Skip to content

Latest commit

 

History

History
502 lines (392 loc) · 27.9 KB

README.md

File metadata and controls

502 lines (392 loc) · 27.9 KB
Raw

kas-installer

KAS Installer allows the deployment and configuration of Managed Kafka Service in a single K8s cluster.

Contents

Prerequisites

  • jq
  • curl
  • OpenShift. In the future there are plans to make it compatible with native K8s. Currently an OpenShift dedicated based environment is needed (Currently needs to be a multi-zone cluster if you want to create a Kafka instance through the fleet manager by using managed_kafka.sh).
  • git
  • opm required to build custom kas-fleetshard OLM bundle from source (see Custom Components)
  • oc
  • kubectl
  • openssl CLI tool
  • rhoas CLI (https://github.com/redhat-developer/app-services-cli)
  • A user with administrative privileges in the OpenShift cluster and is logged in using oc or kubectl
  • yq if kas-fleet-manager-service-template-params is provided
  • OSD Cluster with the following specs. Clusters with fewer/smaller compute nodes may work, but have not been verified with kas-installer.
    • Plan developer.x1
      • 6 compute nodes
      • Size: m5.2xlarge
      • MultiAz: N/A
    • Plan standard.x1
      • 9 compute nodes (3 per zone)
      • Size: m5.2xlarge
      • MultiAz: True
    • Plan standard.x2
      • 12 compute nodes (4 per zone)
      • Size: m5.2xlarge
      • MultiAz: True

On Mac install:

  • brew gsed
  • brew coreutils
  • brew openssl

Description

KAS Installer deploys and configures the following components that are part of Managed Kafka Service:

  • MAS SSO
  • KAS Fleet Manager
  • Observability Operator (via KAS Fleet Manager)
  • sharded-nlb IngressController
  • KAS Fleet Shard and Strimzi Operators (via KAS Fleet Manager)

It deploys and configures the components to the cluster set in the user's kubeconfig file.

Additionally, a single Data Plane cluster is configured ready to be used, in the same cluster set in the user's kubeconfig file.

Usage

Deploy Managed Kafka Service

  1. Create and fill the KAS installer configuration file kas-installer.env. Minimally, the values identified as [required] in kas-installer-defaults.env must be configured.
  2. make sure you have run oc login --server=<api cluster url|https://api.xxx.openshiftapps.com:6443> to your target OSD cluster. You will be asked a password or a token
  3. Run the KAS installer kas-installer.sh to deploy and configure Managed Kafka Service
  4. Run uninstall.sh to remove KAS from the cluster. You should remove any deployed Kafkas before runnig this script.

Troubleshooting: If the installer crashed due to configuration error in kas-installer.env, you often can rerun the installer after fixing the config issue. It is not necessary to run uninstall before retrying.

Installation Modes

Standalone

Deploying a cluster with kas-fleet-manager in standalone is the default for kas-installer (when OCM_SERVICE_TOKEN is not defined). In this mode, the fleet manager deploys the data plane components from OLM bundles.

NOTE: In standalone mode, predefined bundles are used for Strimzi and KAS Fleetshard operators. To use a different bundle you'll need to build a dev bundle and set either STRIMZI_OLM_INDEX_IMAGE or KAS_FLEETSHARD_OLM_INDEX_IMAGE environment variables.

OCM

Installation with OCM mode allows kas-fleet-manager to deploy the data plane components as OCM addons. This mode can be used by setting OCM_SERVICE_TOKEN to your OCM offline token and also setting OCM_CLUSTER_ID to the idenfier for the OSD cluster used in deployment.

NOTE: In OCM mode, it may take up to 10 minutes after the kas-installer.sh completes before the addon installations are ready. Until ready, the fleet-manager API will reject new Kafka requests.

Enterprise

In addition to the configurations required by OCM Mode, Enterprise mode is enabled by setting ENTERPRISE_ENABLED='true' either via kas-installer.env or the environment.

