Rancher 2.0 and ETCD question


#1

I read that Rancher 2.0 is using MySQL instead of etcd.
Are you using a distributed MySQL system to avoid a single point of failure?
+
ETCD is a distributed design but MySQL is not really a distributed database?


#2

I believe that the MySQL implementation will follow the 1.x method, and the user is required to take care of any required MySQL master/slave/clustering.


#3

So this will be the most weak point of Rancher2.0.
But:
Integration of Longhorn inside Rancher2.0 is foreseen and Longhorn is using ETCD.
It would be logical to use etcd in place of MySQL.


#4

Hi everyone. Rancher plans to support using an external etcd cluster with our embedded K8s, MySQL is simply an option for those who don’t need the features of etcd.


#5

When will this external etcd be coming as part of rancher 2?

Replacing MySQL with ETCD is an important feature.


#6

I don’t think everyone is on the same page here… Kubernetes HAS to use etcd, it is the required datastore for the kubernetes api server. However the Rancher Server needs a relational database which is mysql. These are two different pieces to the system. etcd and mysql are not interchangeable. One is a distributed/concurrent key value store, later is a relational database.


#7

This is all outdated. 2.0 preview2+ (alpha12) does not have a MySQL or any other relational DB. There is an embedded etcd for the management stack, and each cluster registered to it has its own etcd like any other k8s installation. Future builds will likely add options to

  • Use an external etcd for management instead of the built-in one

and

  • (Re)Use the management etcd for clusters instead of requiring each cluster to have a separate one (“Embedded” vs “Dedicated”)

#8

Hey there,

in which version can we expect this feature? “Use an external etcd for management instead of the built-in one”

regards
Zoltan


#9

BTW, etcd is the weak link now :slight_smile:

I had k8s admin course and k8s production grade workshop, it was explained that etcd was the weak link but nobody said anything about this blog post or how robust to make etcd:


(full stack: https://monzo.com/blog/2016/09/19/building-a-modern-bank-backend/ )

The upshot is you need 9 etcds in 3 zones (or “data centers”).

here’s another take on etcd:

https://crewjam.com/etcd-aws/

Both of these approaches wrap etcd in some way, but the crewjam example just uses 3 zones (not 3 in each zone).

Wonder what Rancher 2.0 does… soon will find out.

Regards
Rich