[jira] [Comment Edited] (HAWQ-1436) Implement RPS High availability on HAWQ

["Lili Ma (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/HAWQ-1436?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15984355#comment-15984355 ] 

Lili Ma edited comment on HAWQ-1436 at 4/26/17 8:20 AM:
--------------------------------------------------------

I suggest we do solution 1) to simply the implementation as the first step.

[~Paul Guo]
Since RPS will only be called by master and a query usually only raises several RPS request, there won't be a lot of requests to RPS. So I think load balancer may be over-design. 

For services which support a lot of concurrent requests, a proxy server at the front of multiple web Service is an ideal design, since it will be convenient for both HA and load balance.

[~lei_chang]
It's a good suggestion for auto-discover RPS failure and auto-restart, but I think RPS is a little different from Resource Manager process. We need add special processing for it since it's a Web service. Do you have any suggestion on the detailed implementation?


was (Author: lilima):
I suggest we do solution 1) to simply the implementation as the first step.

Since RPS will only be called by master and a query usually only raises several RPS request, there won't be a lot of requests to RPS. So I think load balancer may be over-design. 

For services which support a lot of concurrent requests, a proxy server at the front of multiple web Service is an ideal design, since it will be convenient for both HA and load balance.


> Implement RPS High availability on HAWQ
> ---------------------------------------
>
>                 Key: HAWQ-1436
>                 URL: https://issues.apache.org/jira/browse/HAWQ-1436
>             Project: Apache HAWQ
>          Issue Type: Sub-task
>          Components: Security
>            Reporter: Hongxu Ma
>            Assignee: Hongxu Ma
>             Fix For: backlog
>
>         Attachments: RPSHADesign_v0.1.pdf
>
>
> Once Ranger is configured, HAWQ will rely on RPS to connect to Ranger. A single point RPS may influence the robustness of HAWQ. 
> Thus We need to investigate and design out the way to implement RPS High availability. 



--
This message was sent by Atlassian JIRA
(v6.3.15#6346)