Enterprise is a variation of OCM mode that allows for simple installation of only the control plane components, MAS SSO and kas-fleet-manager. Following installation this mode, the fleet manager's cluster list is (unless overridden) an empty list and no data plane clusters are created or reconciled initially.

Registering a Cluster

In this mode, users must register their own data plane clusters using the rhoas CLI's dedicated register-cluster command with arguments --cluster-mgmt-api-url set to https://api.stage.openshift.com and --access-token set to a production ocm token value. Assuming an active ocm session, the following command will register a new data plane cluster, making it available for placement of new Kafka instances.

rhoas dedicated register-cluster -v --cluster-mgmt-api-url https://api.stage.openshift.com --access-token ${OCM_PRODUCTION_TOKEN}

See the using rhoas CLI for more information on how to login and use rhoas with kas-installer.

NOTE: Registering a data plane cluster with kas-fleet-manager requires the KFM client to be an organization admin (JWT claim is_org_admin: true). If you are not an org admin in either RH SSO or RH SSO stage, you must configure SSO_PROVIDER_TYPE='mas_sso' and login to rhoas using your RH_USERNAME as username and password for this functionality to work.

De-Registering a Cluster

Enterprise/dedicated clusters may be de-registered using the deregister_cluster.sh script. This process will remove all Kafka instances from the cluster first, de-register it from kas-fleet-manager, and finally removal all remaining dataplane components such as CRDs and ingress resources specific to Managed Kafka. When there is only a single dedicated cluster registered, no cluster ID parameter needs to be given to deregister_cluster.sh. Otherwise, when multiple are registered, the ID to de-register is required.

NOTE: Using the uninstall.sh script will automatically delete all Kafka instances and de-register all registered clusters known to the installed kas-fleet-manager prior to uninstalling the Managed Kafka resources.

Fleet Manager Parameter Customization

Service Customization

The kas-installer.sh process will check for the presence of an executable named kas-fleet-manager-service-template-params in the project root. When available, it will be executed with the expection that key/value pairs will be output to stdout. The output will be used when processing the kas-fleet-manager's service-template.yml. The name of the executable is intentially missing an extension to indicate that any language may be used that is known to the user to be supported in their own environment.

Note that oc process requires that individual parameters are specified on a single line.

For example, to provide a custom KAS_FLEETSHARD_OPERATOR_SUBSCRIPTION_CONFIG parameter value to the fleet manager template, something like the following may be used for a kas-fleet-manager-service-template-params executable:

#!/bin/bash

# Declare the subscription config using multi-line YAML
MY_CUSTOM_SUBSCRIPTION_CONFIG='---
resources:
  requests:
    memory: "1Gi"
    cpu: "500m"
  limits:
    memory: "1Gi"
    cpu: "500m"
env:
- name: SSO_ENABLED
  value: "true"
- name: MANAGEDKAFKA_ADMINSERVER_EDGE_TLS_ENABLED
  value: "true"
- name: STANDARD_KAFKA_CONTAINER_CPU
  value: 500m
- name: STANDARD_KAFKA_CONTAINER_MEMORY
  value: 4Gi
- name: STANDARD_KAFKA_JVM_XMS
  value: 1G
- name: STANDARD_ZOOKEEPER_CONTAINER_CPU
  value: 500m
'

# Quota override for my organization
MY_CUSTOM_ORG_QUOTA='
- id: <YOUR ORG HERE>
  max_allowed_instances: 10
  any_user: true
  registered_users: []
  granted_quota:
  - instance_type_id: standard
    kafka_billing_models:
      - id: standard
        max_allowed_instances: 10
      - id: enterprise
        max_allowed_instances: 10
      - id: eval
        max_allowed_instances: 10
'

# Serialize to a single line as JSON (subset of YAML) to standard output
echo "KAS_FLEETSHARD_OPERATOR_SUBSCRIPTION_CONFIG='$(echo "${MY_CUSTOM_SUBSCRIPTION_CONFIG}" | yq e -o=json -I=0)'"

