Deploy an AWS Lambda to disable a non-responding site

This guide explains how to resolve a split-brain scenarios between two sites in a multi-site deployment. It also disables replication if one site fails, so the other site can continue to serve requests.

This deployment is intended to be used with the setup described in the Concepts for multi-site deployments guide. Use this deployment with the other building blocks outlined in the Building blocks multi-site deployments guide.

We provide these blueprints to show a minimal functionally complete example with a good baseline performance for regular installations. You would still need to adapt it to your environment and your organization’s standards and security best practices.

Architecture

In the event of a network communication failure between sites in a multi-site deployment, it is no longer possible for the two sites to continue to replicate data between them. The Infinispan is configured with a FAIL failure policy, which ensures consistency over availability. Consequently, all user requests are served with an error message until the failure is resolved, either by restoring the network connection or by disabling cross-site replication.

In such scenarios, a quorum is commonly used to determine which sites are marked as online or offline. However, as multi-site deployments only consist of two sites, this is not possible. Instead, we leverage “fencing” to ensure that when one of the sites is unable to connect to the other site, only one site remains in the loadbalancer configuration, and hence only this site is able to serve subsequent users requests.

In addition to the loadbalancer configuration, the fencing procedure disables replication between the two Infinispan clusters to allow serving user requests from the site that remains in the loadbalancer configuration. As a result, the sites will be out-of-sync once the replication has been disabled.

To recover from the out-of-sync state, a manual re-sync is necessary as described in Synchronize Sites. This is why a site which is removed via fencing will not be re-added automatically when the network communication failure is resolved. The remove site should only be re-added once the two sites have been synchronized using the outlined procedure Bring site online.

In this guide we describe how to implement fencing using a combination of Prometheus Alerts and AWS Lambda functions. A Prometheus Alert is triggered when split-brain is detected by the Infinispan server metrics, which results in the Prometheus AlertManager calling the AWS Lambda based webhook. The triggered Lambda function inspects the current Global Accelerator configuration and removes the site reported to be offline.

In a true split-brain scenario, where both sites are still up but network communication is down, it is possible that both sites will trigger the webhook simultaneously. We guard against this by ensuring that only a single Lambda instance can be executed at a given time. The logic in the AWS Lambda ensures that always one site entry remains in the loadbalancer configuration.

Prerequisites

ROSA HCP based multi-site Keycloak deployment
AWS CLI Installed
AWS Global Accelerator loadbalancer
jq tool installed

Procedure

Enable Openshift user alert routing

Command:

kubectl apply -f - << EOF
apiVersion: v1
kind: ConfigMap
metadata:
  name: user-workload-monitoring-config
  namespace: openshift-user-workload-monitoring
data:
  config.yaml: |
    alertmanager:
      enabled: true
      enableAlertmanagerConfig: true
EOF
kubectl -n openshift-user-workload-monitoring rollout status --watch statefulset.apps/alertmanager-user-workload

Decide upon a username/password combination which will be used to authenticate the Lambda webhook and create an AWS Secret storing the password
Command:
```
aws secretsmanager create-secret \
  --name webhook-password \ (1)
  --secret-string changeme \ (2)
  --region eu-west-1 (3)
```
1 The name of the secret

2 The password to be used for authentication

3 The AWS region that hosts the secret

Create the Role used to execute the Lambda.

Command:

FUNCTION_NAME= (1)
ROLE_ARN=$(aws iam create-role \
  --role-name ${FUNCTION_NAME} \
  --assume-role-policy-document \
  '{
    "Version": "2012-10-17",
    "Statement": [
      {
        "Effect": "Allow",
        "Principal": {
          "Service": "lambda.amazonaws.com"
        },
        "Action": "sts:AssumeRole"
      }
    ]
  }' \
  --query 'Role.Arn' \
  --region eu-west-1 \ (2)
  --output text
)

1	A name of your choice to associate with the Lambda and related resources
2	The AWS Region hosting your Kubernetes clusters

