EventSource High Availability¶

EventSource controller creates a k8s deployment (replica number defaults to 1) for each EventSource object to watch the events. HA can be achieved by setting spec.replicas to a number greater than 1.

Some types of the event sources do not allow multiple live clients with same attributes (i.e. multiple clients with same clientID connecting to a NATS server), or multiple event source PODs will generate duplicated events to downstream, so the HA strategies are different for different event sources.

Please DO NOT manually scale up the replicas, that might cause unexpected behaviors!

Active-Active¶

Active-Active strategy is applied to the following EventSource types.

AWS SNS
AWS SQS
Bitbucket
Bitbucket Server
GitHub
GitLab
NetApp Storage GRID
Slack
Stripe
Webhook

When spec.replicas is set to N (N > 1), all the N Pods serve traffic.

Active-Passive¶

If following EventSource types have spec.replicas > 1, Active-Passive strategy is used, which means only one Pod serves traffic and the rest ones stand by. One of standby Pods will be automatically elected to be active if the old one is gone.

AMQP
Azure Events Hub
Calendar
Emitter
GCP PubSub
Generic
File
HDFS
Kafka
Minio
MQTT
NATS
NSQ
Pulsar
Redis
Resource

Kubernetes Leader Election¶

By default, Argo Events will use NATS for the HA leader election except when using a Kafka Eventbus, in which case a kubernetes leader election will be used. If using a different EventBus you can opt-in to a Kubernetes native leader election by specifying the following annotation.

annotations:
  events.argoproj.io/leader-election: k8s

To use Kubernetes leader election the following RBAC rules need to be associated with the EventSource ServiceAccount.

apiVersion: rbac.authorization.k8s.io/v1
kind: Role
metadata:
  name: argo-events-leaderelection-role
rules:
- apiGroups: ["coordination.k8s.io"]
  resources: ["leases"]
  verbs:     ["get", "create", "update"]

More¶

Click here to learn more information about Argo Events DR/HA recommendations.