[
https://issues.apache.org/jira/browse/HBASE-27005?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17600075#comment-17600075
]
Duo Zhang commented on HBASE-27005:
-----------------------------------
Discussed with [~xiegang112] several times about how to implement the
ClusterManager, if we want to do this in native K8s way, there is no choice
other than implementing our own operator or controller, where we could do
something like 'shutdown a region server on a specific host and reduce the
replica number at the same time'. And it is also a pain that, when starting a
new region server, it is not likely that we could know its host before actually
starting, or we could, but it also requires special overlay support and also, a
special oeprator...
So for simplify the task here, I prefer we just start the main hbase/hdfs
process by something like a python daemon, and daemon could respond to some rpc
requests to start or stop the process. In this way the pod is still there so
the host will not change and no eviction/scheduling is required. This is not
very 'cloud native' but I think it is the easier way to fit our requirements
for now...
Thanks.
> 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)
