Closed tormodmacleod closed 2 months ago
Hi @tormodmacleod
Thanks for creating this issue. We're going to evaluate your proposal and your PR during the following days.
Because we're very close to Christmas we may give you feedback later than usual.
This issue won't be actioned.
Description
all I want is a single infrastructure agent pod running. i don't need
ksm
orcontrolPlane
running so i disable them as below.however, i'm still left with an
agent
and akubelet
container running in akubelet
pod on each node because this has been implemented as aDaemonSet
i just want a single pod running an infrastructure agent container only to get the length of some redis lists
i already have eks covered in my infrastructure monitoring layer. i need to be able to implement application monitoring
Acceptance Criteria
kubelet
pods to be created as part of aDeployment
rather than aDaemonSet
- togglekubelet
container NOT to be created within the pod - toggleDescribe Alternatives
a new chart which implements only the infrastructure agent?
Dependencies
it might be worth updating the following documents to make it clear to others how you might go about retrieving the lengths of some of your redis keys. i found this was a little unclear if your redis cluster is managed (aws elasticache) rather than running on the cluster
https://docs.newrelic.com/docs/kubernetes-pixie/kubernetes-integration/advanced-configuration/monitor-services/monitor-services-running-kubernetes/ https://docs.newrelic.com/docs/kubernetes-pixie/kubernetes-integration/get-started/introduction-kubernetes-integration/ https://docs.newrelic.com/docs/infrastructure/host-integrations/host-integrations-list/redis/redis-integration/
Additional context
i am using new relic to monitor at different layers of the stack here
infrastructure
at the infrastructure layer, i have (amongst other things) an eks cluster. i've installed the
nri-bundle
chart from https://helm-charts.newrelic.com using the terraform helm provider. this includes thenewrelic-infrastructure
chart with thekubelet
,ksm
et cetera. this is sneding all the telemetry for my eks cluster back to newrelic. i'm getting all of the information about my eks cluster that i want from it and i'm very happyapplication
on my eks cluster i'm running a laravel app. this uses redis to store (amongst other things) it's queue data. i'm using aws elasticache for redis. i've installed the
newrelic-infrastructure
chart from https://newrelic.github.io/nri-kubernetes using the terraform helm provider with the following configurationthis is working and sending the lengths of my queues to newrelic. i now can see if i have a problem with one of my queue workers or the queue is filling up for some other reason.
why this matters
these two layers shouldn't know about each other i could try to implement this all at the infrastructure layer. but then i'd need to tell my infrastructure layer about which redis keys to monitor - you can't use wildcards. whenever i create a new tenant, i'll want to add the redis keys for their queues to the list of keys i'm monitoring. when i remove the tenant i'll no longer want to monitor their queues. if i implement this at the infrastructure layer, i'll need to have terraform call the api of the thing that manages my tenants to get a list of the queues for all of my tenants. the terraform that provisions my infrastructure layer is run manually rather than being event driven. it this could be changed but due to it's nature i don't particularly want this running off events. it shouldn't know anything about tenants or queues
i only need one pod even if i did decide to have my infrastructure layer configuring the redis integration to monitor my queues, the
agent
runs within thekubelet
pod as part of aDaemonset
. this means that as my cluster scales out and more and more nodes are added i'll have more and more pods. these pods will all be retrieving the same information from redis and sending it to newrelic. this puts unnecessary load on my redis cluster. it also sends duplicate information to newrelic increasing my costs without providing any additional valueEstimates
i'd say S = 1-3 days
For Maintainers Only or Hero Triaging this bug
Suggested Priority (P1,P2,P3,P4,P5): Suggested T-Shirt size (S, M, L, XL, Unknown):