[
https://issues.apache.org/jira/browse/HBASE-27005?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17579421#comment-17579421
]
Duo Zhang commented on HBASE-27005:
-----------------------------------
With the help [~xiegang112], I was able to start a zookeeper with k3s on
jenkins-hbase23. [~xiegang112] is a HDFS contributer in the past and later
moved on to build cloud native platforms.
The related docker files and k8s configs can be found here:
https://github.com/Apache9/hbase-k3s
Most of the problems are env related. For example, we have a kubectl command
installed so k3s just skips installing kubectl but the existing kubectl is too
old. And also, newer version of k3s requires containerd instead of docker, but
we installed a containerd which is not compatible with k3s too. Not sure why we
install these things but never use them...
About k3s, see here https://k3s.io/
I will start HDFS and then HBase on jenkins23 too, with k3s. But if we want to
use this cluster to run ITBLL, there are still some technical problems. Our
ClusterManager requires start/stop/restart/kill a region server with host port
pair, but in the K8s world, this is kinda difficult. You can easily reduce the
region server number, for example, from 5 to 4, just reduce the replicas, but
you are not easy to control which one will be killed. And also, there is no way
to start a region server on a given 'host', we do not know the ip address
before launching the pod...
So maybe we need to redefine the ClusterManager, to make it easier with the K8s
concepts...
> move nightly cluster tests to run on k8s
> ----------------------------------------
>
> Key: HBASE-27005
> URL: https://issues.apache.org/jira/browse/HBASE-27005
> Project: HBase
> Issue Type: Umbrella
> Components: integration tests
> Reporter: Sean Busbey
> Priority: Major
>
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
