soft-pod-memory-evicter

soft-pod-memory-evicter

A Kubernetes Controller that proactively evicts Pods approaching their memory limits, enabling graceful shutdowns before OOM (Out of Memory) termination occurs.

Supported Versions

The controller has been tested and built with the following components:

• Kubernetes >= 1.32.1
• Metric Server >= 0.7.2

Installation

helm repo add maxlaverse https://maxlaverse.github.io/helm-charts/
helm repo update
helm install soft-pod-memory-evicter maxlaverse/soft-pod-memory-evicter

The controller can be configured by setting Helm values under extraArgs: [] with any of the options listed in the Usage section below. For a complete list of configurable Helm values, please refer to the values.yaml file.

Usage

NAME:
   soft-pod-memory-evicter - Gracefully evict Pods before they get OOM killed

USAGE:
   soft-pod-memory-evicter [global options] command [command options]

COMMANDS:
   help, h  Shows a list of commands or help for one command

GLOBAL OPTIONS:
   --dry-run                                              Output additional debug lines (default: false)
   --pod-selector value                                   Evict only Pods matching this label selector
   --eviction-pause value                                 Pause duration between evictions (default: 5m0s)
   --memory-usage-check-interval value                    Interval at which the Pod metrics are checked (default: 3m0s)
   --memory-usage-threshold value                         Memory usage eviction threshold (0-100) (default: 95)
   --channel-queue-size value                             Size of the queue for pod eviction (default: 100)
   --ignore-namespace value [ --ignore-namespace value ]  Do not evict Pods from this namespace. Can be used multiple times
   --enable-metrics value                                 Expose Prometheus metrics endpoint (default: false)
   --metrics-bind-address value                           Bind address for the Prometheus exporter (default: :9288)
   --loglevel value, -v value                             Log Level (default: 0)
   --help, -h                                             show help

Metrics

Enable the exporter with --enable-metrics and set --metrics-bind-address (for example, :9288). The endpoint is served at /metrics and currently publishes a single counter:

• soft_pod_memory_evicter_evicted_pods_total{affected_namespace="<ns>",affected_app_kubernetes_io_name="<name>",affected_app_kubernetes_io_instance="<instance>"} — counts evictions per workload (namespace + app.kubernetes.io/name + app.kubernetes.io/instance).

You might also want to add the following annotations into Helm values to let Prometheus know about the exporter:

podAnnotations:
  "prometheus.io/scrape": "true"
  "prometheus.io/port": "9288"

Pod-specific configurations

Pods can be configured individually using annotations to override the default behavior:

• soft-pod-memory-evicter.laverse.net/memory-usage-threshold: Specifies the memory usage threshold (0-100) at which the Pod should be evicted. This value overrides the global threshold set by the --memory-usage-threshold flag.

Example:

apiVersion: v1
kind: Pod
metadata:
  annotations:
    soft-pod-memory-evicter.laverse.net/memory-usage-threshold: "80"

License

Distributed under the Apache License. See LICENSE for more information.