# Disable organization quotas (allows deployment of developer instances)
echo "REGISTERED_USERS_PER_ORGANISATION='[]'"
# Custom organization quota (allows deployment of standard instances), disabled/commented here
#echo "REGISTERED_USERS_PER_ORGANISATION='$(echo "${MY_CUSTOM_ORG_QUOTA}" | yq e -o=json -I=0)'"

Secret Customization

Similar to the kas-fleet-manager-service-template-params previously described, the kas-installer.sh process will check for the presence of an executable named kas-fleet-manager-secrets-template-params in the project root. When available, it will be executed with the expection that key/value pairs will be output to stdout. The output will be used when processing the kas-fleet-manager's secrets-template.yml.

Instance Types

The default kas-fleet-manager configuration enables the deployment of two instance types, standard and developer. Permission to create an instance of a particular type depends on the organization's quota for the user creating the instance.

  1. Creating a developer instance type requires that the user creating the instance does not have an instance quota. Use parameter customization to set the REGISTERED_USERS_PER_ORGANISATION property to an empty array [].
  2. Creating a standard instance type requires that the user creating the instance does have an instance quota. Use parameter customization to set the REGISTERED_USERS_PER_ORGANISATION property to an array containing the user's org. See the MY_CUSTOM_ORG_QUOTA variable in the sample script for an example.

To create a standard.x2 (or other non-default types) instance type, the plan must be provided to the Kafka create request. If using the managed_kafka.sh script, the --plan argument may be used:

managed_kafka.sh --create mykafka --plan standard.x2

Deploying to a single provider and region

The default installation configurations will deploy kas-fleet-manager with a single cloud provider and region. The provider name and region may be configured in the environment using the CLOUD_PROVIDER and REGION variables. See the kas-installer-defaults.env for the default values.

Depending on the cloud provider configured, additional provider-specific configuration may be required, for example GCP_API_CREDENTIALS for the gcp CLOUD_PROVIDER. Please see the documentation corresponding to the version of kas-fleet-manager in use for details.

Deploying to multiple clusters/regions

In order to allow clients to deploy Kafka clusters to multiple clusters and/or regions, the following parameters must be customized:

  • KUBE_CONFIG in kas-fleet-manager-secrets-template-params (standalone only)
  • SUPPORTED_CLOUD_PROVIDERS in kas-fleet-manager-service-template-params
  • CLUSTER_LIST in kas-fleet-manager-service-template-params

The following sections describe the contents of these parameters in detail.

SUPPORTED_CLOUD_PROVIDERS

The SUPPORTED_CLOUD_PROVIDERS parameter contains a list of supported cloud providers in a yaml format. See deploying-kas-fleet-manager-to-openshift.md for more details.

CLUSTER_LIST

The CLUSTER_LIST parameter contains a list clusters for kas fleet manager. See data-plane-osd-cluster-options.md for more details.

KUBE_CONFIG (standalone only)

When running in standalone mode, the KUBE_CONFIG contains cluster connection information for each cluster to which Kafka can be deployed. There should be one entry in KUBE_CONFIG for each entry in CLUSTER_LIST. For example, to generate a KUBE_CONFIG for two clusters:

oc login -u kubeadmin -p <first_cluster_password> <first_cluster_url>
oc config view --minify --raw > /tmp/c1.yaml
oc login -u kubeadmin -p <second_cluster_password> <second_cluster_url>
oc config view --minify --raw > /tmp/c2.yaml
(KUBECONFIG=/tmp/c1.yaml:/tmp/c2.yaml oc config view --raw)

The KUBE_CONFIG value must be base64 encoded. For example, the following may be used to encode the value in the kas-fleet-manager-secrets-template-params:

echo "KUBE_CONFIG='$(echo "${KUBE_CONFIG}"  | yq -o=json -I=0 |  ${BASE64} -w0)'"

Custom Domain for the Data Plane

Custom domain name registration for the data plane routes may be enabled using the following configurations in the kas-fleet-manager customization scripts. If KAFKA_DOMAIN_NAME is not specified, the default value of in the KFM template will be used.

