Servers are deployed to Kubernetes using the official Docker image for CB 7.0.0.
Fresh installation of 2 nodes: after adding a default bucket with 1 replica, nodes can not be rebalanced and bucket + nodes are pending:
Hi @zoltan.zvara, Thanks for trying out Couchbase Server 7.0 beta!
Can you tell us how you have set up your Couchbase Cluster on Kubernetes? Can you send you the configuration and also the machine configuration?
Thanks!
Hi @anil!
I attached the Helm Chart. helm-chart.zip (2.2 KB)
It is deployed as follows:
---
apiVersion: helm.fluxcd.io/v1
kind: HelmRelease
metadata:
name: couchbase-default-development
namespace: development
annotations:
fluxcd.io/automated: "false"
filter.fluxcd.io/chart-image: glob:*
spec:
releaseName: couchbase-default-development
chart:
git: ssh://redacted:redacted/redacted/redacted.git
path: charts/couchbase
ref: master
values:
image:
repository: couchbase
tag: community-7.0.0-beta
persistence:
size: 10Gi
class: local
replicas: 2
Then through VPN, the Couchbase cluster is initialized and nodes are connected on:
- Intel on-prem 4-node blade
- Running CoreOS latest
- Kubernetes 1.17.4
- As you can see it is deployed using Flux, but I think it is irrelevant.
- Kubernetes was deployed using Kubespray, not an experimental cluster, running 100s of stuff pretty well, I don’t think there is anything wrong with the networking.
It works with:
- Couchbase 6.0 Community
- Couchbase 6.5 Community
- Couchbase 6.6 Community
It does not work with:
- Couchbase 6.5 Enterprise
- Couchbase 7.0 Beta Community
“Does not work” means that a bucket with > 1 replica copies are never gets initialized.
Let me know if you need more.
Thanks!
This is a hectic problem and is due to having long Couchbase server hostnames. Couchbase will fail to initialize replication protocol between servers for replicas when the server hostnames are longer than 200 characters, including protocol and port (?).