publicissapient-france / xebikart-infra Goto Github PK

And potentially add TLS cert handling and ExternalDNS record on it!

• https://circleci.com/docs/2.0/contexts/

Context called XebiKart - GCP/GKE Deployment with:

You may avoid 2 distinct deployments by using Helm hooks: https://github.com/helm/helm/blob/master/docs/charts_hooks.md#defining-a-crd-with-the-crd-install-hook.

Originally posted by @clook in #43 (comment)

See https://www.terraform.io/docs/providers/github/r/repository.html

In order to kill unhealthy pods automatically, we should declare a livenessProbe and a readinessProbe for RabbitMQ pods

Some pointers

• rabbitmqctl status
• rabbitmqctl node_health_check -n rabbit@hostname
• RabbitMQ REST API itself: /api/aliveness-test/<vhost | urlEncoded>
• Prometheus metrics

At this moment, RabbitMQ k8s ressources have "limitless" ressources. We need to set up coherent limits in order to optimize ressources utilisation. To do so, we need to modify Helm Chart

CF Terraform Doc recommendations.

For example, let's say a cluster service such as Tiller (for Helm) need some RBAC config (ServiceAccount, ClusterRoleBindings, etc) and a Deployment. Should we keep all of this in a single yaml file or should we split it into multiple YAML files?

This issue as a reminder as asked by @Horgix during XKE.

See https://www.terraform.io/upgrade-guides/0-12.html

When we use many RabbitMQ replicas, some messages are lost.This bug have been detected by @sergiords.

How to reproduce the bug:

• Verify that there're more than one replica of RabbitMQ deployed in GKE
• Use a MQTT client (like mosquito_sub) to subscribe to a topic
• Run a little script that publishes somme messages into a topic MQTT with mosquito_pub or other MQTT client
• Verify that somme messages are lost

This behavior is not reproductible with only one replica or when installing Helm Chart in minikube

As stated in the ADR/006, the user/password authentication is a tradeoff regarding security VS availability for teams that will need to be adressed.