Using Google Container Engine with GCP free tier

Question

Is it possible to use Google Container Engine with Google Cloud free tier?
(I mean the "Always Free" usage limit, not the $300 free credit)

The docs for GKE says:

The basic cluster is free but each node is charged at standard Compute Engine pricing

But the Compute Engine also have a free instance. Is it possible to use them together?

Answer 1

Unfortuniately, this is no longer a correct answer, as GKE no longer (as of December 2020 if not earlier) supports f1-micro instances for node pools as they do not have sufficient memory (as alluded to in my original answer below where enabling stackdrier would make the cluster unstable). Therefore, it is not possible to run a GKE cluster fully within the free tier.

Previously, this was possible. See explanation below.

---

Yes, you can use GKE with the free tier. GKE only charges for the underlying compute engine resources, which are directly billed by compute engine. (Note that after June 6, 2020, the free tier only includes one free GKE zonal cluster -- not an unlimited number of clusters).

GKE will likely require you to run 3 free f1-micro instances concurrently to get the cluster to a minimum size, but as long as the cluster is in one of the free regions and the total usage in a month is under the total number of hours per month it will still be free (that is, you can run 3 f1-micros for a bit under 250 hours and still be in the free tier). Make sure to shut off your instances when you are not using them. See more at https://cloud.google.com/free/docs/gcp-free-tier#always-free-usage-limits (especially the notes about the limit being time rather than instance count).

You may also want to ensure that the persistant disks are not kept around while the cluster isn't running, as the free tier only allows 3 10GB disks over the course of the month.

If you happen to run over the usage, you will only be charged for the usage beyond the free tier.

Of course, this all assumes that f1-micro instances are suitable to your use case. They are quite limited, and once GKE is on them, they have almost nothing remaining in terms of RAM: https://cloud.google.com/kubernetes-engine/docs/concepts/cluster-architecture#memory_cpu

Finally, it has been my experience that setting up the stackdriver support when you create the cluster if you only have micro instances can cause the cluster to struggle greatly -- the stackdriver monitoring alone (or with even minimal other applications) start to cause the nodes to be throttled and time out.