Qliksense consumer nodes CPU utilization spiking to 100 %

[Nilesh_999]

Hi All, 

Recently, we encountered a situation where the CPU utilization on the consumer nodes within the Qlik Sense Enterprise production environment reached 100% Wherein the Repository.exe on the consumer node is taking all the CPU. As a consequence, the Qlik Sense Hub became either inaccessible or significantly slowed down for all end users

The current interim solution involves rebooting the consumer nodes daily in the morning. However, we are actively seeking a permanent resolution to this issue. If anyone has encountered a similar problem within the Qlik Sense Enterprise environment and found a solution, your insights would be greatly appreciated.

Config- We have QS Nov 2022 installed with Cloud SQL instance Postgres as Repository.

[AdamBS]

We experienced similar at a large client.  Qlik support helped by using the Telemetry App - see help. It showed that we had a fload of user spiking the RAM, calls to the Repository got queued up and resulted in the Repository service spinning and consuming most CPU.  Part of our solution was to drop the App cache in memory duration from the default 8-hours.   This means our unused Apps drop out of RAM faster. Good luck... consider opening a case with Qlik Support, you might need their eyes on this one.

[Nilesh_999]

Hi Adam, 

Thanks for the response. 

When you are referring to time limit for dropping app cache from memory is it something inline with below article - 

https://community.qlik.com/t5/Official-Support-Articles/How-to-clear-the-cache-used-by-QlikView-Serv...

In our case we have one ODAG application that got accessed by thousands of users in early hours of the day and Consumers CPU exhausts during the same window. 

Thanks again for your response. 

[Nilesh_999]

Hi All, 

An update on this post , We identified one of the applications on our Qlik platform as particularly complex within our estate. To address this, we isolated the application to a different node by modifying the existing load balancing rules. Following this adjustment, the platform regained stability, particularly for the consumer nodes. Further investigation revealed that the application contained highly complex measures and expressions, with nested if-else statements up to ten levels deep

We are looking to further optimize the application. Any suggestions on how to measure its performance would be greatly appreciated.