haproxy requests excessive resources by default

Description

User story:

As a DBA when deploying a PXC DBaaS cluster haproxy requests are excessive by default

And as it can't be changed, I would recommend lowering the default to

Requests: cpu: 500m memory: 500M

So that this will help avoid wasting resources and reduce k8s cluster specs as by now 6G are needed only for haproxy to run small databases in k8s.

Acceptance criteria

  • haproxy is default choice for PXC cluster

  • haproxy default requests CPU 500m

  • haproxy default requests memory is 500M

  • Documentation updated to reflect these changes

Details

This is what we get by default in the haproxy section

Requests: cpu: 1 memory: 2G

Recommended defaults would be

Requests: cpu: 500m memory: 500M

 

How to test

None

How to document

None

Smart Checklist

Activity

Show:

Former user August 12, 2022 at 6:44 AM

  • To start a fully-working 3 node XtraDB cluster, consisting of sets of 3x HAProxy, 3x PXC and 6x PMM Client containers, you will need at least 9 vCPU available for minikube. (1x vCPU for HAProxy and PXC and 0.5vCPU for each pmm-client containers).

Docs should be updated https://docs.percona.com/percona-monitoring-and-management/setting-up/server/dbaas.html#start-pmm-server-and-activate-a-dbaas-feature

Andrei Minkin August 2, 2022 at 9:58 AM

Changed default values for both ProxySQL and for HAProxy. FB is here

Andrei Minkin August 2, 2022 at 8:12 AM

I discussed with yesterday about default values for proxysql and he told me that we should keep values the same for both proxysql and haproxy 

Roma Novikov July 28, 2022 at 2:20 PM

, this  global problem. imo, it's ok  to get it "hard coded" as usual now and we need better solution for all PMM 

Andrei Minkin July 28, 2022 at 11:46 AM

I think that we need additional endpoints for PXC and PSMDB to get default values for different cluster sizes because right now we have at least two options to set resources

  1. On grafana frontend

  2. By using fillDefaults method on cluster creation (Works for PXC cluster)

  3. Also, there is an option to set it on dbaas-controller side 

We can use default values provided by either operators definitions or by specifying or own

Done

Details

Assignee

Reporter

Priority

Labels

Needs QA

Yes

Needs Doc

Yes

Planned Version/s

Fix versions

Story Points

Smart Checklist

Created May 13, 2022 at 2:52 PM
Updated March 6, 2024 at 1:13 AM
Resolved October 5, 2022 at 7:22 AM
 

Flag notifications