Create and attach the 'LambdaSecretManager' Policy so that the Lambda can access AWS Secrets

Command:

POLICY_ARN=$(aws iam create-policy \
  --policy-name LambdaSecretManager \
  --policy-document \
  '{
      "Version": "2012-10-17",
      "Statement": [
          {
              "Effect": "Allow",
              "Action": [
                  "secretsmanager:GetSecretValue"
              ],
              "Resource": "*"
          }
      ]
  }' \
  --query 'Policy.Arn' \
  --output text
)
aws iam attach-role-policy \
  --role-name ${FUNCTION_NAME} \
  --policy-arn ${POLICY_ARN}

Attach the ElasticLoadBalancingReadOnly policy so that the Lambda can query the provisioned Network Load Balancers

Command:

aws iam attach-role-policy \
  --role-name ${FUNCTION_NAME} \
  --policy-arn arn:aws:iam::aws:policy/ElasticLoadBalancingReadOnly

Attach the GlobalAcceleratorFullAccess policy so that the Lambda can update the Global Accelerator EndpointGroup

Command:

aws iam attach-role-policy \
  --role-name ${FUNCTION_NAME} \
  --policy-arn arn:aws:iam::aws:policy/GlobalAcceleratorFullAccess

Create a Lambda ZIP file containing the required fencing logic

Command:

LAMBDA_ZIP=/tmp/lambda.zip
cat << EOF > /tmp/lambda.py

from urllib.error import HTTPError

import boto3
import jmespath
import json
import os
import urllib3

from base64 import b64decode
from urllib.parse import unquote

# Prevent unverified HTTPS connection warning
urllib3.disable_warnings(urllib3.exceptions.InsecureRequestWarning)


class MissingEnvironmentVariable(Exception):
    pass


class MissingSiteUrl(Exception):
    pass


def env(name):
    if name in os.environ:
        return os.environ[name]
    raise MissingEnvironmentVariable(f"Environment Variable '{name}' must be set")


def handle_site_offline(labels):
    a_client = boto3.client('globalaccelerator', region_name='us-west-2')

    acceleratorDNS = labels['accelerator']
    accelerator = jmespath.search(f"Accelerators[?(DnsName=='{acceleratorDNS}'|| DualStackDnsName=='{acceleratorDNS}')]", a_client.list_accelerators())
    if not accelerator:
        print(f"Ignoring SiteOffline alert as accelerator with DnsName '{acceleratorDNS}' not found")
        return

    accelerator_arn = accelerator[0]['AcceleratorArn']
    listener_arn = a_client.list_listeners(AcceleratorArn=accelerator_arn)['Listeners'][0]['ListenerArn']

    endpoint_group = a_client.list_endpoint_groups(ListenerArn=listener_arn)['EndpointGroups'][0]
    endpoints = endpoint_group['EndpointDescriptions']

    # Only update accelerator endpoints if two entries exist
    if len(endpoints) > 1:
        # If the reporter endpoint is not healthy then do nothing for now
        # A Lambda will eventually be triggered by the other offline site for this reporter
        reporter = labels['reporter']
        reporter_endpoint = [e for e in endpoints if endpoint_belongs_to_site(e, reporter)][0]
        if reporter_endpoint['HealthState'] == 'UNHEALTHY':
            print(f"Ignoring SiteOffline alert as reporter '{reporter}' endpoint is marked UNHEALTHY")
            return

        offline_site = labels['site']
        endpoints = [e for e in endpoints if not endpoint_belongs_to_site(e, offline_site)]
        del reporter_endpoint['HealthState']
        a_client.update_endpoint_group(
            EndpointGroupArn=endpoint_group['EndpointGroupArn'],
            EndpointConfigurations=endpoints
        )
        print(f"Removed site={offline_site} from Accelerator EndpointGroup")