  1. kas-fleet-manager-service-template-params 
    echo "ENABLE_KAFKA_CNAME_REGISTRATION='true'"
echo "KAFKA_DOMAIN_NAME='<your domain here>'"
  2. kas-fleet-manager-secrets-template-params 
     echo "ROUTE53_ACCESS_KEY='<your AWS Route53 access key>'"
 echo "ROUTE53_SECRET_ACCESS_KEY='<your AWS Route53 secret access key>'"

With this configuration, you will likely also want to provide a certificate and key, either by generating one using the custom TLS instructions, or by directly providing values for KAFKA_TLS_CERT and KAFKA_TLS_KEY in the kas-installer.env file or environment.

Observability Operator

The version of the Observability Operator deployed by kas-fleet-manager can be configured with the following parameters in kas-fleet-manager-service-template-params:

  • OBSERVABILITY_OPERATOR_INDEX_IMAGE
  • OBSERVABILITY_OPERATOR_STARTING_CSV

SSO Providers

Configuration of kas-fleet-manager's SSO providers is done by setting the SSO_PROVIDER_TYPE configuration variable. When not set, the default provider is mas_sso. To use RH SSO, the variable may be set to redhat_sso and additional configuration can be provided for REDHAT_SSO_HOSTNAME (default sso.stage.redhat.com), REDHAT_SSO_REALM (default redhat-external), REDHAT_SSO_CLIENT_ID (required), and REDHAT_SSO_CLIENT_SECRET (required). See the description for each variable in the kas-installer-defaults.env file for more information.

kas-fleet-manager admin API SSO

The kas-fleet-manager will use MAS-SSO as an identity provider for its admin API endpoints by default. An alternate IdP may be provided by setting ADMIN_API_SSO_BASE_URL, ADMIN_API_SSO_REALM, and ADMIN_API_SSO_ENDPOINT_URI in your kas-installer.env configuration, or in the environment. See kas-installer-defaults.env for the default values.

Users that need to configure a different set of client roles to authorize the admin API operation should use the ADMIN_AUTHZ_CONFIG variable to override the defaults. This variable can be set using fleet manager parameter customization. See the ADMIN_AUTHZ_CONFIG parameter in kas-fleet-manager's templates/service-template.yml file for the default value.

A token for use with the admin API endpoints by calling ./get_access_token.sh --sre-admin (this is used internally by managed_kafka.sh when the --admin flag is provided). By default, the script is configured for use with a local MAS-SSO instance which has a client pre-configured for this purpose, kafka-admin. To provide a custom service account for use with the admin endpoints, provide values for the ADMIN_API_CLIENT_ID and ADMIN_API_CLIENT_SECRET configuration values.

MAS SSO Resources

MAS SSO (must always available to support kas-fleet-manager admin operations) may optionally be installed with modified resource requests and limits. The two environment variable should be in JSON format on a single line.

Configuration of the SSO operator can be done using the MAS_SSO_OPERATOR_SUBSCRIPTION_CONFIG variable, the format of which must conform to the Subscription Config object. Example with custom CPU and memory requests:

MAS_SSO_OPERATOR_SUBSCRIPTION_CONFIG='{"resources": { "requests": { "cpu": "500m", "memory": "512Mi" }, "limits": { "cpu": "500m", "memory": "512Mi" }}}'

The MAS SSO Keycloak instance may be configured using the MAS_SSO_KEYCLOAK_RESOURCES variable. The format is the same as the Kubernetes resources object for other types. See mas-sso/keycloak.yaml for the defaults. Example custom CPU and memory requests:

MAS_SSO_KEYCLOAK_RESOURCES='{ "requests": { "cpu": "500m", "memory": "768Mi" }, "limits": { "cpu": "500m", "memory": "768Mi" }}'

Custom Components

Custom-built components are supported for kas-fleet-manager and kas-fleetshard.

