CODEWITHSUNDEEP
google-cloud-platformgoogle-kubernetes-engine
jlim
jlim

Reputation: 1039

GKE is spinning all node pool in one single zone safe?

This is a question specifically to public cloud which I'm familiar with such as GCP. Not too sure whether it applies to AWS, Azure, DigitalOcean, or other cloud provider.

So my question is, assuming I have provision 3 Kafka instances, 3 Redis instances (HA) and 3 MongoDB instance. Instead of separating each instance into separate zone, it is safe to provision all instances in one single zone? How stable or how often a cloud provider zone goes down ?

For example

Kafka

When region us-west1 in zone-a goes down, does that mean all my 3 instances will be down? My question is, has this ever happen in GCP ? If yes, how long does it need to recover in that zone?

instance1 - us-west1-a
instance2 - us-west1-a
instance3 - us-west1-a

In GCP, when we provision GKE regional cluster, there are 2 ways we can provision the node-pool namely:

https://cloud.google.com/kubernetes-engine/docs/how-to/creating-a-regional-cluster

Another question is, in what situation we should provision single zone, and when we should provision multizone nodepool.

Upvotes: 1

Views: 444

Answers (1)

dany L
dany L

Reputation: 2654

Deploying a regional cluster is "safer". A zone is a deployment area for resources for a region, and is isolated from each other. So zones are single points of failures.

Failures always happen however have a look at Google SLAs for GKE in order to get a sense of what is "safer":

Zonal Cluster (control plane)   99.5%
Regional Cluster (control plane)    99.95%

For recovery, I would say short, as Google has the infrastructure and resources for a fast recovery.

Also have a look at the previous 365 days of the status dashboard for GCP to get a sense of how often associated products are down.

Upvotes: 1

Related Questions