topic Re: Qlik Sense using Round Robin ? in Management & Governance

Qlik Sense using Round Robin ?

sharanh — Sat, 16 Nov 2024 06:35:09 GMT

Hi All,

Qlik Multi node setup I have- Central node and RIM node (also fail over candidate)

Both nodes added in Load balancing of Central and RIM node virtual proxies.

My question is 'HOW Qlik uses both servers to assign and run scheduler tasks and user logins?'

I understand qlik uses round robin algorithm for this purpose and the server used at time of user login will be used for subsequent sessions.

I need more clarify on this.

Example-

I have 4 tasks - Task A, Task B, Task C and Task D. All scheduler to run at different time everyday. How will qlik use the two nodes in running the 4 tasks?

The basic aim is to know how is CPU utilized and memory used of two nodes in shared persistence.

Thanks in advance

Re: Qlik Sense using Round Robin ?

Daniele_Purrone — Fri, 15 Feb 2019 09:35:47 GMT

Hi!

"the server used at time of user login will be used for subsequent session": this is only for virtual proxy to engine connections, not for scheduled tasks.

So, scheduler load balancing is round-robin only. However, all the tasks in the same chain are currently sent to the same engine, and this is getting investigated for improvements.

We also have a feature request about introducing dynamic load balancing: https://support.qlik.com/articles/000061586

I hope this helps!

Daniele

Re: Qlik Sense using Round Robin ?

Levi_Turner — Sun, 17 Feb 2019 15:05:39 GMT

So the Scheduler architecture is well described on https://community.qlik.com/t5/Qlik-Sense-Deployment-Management/Qliksense-Reload-background-activites/m-p/64375/highlight/true#M1128

The key part is that the Scheduler does not do "round robin". It does a check to see which Slave is least busy and allocated based on the response.

Does that clarify things?

Re: Qlik Sense using Round Robin ?

sharanh — Mon, 18 Feb 2019 01:05:26 GMT

Thanks Daniele and Leviturner.

I understand that the two different approaches that you both have mentioned utilize the slave or RIM node in different ways and CPU utilization is kind of same for main and RIM nodes with no server over utilized.

Somehow, the approaches have different mechanisms to assign tasks to slave.

It will be great if we could please have a statement /approach to which all of us can mutually agree on?