored in one of the configured physical topics.
If a topic creation request is made with a cleanup.policy that isn’t configured in the ConcentrationRule, topic creation will fail.
It is not possible to update cleanup.policy of a concentrated topic.
If spec.autoManaged is set to true, the underlying physical topics and configurations will be automatically created and/or extended to honour the topics configurations.
If spec.offsetCorrectness is set to true, Gateway will maintain a list of offsets for each of the concentrated topic records.
- This allows for a proper calculation of message count and consumer group lag.
- There are some limitations with offset correctness.
If spec.offsetCorrectness is set to false, Gateway will report the offsets of the backing topic records.

If a ConcentrationRule spec changes, it will not affect previously created concentrated topics, it will only affect the topics created after the change.

VirtualCluster

A Virtual Cluster allows you to isolate one or more service accounts within a logical cluster. Any topic or consumer group created within a Virtual Cluster will be accessible only to that specific Virtual Cluster.

Managed with: API, CLI

---
apiVersion: gateway/v2
kind: VirtualCluster
metadata:
 name: "mon-app-A"
spec:
 aclEnabled: false

VirtualCluster checks:

metadata.name must be a valid topic prefix as all vcluster topics and consumer groups will be created on the physical kafka cluster with this as the prefix (they will appear on the vcluster without the prefix).
spec.aclEnabled is optional (default false). When unset or false,
- there are no authorization checks for clients connecting to the vcluster. Clients are free to perform any operation on resources within the vcluster
- The fields acls and superUsers cannot be set
spec.type must be either Standard or Partner (default if not set is Standard)

The next two sections describe the behaviour of spec.aclEnabled when it is set to true

ACLs configurable via Kafka

When a vcluster is configured to have ACLs set via Kafka (aclMode: KAFKA_API) an administrator can connect to the vcluster as any service account named in the superUsers list and fully manage the ACLs of other service accounts within the vcluster using the Kafka Admin API. This is the same way you would manage a real Kafka Cluster.

---
apiVersion: gateway/v2
kind: VirtualCluster
metadata:
 name: "mon-app-A"
spec:
 aclEnabled: true
 aclMode: KAFKA_API
 superUsers:
 - username1
 - username2

when spec.aclMode is set to KAFKA_API (default if aclEnabled: true)
- it cannot be changed
- acls cannot be set
- spec.superUsers must be set
spec.superUsers is the list of usernames for which the associated service accounts in this virtual cluster can bypass ACLs.
- Specified usernames are not created automatically. You still need to create the associated gateway service accounts on the vcluster for them to work.

ACLs configurable via REST

Managing ACLs with the Kafka Admin API is extremely powerful, but can also be cumbersome. For some use-cases it is desirable to configure everything using the REST api instead.

---
apiVersion: gateway/v2
kind: VirtualCluster
metadata:
 name: "mon-app-A"
spec:
 aclEnabled: true
 aclMode: REST_API
 acls:
  - resourcePattern:
      resourceType: TOPIC
      name: customers
      patternType: LITERAL
    principal: User:username1
    host: "*"
    operation: READ
    permissionType: ALLOW
  - resourcePattern:
      resourceType: TOPIC
      name: customers
      patternType: LITERAL
    principal: User:username1
    host: "*"
    operation: WRITE
    permissionType: ALLOW

when spec.aclMode is set to REST_API
- it cannot be changed
- spec.acls must be set
- spec.superUsers cannot be set
spec.acls is the complete list of ACL bindings for the vcluster and allows (nearly) any valid Kafka API ACL binding (inc * wildcards) to be set. For a complete list of valid ACL bindings checkout the Open API schema.

We do not allow the aclMode to be changed once it is set because KAFKA_API and REST_API have incompatible mutation processes (KAFKA_API mode changes are cumulative whereas REST_API mode changes are idempotent).

AliasTopic

An Alias Topic allows a real Kafka topic to appear as a logical topic within Gateway. This is useful for aliasing topics or making a topic accessible within a Virtual Cluster.

---
apiVersion: gateway/v2
kind: AliasTopic
metadata:
  name: name1
  vCluster: vCluster1
spec:
  physicalName: physicalName1

Reference

​Deploy Interceptor

​Interceptor targeting

​Examples

​GatewayServiceAccount

​GatewayGroup

​ConcentrationRule

​VirtualCluster

​ACLs configurable via Kafka

​ACLs configurable via REST

​AliasTopic

​Related resources

Deploy Interceptor

Interceptor targeting

Examples

GatewayServiceAccount

GatewayGroup

ConcentrationRule

VirtualCluster

ACLs configurable via Kafka

ACLs configurable via REST

AliasTopic

Related resources