  1. kas-fleet-manager: see the documentation for KAS_FLEET_MANAGER_IMAGE_BUILD in the kas-installer-defaults.env file.
  2. kas-fleetshard: prior to running kas-installer.sh, execute operators/generate-kas-fleetshard-olm-bundle.sh, passing the required parameters (see operators/generate-kas-fleetshard-olm-bundle.sh --help for details). The script will build kas-fleetshard from source and an OLM bundle index, optionally updating the kas-installer.env file with the configuration. Running kas-installer.sh will deploy the custom-built operator.

Using rhoas CLI

Use ./rhoas_login.sh as a short cut to login to the CLI. Login using the username you specified as RH_USERNAME in the env file. The password is the same as the RH_USERNAME value. Alternatively (and likely preferably), use ./login-all.sh which makes sure you are also logged to your OpenShift Dedicated cluster.

There are a couple of things that are expected not to work when using the RHOAS CLI with a kas-installer installed instance. These are noted below.

Service Account Maintenace

  1. To create an account, run rhoas service-account create --short-description foo --file-format properties.
  2. To list existing service accounts, run rhoas service-account list.
  3. To remove an existing service account, run rhoas service-account delete --id=<ID of service account>.

Kafka Instance Maintenance

  1. To create a cluster, run rhoas kafka create --bypass-checks --provider aws --region us-east-1 --name <clustername>. Note that --bypass-checks is required as the T&Cs endpoint will not exist in your environment. The provider and region must be passed on the command line.
  2. To list existing clusters, run rhoas kafka list
  3. To remove an existing cluster, run rhoas kafka delete --name <clustername>.

Kafka topics / consumergroups / ACL

To use these cli featurs, you must set MANAGEDKAFKA_ADMINSERVER_EDGE_TLS_ENABLED=true in your kas-installer.env so that the admin-server will run over TLS (edge terminated).

  1. To create a topic rhoas kafka topic create --name=foo
  2. To grant access rhoas kafka acl grant-access --topic=foo --all-accounts --producer etc.

Running the User Interface

Only with SSO_PROVIDER_TYPE=redhat_sso and REDHAT_SSO_HOSTNAME=sso.redhat.com (production)

See the Running the UI wiki page for more detailed instructions.

Local instances of the app-services-ui, kas-ui, and kafka-ui may be run locally by using the ui/install.sh script. Running the UI installation will start three containers using podman or docker (auto-detected, but can be forced by setting CONTAINER_CLI to docker or podman) with a main entrypoint of https://127.0.0.1:1337. The IP must be configured with name prod.foo.redhat.com in the user's local /etc/hosts file.

...
127.0.0.1 prod.foo.redhat.com
...

The repository and branch (or tag/commit) maybe configured in the kas-installer.env file using the following variables

  • APP_SERVICES_UI_GIT_URL
  • APP_SERVICES_UI_GIT_REF
  • KAS_UI_GIT_URL
  • KAS_UI_GIT_REF
  • KAFKA_UI_GIT_URL
  • KAFKA_UI_GIT_REF

Note, when navigating to https://prod.foo.redhat.com:1337/, you may be prompted to login to sso.redhat.com as well as the local MAS-SSO instance. The credentials for MAS-SSO should be the value of the RH_USERNAME variable for both username and password.

Custom TLS

Users may provide custom-generated TLS certificates using the gen-certs.sh script. The output is placed in the certs directory (ignored by git) and includes a CA certificate, CA key, server certificate, and server key. Each time the script is run, the server files will be replaced, but the CA certificate and key will be retained. The server certificate is issued specifically for the current session's domain name, as determined by the K8S_CLUSTER_DOMAIN variable. To configure the certificates for a Kafka instance, set the following variables in kas-installer.env, where KAS_INSTALLER_HOME is the path to the project root:

KAFKA_TLS_CERT="$(cat ${KAS_INSTALLER_HOME}/certs/server-cert.pem)"
KAFKA_TLS_KEY="$(cat ${KAS_INSTALLER_HOME}/certs/server-key.pem)"

The CA certificate may be (for example) imported to your browser, enabling the generated server certificates used by the admin API endpoint and the UI to be trusted during testing without the needing to trust them individually.