        take_infinispan_site_offline(reporter, offline_site)
        print(f"Backup site={offline_site} caches taken offline")
    else:
        print("Ignoring SiteOffline alert only one Endpoint defined in the EndpointGroup")


def endpoint_belongs_to_site(endpoint, site):
    lb_arn = endpoint['EndpointId']
    region = lb_arn.split(':')[3]
    client = boto3.client('elbv2', region_name=region)
    tags = client.describe_tags(ResourceArns=[lb_arn])['TagDescriptions'][0]['Tags']
    for tag in tags:
        if tag['Key'] == 'site':
            return tag['Value'] == site
    return false


def take_infinispan_site_offline(reporter, offlinesite):
    endpoints = json.loads(INFINISPAN_SITE_ENDPOINTS)
    if reporter not in endpoints:
        raise MissingSiteUrl(f"Missing URL for site '{reporter}' in 'INFINISPAN_SITE_ENDPOINTS' json")

    endpoint = endpoints[reporter]
    password = get_secret(INFINISPAN_USER_SECRET)
    url = f"https://{endpoint}/rest/v2/container/x-site/backups/{offlinesite}?action=take-offline"
    http = urllib3.PoolManager(cert_reqs='CERT_NONE')
    headers = urllib3.make_headers(basic_auth=f"{INFINISPAN_USER}:{password}")
    try:
        rsp = http.request("POST", url, headers=headers)
        if rsp.status >= 400:
            raise HTTPError(f"Unexpected response status '%d' when taking site offline", rsp.status)
        rsp.release_conn()
    except HTTPError as e:
        print(f"HTTP error encountered: {e}")


def get_secret(secret_name):
    session = boto3.session.Session()
    client = session.client(
        service_name='secretsmanager',
        region_name=SECRETS_REGION
    )
    return client.get_secret_value(SecretId=secret_name)['SecretString']


def decode_basic_auth_header(encoded_str):
    split = encoded_str.strip().split(' ')
    if len(split) == 2:
        if split[0].strip().lower() == 'basic':
            try:
                username, password = b64decode(split[1]).decode().split(':', 1)
            except:
                raise DecodeError
        else:
            raise DecodeError
    else:
        raise DecodeError

    return unquote(username), unquote(password)


def handler(event, context):
    print(json.dumps(event))

    authorization = event['headers'].get('authorization')
    if authorization is None:
        print("'Authorization' header missing from request")
        return {
            "statusCode": 401
        }

    expectedPass = get_secret(WEBHOOK_USER_SECRET)
    username, password = decode_basic_auth_header(authorization)
    if username != WEBHOOK_USER and password != expectedPass:
        print('Invalid username/password combination')
        return {
            "statusCode": 403
        }

    body = event.get('body')
    if body is None:
        raise Exception('Empty request body')

    body = json.loads(body)
    print(json.dumps(body))

    if body['status'] != 'firing':
        print("Ignoring alert as status is not 'firing', status was: '%s'" % body['status'])
        return {
            "statusCode": 204
        }

    for alert in body['alerts']:
        labels = alert['labels']
        if labels['alertname'] == 'SiteOffline':
            handle_site_offline(labels)

    return {
        "statusCode": 204
    }


INFINISPAN_USER = env('INFINISPAN_USER')
INFINISPAN_USER_SECRET = env('INFINISPAN_USER_SECRET')
INFINISPAN_SITE_ENDPOINTS = env('INFINISPAN_SITE_ENDPOINTS')
SECRETS_REGION = env('SECRETS_REGION')
WEBHOOK_USER = env('WEBHOOK_USER')
WEBHOOK_USER_SECRET = env('WEBHOOK_USER_SECRET')

EOF
zip -FS --junk-paths ${LAMBDA_ZIP} /tmp/lambda.py

Create the Lambda function.

