allow restricting Kube metadata to local node only #1440
Conversation
Codecov ReportAttention: Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## main #1440 +/- ##
===========================================
- Coverage 80.97% 63.22% -17.76%
===========================================
Files 149 145 -4
Lines 15255 15140 -115
===========================================
- Hits 12353 9572 -2781
- Misses 2293 4895 +2602
- Partials 609 673 +64
Flags with carried forward coverage won't be shown. Click here to find out more. ☔ View full report in Codecov by Sentry. |
mariomac
requested review from
grcevski,
marctc and
rafaelroquetto
as code owners
|
Hello @mariomac!
Please, if the current pull request addresses a bug fix, label it with the |
|
This PR must be merged before a backport PR will be created. |
There was a problem hiding this comment.
LGTM! I think the test failure might be related to aggressive expiration of metrics, or maybe confusion of what the name should be. I see an expiration of these values, where we have testserver-....
2024-12-12T16:40:35.912684579Z stdout F time=2024-12-12T16:40:35.912Z level=DEBUG msg="storing new metric label set" component=otel.Expirer type=*metric.int64Inst labelValues="[172.18.0.2 43558 request 10.244.0.5 10.244.0.0/16 testserver-858cdf668b-xpnlx 8080 egress my-kube testserver-858cdf668b-xpnlx default 172.18.0.2 test-kind-cluster-netolly-control-plane testserver Deployment Pod internal-pinger-net default 172.18.0.2 test-kind-cluster-netolly-control-plane internal-pinger-net Pod Pod 8080 10.244.0.9 10.244.0.0/16 internal-pinger-net TCP]"
However the next label is recorded with testserver as a name, i.e. no dash something:
2024-12-12T16:41:02.912577238Z stdout F time=2024-12-12T16:41:02.912Z level=DEBUG msg="storing new metric label set" component=otel.Expirer type=*metric.int64Inst labelValues="[172.18.0.2 43558 request 10.96.94.38 10.96.0.0/16 testserver 8080 egress my-kube testserver default testserver Service Service internal-pinger-net default 172.18.0.2 test-kind-cluster-netolly-control-plane internal-pinger-net Pod Pod 8080 10.244.0.9 10.244.0.0/16 internal-pinger-net TCP]"
All subsequent labels are without the dash...
pkg/kubecache/meta/informers_init.go
Outdated
| @@ -470,6 +521,10 @@ func (inf *Informers) ipInfoEventHandler(ctx context.Context) *cache.ResourceEve | |||
| return &cache.ResourceEventHandlerFuncs{ | |||
| AddFunc: func(obj interface{}) { | |||
| metrics.InformerNew() | |||
| em := obj.(*indexableEntity).EncodedMeta | |||
| for _, ip := range em.Ips { | |||
There was a problem hiding this comment.
I think we probably want to remove this debug print :)
|
Good catch @grcevski ! The duplicity of messages might be because testservice is captured either via a Pod (name with suffix) and a Service (name without suffix), but I'll anyway check that the IPs do not collide and that the expiration is properly set. |
Adds the
BEYLA_KUBE_META_RESTRICT_LOCAL_NODEconfiguration option that allows configuring the local informer to only watch the Kubernetes Pods from the local node. This will alleviate the memory load, especially during startup.