The module is enabled by default in the following bundles: Default
, Managed
.
The module is disabled by default in the Minimal
bundle.
The module is configured using the ModuleConfig custom resource named extended-monitoring
(learn more about setting up Deckhouse…).
Example of the ModuleConfig/extended-monitoring
resource for configuring the module:
apiVersion: deckhouse.io/v1alpha1
kind: ModuleConfig
metadata:
name: extended-monitoring
spec:
version: 1
enabled: true
settings: # <-- Module parameters from the "Parameters" section below.
Parameters
Schema version: 1
- certificatesobject
Settings for monitoring the certificates in the Kubernetes cluster.
- certificates.exporterEnabledboolean
Enables certExporter.
Default:
false
- certificates.exporterEnabledboolean
- eventsobject
Settings for monitoring the events in the Kubernetes cluster.
- events.exporterEnabledboolean
Enables eventsExporter.
Default:
false
- events.severityLevelstring
Whether to expose only crutial events.
Default:
"OnlyWarnings"
Allowed values:
All
,OnlyWarnings
- events.exporterEnabledboolean
- imageAvailabilityobject
Settings for monitoring the availability of images in the cluster.
- imageAvailability.exporterEnabledboolean
Enables imageAvailabilityExporter.
Default:
true
- imageAvailability.ignoredImagesarray of strings
A list of images to ignore when checking the presence in the registry, e.g.,
alpine:3.12
orquay.io/test/test:v1.1
.Example:
ignoredImages: - alpine:3.10 - alpine:3.2
- imageAvailability.skipRegistryCertVerificationboolean
Whether to skip the verification of the container registry certificate.
Default:
false
- imageAvailability.exporterEnabledboolean
- nodeSelectorobject
The same as in the pods’
spec.nodeSelector
parameter in Kubernetes.If the parameter is omitted or
false
, it will be determined automatically. - tolerationsarray of objects
The same as in the pods’
spec.tolerations
parameter in Kubernetes;If the parameter is omitted or
false
, it will be determined automatically.- tolerations.effectstring
- tolerations.keystring
- tolerations.operatorstring
- tolerations.tolerationSecondsinteger
- tolerations.valuestring
How to use extended-monitoring-exporter
Attach the extended-monitoring.flant.com/enabled
annotation to the Namespace to enable the export of extended monitoring metrics. You can do it by:
- adding the appropriate helm-chart to the project (recommended method);
- adding it to
.gitlab-ci.yml
(kubectl patch/create); - attaching it manually (
kubectl annotate namespace my-app-production extended-monitoring.flant.com/enabled=""
). - configuring via namespace-configurator module.
Any of the methods above would result in the emergence of the default metrics (+ any custom metrics with the threshold.extended-monitoring.flant.com/
prefix) for all supported Kubernetes objects in the target Namespace. Note that monitoring and standard annotations are enabled automatically for a number of non-namespaced Kubernetes objects described below.
You can also add custom annotations with the specified value to threshold.extended-monitoring.flant.com/something
Kubernetes objects, e.g., kubectl annotate pod test threshold.extended-monitoring.flant.com/disk-inodes-warning-threshold=30
.
In this case, the annotation value will replace the default one.
You can disable monitoring on a per-object basis by adding the extended-monitoring.flant.com/enabled=false
annotation to it. Thus, the default annotations will also be disabled (as well as annotation-based alerts).
Standard annotations and supported Kubernetes objects
Below is the list of annotations used in Prometheus Rules and their default values.
Caution! All annotations:
- Start with a
threshold.extended-monitoring.flant.com/
prefix; - Have an integer value (except for the
extended-monitoring.flant.com/enabled
Namespace-annotation — its value can be omitted). The specified value defines the alert threshold.
Non-namespaced Kubernetes objects
Do not require a Namespace annotation and are enabled by default.
Node
Annotation | Type | Default value |
---|---|---|
disk-bytes-warning | int (percent) | 70 |
disk-bytes-critical | int (percent) | 80 |
disk-inodes-warning | int (percent) | 85 |
disk-inodes-critical | int (percent) | 90 |
load-average-per-core-warning | int | 3 |
load-average-per-core-critical | int | 10 |
Caution! These annotations do not apply to
imagefs
(/var/lib/docker
by default) andnodefs
(/var/lib/kubelet
by default) volumes. The thresholds for these volumes are configured completely automatically according to the kubelet’s eviction thresholds. The default values are available here; for more info, see the exporter.
Namespaced Kubernetes objects
Pod
Annotation | Type | Default value |
---|---|---|
disk-bytes-warning | int (percent) | 85 |
disk-bytes-critical | int (percent) | 95 |
disk-inodes-warning | int (percent) | 85 |
disk-inodes-critical | int (percent) | 90 |
container-throttling-warning | int (percent) | 25 |
container-throttling-critical | int (percent) | 50 |
container-cores-throttling-warning | int (cores) | |
container-cores-throttling-critical | int (cores) |
Ingress
Annotation | Type | Default value |
---|---|---|
5xx-warning | int (percent) | 10 |
5xx-critical | int (percent) | 20 |
Deployment
Annotation | Type | Default value |
---|---|---|
replicas-not-ready | int (count) | 0 |
The threshold implies the number of unavailable replicas in addition to maxUnavailable. This threshold will be triggered if the number of unavailable replicas is greater than maxUnavailable
by the amount specified. Suppose replicas-not-ready
is 0. In this case, the threshold will be triggered if the number of unavailable replicas is greater than maxUnavailable
. If replicas-not-ready
is set to 1, then the threshold will be triggered if the number of unavailable replicas is greater than maxUnavailable
+ 1. This way, you can fine-tune this parameter for specific Deployments (that may be unavailable) in the Namespace with the extended monitoring enabled to avoid getting excessive alerts.
Statefulset
Annotation | Type | Default value |
---|---|---|
replicas-not-ready | int (count) | 0 |
The threshold implies the number of unavailable replicas in addition to maxUnavailable (see the comments on Deployment).
DaemonSet
Annotation | Type | Default value |
---|---|---|
replicas-not-ready | int (count) | 0 |
The threshold implies the number of unavailable replicas in addition to maxUnavailable (see the comments on Deployment).
CronJob
Note that only the deactivation using the extended-monitoring.flant.com/enabled=false
annotation is supported.
How does it work?
The module exports specific Kubernetes object annotations to Prometheus. It allows you to improve Prometheus rules by adding the thresholds for triggering alerts. Using metrics that this module exports, you can, e.g., replace the “magic” constants in rules.
Before:
max by (namespace, pod, container) (
(
rate(container_cpu_cfs_throttled_periods_total[5m])
/
rate(container_cpu_cfs_periods_total[5m])
)
> 0.85
)
After:
max by (namespace, pod, container) (
(
rate(container_cpu_cfs_throttled_periods_total[5m])
/
rate(container_cpu_cfs_periods_total[5m])
)
> on (namespace, pod) group_left
max by (namespace, pod) (extended_monitoring_pod_threshold{threshold="container-throttling-critical"}) / 100
)