Command:

aws lambda create-function \
  --function-name ${FUNCTION_NAME} \
  --zip-file fileb://${LAMBDA_ZIP} \
  --handler lambda.handler \
  --runtime python3.12 \
  --role ${ROLE_ARN} \
  --region eu-west-1 (1)

1	The AWS Region hosting your Kubernetes clusters

Expose a Function URL so the Lambda can be triggered as webhook

Command:

aws lambda create-function-url-config \
  --function-name ${FUNCTION_NAME} \
  --auth-type NONE \
  --region eu-west-1 (1)

1	The AWS Region hosting your Kubernetes clusters

Allow public invocations of the Function URL

Command:

aws lambda add-permission \
  --action "lambda:InvokeFunctionUrl" \
  --function-name ${FUNCTION_NAME} \
  --principal "*" \
  --statement-id FunctionURLAllowPublicAccess \
  --function-url-auth-type NONE \
  --region eu-west-1 (1)

1	The AWS Region hosting your Kubernetes clusters

Configure the Lambda’s Environment variables:

In each Kubernetes cluster, retrieve the exposed Infinispan URL endpoint:
```
kubectl -n ${NAMESPACE} get route infinispan-external -o jsonpath='{.status.ingress[].host}' (1)
```
1 Replace ${NAMESPACE} with the namespace containing your Infinispan server

Upload the desired Environment variables

ACCELERATOR_NAME= (1)
LAMBDA_REGION= (2)
CLUSTER_1_NAME= (3)
CLUSTER_1_ISPN_ENDPOINT= (4)
CLUSTER_2_NAME= (5)
CLUSTER_2_ISPN_ENDPOINT= (6)
INFINISPAN_USER= (7)
INFINISPAN_USER_SECRET= (8)
WEBHOOK_USER= (9)
WEBHOOK_USER_SECRET= (10)

INFINISPAN_SITE_ENDPOINTS=$(echo "{\"${CLUSTER_NAME_1}\":\"${CLUSTER_1_ISPN_ENDPOINT}\",\"${CLUSTER_2_NAME}\":\"${CLUSTER_2_ISPN_ENDPOINT\"}" | jq tostring)
aws lambda update-function-configuration \
    --function-name ${ACCELERATOR_NAME} \
    --region ${LAMBDA_REGION} \
    --environment "{
      \"Variables\": {
        \"INFINISPAN_USER\" : \"${INFINISPAN_USER}\",
        \"INFINISPAN_USER_SECRET\" : \"${INFINISPAN_USER_SECRET}\",
        \"INFINISPAN_SITE_ENDPOINTS\" : ${INFINISPAN_SITE_ENDPOINTS},
        \"WEBHOOK_USER\" : \"${WEBHOOK_USER}\",
        \"WEBHOOK_USER_SECRET\" : \"${WEBHOOK_USER_SECERT}\",
        \"SECRETS_REGION\" : \"eu-central-1\"
      }
    }"

1	The name of the AWS Global Accelerator used by your deployment
2	The AWS Region hosting your Kubernetes cluster and Lambda function
3	The name of one of your Infinispan sites as defined in Deploy Infinispan for HA with the Infinispan Operator
4	The Infinispan endpoint URL associated with the CLUSER_1_NAME site
5	The name of the second Infinispan site
6	The Infinispan endpoint URL associated with the CLUSER_2_NAME site
7	The username of a Infinispan user which has sufficient privileges to perform REST requests on the server
8	The name of the AWS secret containing the password associated with the Infinispan user
9	The username used to authenticate requests to the Lambda Function
10	The name of the AWS secret containing the password used to authenticate requests to the Lambda function

Retrieve the Lambda Function URL

Command:

aws lambda get-function-url-config \
  --function-name ${FUNCTION_NAME} \
  --query "FunctionUrl" \
  --region eu-west-1 \(1)
  --output text

1	The AWS region where the Lambda was created

Output:

https://tjqr2vgc664b6noj6vugprakoq0oausj.lambda-url.eu-west-1.on.aws

In each Kubernetes cluster, configure a Prometheus Alert routing to trigger the Lambda on split-brain

