[
https://issues.apache.org/jira/browse/KYLIN-4834?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
yoonsung.lee updated KYLIN-4834:
--------------------------------
Description:
Hi. I'm yoonsung.lee developing & operating the statistical storage using Kylin
in LINE Advertising dept.
h1. Current Situation
I understand there is only one active leader job server instance in a Kylin
cluster.
The leader job server is selected by Zookeeper if there are multiple active job
servers in a Kylin cluster.
This is my pain point.
When my leader job server has a fault or shutdown, I want to change the leader
job server automatically and the DNS server route the traffic to the changed
one.
In this case, I should know which one is the leader job server as soon as
possible the leader(active) has changed.
For this one, I try to watch the path of Zookeeper in which the Kylin server
maintains the server list in a cluster.
But, this approach has a fault in essence. It's not an implementation
controlled by Kylin.
h1. Question
Is my approach appropriate for the Kylin cluster?
If not, is there any better idea for fault tolerance of job servers with auto
DNS or LoadBalancer mapping?
h1. Suggestion
So, I suggest this idea.
Any server in a Kylin cluster forwards the API request to the leader job server
in the case that the request should be done by the job server and pass valid
authentication. If then, I just attach all of the servers to my LoadBalancer
without caring about which one is the leader job server.
* Could this feature be harmful or have side-effects?
was:
Hi. I'm yoonsung.lee developing & operating the statistical storage using Kylin
in LINE Advertising dept.
h1. Current Situation
I understand there is only one active(leader) job server instance in a Kylin
cluster.
The active job server is selected by Zookeeper if there are multiple job
servers in a Kylin cluster.
This is my pain point.
When my leader job server has a fault or shutdown, I want to change the leader
job server automatically and the DNS server route the traffic to the changed
one.
In this case, I should know which one is the leader job server as soon as
possible the leader(active) has changed.
For this one, I try to watch the path of zookeeper in which the Kylin server
maintain the server list in a cluster.
But, this approach has a fault in essence. It's not an implementation
controlled by Kylin.
h1. Question
Is my approach appropriate for the Kylin cluster?
If not, is there any better idea for fault tolerance of job servers with auto
DNS or LoadBalancer mapping?
h1. Suggestion
So, I suggest this idea.
Any server in a Kylin cluster forwards the API request to the leader job server
in the case that the request should be done by the job server and pass valid
authentication. If then, I just attach all of the servers to my LoadBalancer
without caring about which one is the leader job server.
* Could this feature be harmful or have side-effects?
> Auto forward the job API request to the available leader job server
> -------------------------------------------------------------------
>
> Key: KYLIN-4834
> URL: https://issues.apache.org/jira/browse/KYLIN-4834
> Project: Kylin
> Issue Type: Wish
> Components: Others
> Reporter: yoonsung.lee
> Priority: Minor
>
> Hi. I'm yoonsung.lee developing & operating the statistical storage using
> Kylin in LINE Advertising dept.
> h1. Current Situation
> I understand there is only one active leader job server instance in a Kylin
> cluster.
> The leader job server is selected by Zookeeper if there are multiple active
> job servers in a Kylin cluster.
> This is my pain point.
> When my leader job server has a fault or shutdown, I want to change the
> leader job server automatically and the DNS server route the traffic to the
> changed one.
> In this case, I should know which one is the leader job server as soon as
> possible the leader(active) has changed.
> For this one, I try to watch the path of Zookeeper in which the Kylin server
> maintains the server list in a cluster.
> But, this approach has a fault in essence. It's not an implementation
> controlled by Kylin.
> h1. Question
> Is my approach appropriate for the Kylin cluster?
> If not, is there any better idea for fault tolerance of job servers with auto
> DNS or LoadBalancer mapping?
> h1. Suggestion
> So, I suggest this idea.
> Any server in a Kylin cluster forwards the API request to the leader job
> server in the case that the request should be done by the job server and pass
> valid authentication. If then, I just attach all of the servers to my
> LoadBalancer without caring about which one is the leader job server.
> * Could this feature be harmful or have side-effects?
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