Legacy scripts

Please favour using the rhoas command line. These scripts will be remove at some point soon.

Service Account Maintenance

The service_account.sh script supports creating, listing, and deleting service accounts.

  1. To create an account, run service_account.sh --create. The new service account information will be printed to the console. Be sure to retain the clientID and clientSecret values to use when generating an access token or for connecting to Kafka directly.
  2. To list existing service accounts, run service_account.sh --list.
  3. To remove an existing service account, run service_account.sh --delete <ID of service account>.

Generate an Access Token

  1. Run get_access_token.sh using the clientID and clientSecret as the first and second arguments. The generated access token and its expiration date and time will be printed to the console.

Kafka Instance Maintenance

The managed_kafka.sh script supports creating, listing, and deleting Kafka clusters.

  1. To create a cluster, run managed_kafka.sh --create <cluster name>. Progress will be printed as the cluster is prepared and provisioned. If kas-fleet-manager has been configured with more than one cloud provider or region, the --provider and/or --region arguments must be provided to the managed_kafka.sh script. Otherwise, the script will discover the single provider and region using the fleet manager's cloud provider endpoints.
  2. To list existing clusters, run managed_kafka.sh --list.
  3. To remove an existing cluster, run managed_kafka.sh --delete <cluster ID>.
  4. To patch an existing cluster (for instance changing a strimzi version), run managed_kafka.sh --admin --patch <cluster ID> '{ "strimzi_version": "strimzi-cluster-operator.v0.23.0-3" }'
  5. To use kafka bin scripts against pre existing kafka cluster, run managed_kafka.sh --certgen <kafka id> <Service_Account_ID> <Service_Account_Secret>. If you do not pass the <Service_Account_ID> <Service_Account_Secret> arguments, the script will attempt to create a Service_Account for you. The cert generation is already performed at the end of --create. Point the --command-config flag to the generated app-services.properties in the working directory.
  • If there is already 2 service accounts pre-existing you must delete 1 of them for this script to work

Access the Kafka Cluster using command line tools

To use the Kafka Cluster that is created with the managed_kafka.sh script with command line tools like kafka-topics.sh or kafka-console-consumer.sh do the following.

  1. run tool_access.sh <cluster name>. This will generate the certificate / truststore and app-services.properties file. It will also create a service account and grant GROUP and TOPIC permissions. The bootstrap host will be displayed upon completion, and is also in the properties file as bootstrap.servers.
  2. Execute your tool like kafka-topics.sh --bootstrap-server <bootstrap-host>:443 --command-config app-services.properties --topic foo --create --partitions 9
  3. If you want to use a different service account, you may edit the app-services.properties file and update the username and password with clientID and clientSecret

Running E2E Test Suite (experimental)

  1. Install all cluster components using kas-installer.sh, ensure you enable MANAGEDKAFKA_ADMINSERVER_EDGE_TLS_ENABLED as documented in service-customization

  2. Clone the e2e-test-suite repository locally and change directory to the test suite project root

  3. Generate the test suite configuration with ${KAS_INSTALLER_DIR}/e2e-test-config.sh > config.json. When using kas-installer.sh with the redhat_sso provider type, the user accounts to be used by the E2E tests must be preconfigured. The e2e-test-config.sh script will set all variables named like E2E_USER_ in the test suite's config.json file. The E2E_USER_ prefix is not included in the JSON configuration.

    For example, an environment variable E2E_USER_PRIMARY_USERNAME='my-primary-user' will be added to config.json as key/value pair "PRIMARY_USERNAME": "my-primary-user".

  4. Execute individual test classes:

    • ./hack/testrunner.sh test KafkaAdminPermissionTest
    • ./hack/testrunner.sh test KafkaInstanceAPITest
    • ./hack/testrunner.sh test KafkaCLITest