Command:

NAMESPACE= # The namespace containing your deployments
kubectl apply -n ${NAMESPACE} -f - << EOF
apiVersion: v1
kind: Secret
type: kubernetes.io/basic-auth
metadata:
  name: webhook-credentials
stringData:
  username: 'keycloak' (1)
  password: 'changme' (2)
---
apiVersion: monitoring.coreos.com/v1beta1
kind: AlertmanagerConfig
metadata:
  name: example-routing
spec:
  route:
    receiver: default
    groupBy:
      - accelerator
    groupInterval: 90s
    groupWait: 60s
    matchers:
      - matchType: =
        name: alertname
        value: SiteOffline
  receivers:
    - name: default
      webhookConfigs:
        - url: 'https://tjqr2vgc664b6noj6vugprakoq0oausj.lambda-url.eu-west-1.on.aws/' (3)
          httpConfig:
            basicAuth:
              username:
                key: username
                name: webhook-credentials
              password:
                key: password
                name: webhook-credentials
            tlsConfig:
              insecureSkipVerify: true
---
apiVersion: monitoring.coreos.com/v1
kind: PrometheusRule
metadata:
  name: xsite-status
spec:
  groups:
    - name: xsite-status
      rules:
        - alert: SiteOffline
          expr: 'min by (namespace, site) (vendor_jgroups_site_view_status{namespace="default",site="site-b"}) == 0' (4)
          labels:
            severity: critical
            reporter: site-a (5)
            accelerator: a3da6a6cbd4e27b02.awsglobalaccelerator.com (6)

1	The username required to authenticate Lambda requests
2	The password required to authenticate Lambda requests
3	The Lambda Function URL
4	The namespace value should be the namespace hosting the Infinispan CR and the site should be the remote site defined by `spec.service.sites.locations[0].name` in your Infinispan CR
5	The name of your local site defined by `spec.service.sites.local.name` in your Infinispan CR
6	The DNS of your Global Accelerator

Verify

To test that the Prometheus alert triggers the webhook as expected, perform the following steps to simulate a split-brain:

In each of your clusters execute the following:

Command:

kubectl -n openshift-operators scale --replicas=0 deployment/infinispan-operator-controller-manager (1)
kubectl -n openshift-operators rollout status -w deployment/infinispan-operator-controller-manager
kubectl -n ${NAMESPACE} scale --replicas=0 deployment/infinispan-router (2)
kubectl -n ${NAMESPACE} rollout status -w deployment/infinispan-router

1	Scale down the Infinispan Operator so that the next step does not result in the deployment being recreated by the operator
2	Scale down the Gossip Router deployment.Replace `${NAMESPACE}` with the namespace containing your Infinispan server

Verify the SiteOffline event has been fired on a cluster by inspecting the Observe → Alerting menu in the Openshift console
Inspect the Global Accelerator EndpointGroup in the AWS console and there should only be a single endpoint present

Scale up the Infinispan Operator and Gossip Router to re-establish a connection between sites:

Command:

kubectl -n openshift-operators scale --replicas=1 deployment/infinispan-operator-controller-manager
kubectl -n openshift-operators rollout status -w deployment/infinispan-operator-controller-manager
kubectl -n ${NAMESPACE} scale --replicas=1 deployment/infinispan-router (1)
kubectl -n ${NAMESPACE} rollout status -w deployment/infinispan-router

1	Replace `${NAMESPACE}` with the namespace containing your Infinispan server

Inspect the vendor_jgroups_site_view_status metric in each site. A value of 1 indicates that the site is reachable.
Update the Accelerator EndpointGroup to contain both Endpoints. See the Bring site online guide for details.

1	The name of the secret
2	The password to be used for authentication
3	The AWS region that hosts the secret

Deploy an AWS Lambda to disable a non-responding site

Architecture

Prerequisites

Procedure

Verify

Further reading