You are viewing documentation for Kubernetes version: v1.19

Kubernetes v1.19 documentation is no longer actively maintained. The version you are currently viewing is a static snapshot. For up-to-date documentation, see the latest version.

Configuring kubelet Garbage Collection

Garbage collection is a helpful function of kubelet that will clean up unused images and unused containers. Kubelet will perform garbage collection for containers every minute and garbage collection for images every five minutes.

External garbage collection tools are not recommended as these tools can potentially break the behavior of kubelet by removing containers expected to exist.

Image Collection

Kubernetes manages lifecycle of all images through imageManager, with the cooperation of cadvisor.

The policy for garbage collecting images takes two factors into consideration: HighThresholdPercent and LowThresholdPercent. Disk usage above the high threshold will trigger garbage collection. The garbage collection will delete least recently used images until the low threshold has been met.

Container Collection

The policy for garbage collecting containers considers three user-defined variables. MinAge is the minimum age at which a container can be garbage collected. MaxPerPodContainer is the maximum number of dead containers every single pod (UID, container name) pair is allowed to have. MaxContainers is the maximum number of total dead containers. These variables can be individually disabled by setting MinAge to zero and setting MaxPerPodContainer and MaxContainers respectively to less than zero.

Kubelet will act on containers that are unidentified, deleted, or outside of the boundaries set by the previously mentioned flags. The oldest containers will generally be removed first. MaxPerPodContainer and MaxContainer may potentially conflict with each other in situations where retaining the maximum number of containers per pod (MaxPerPodContainer) would go outside the allowable range of global dead containers (MaxContainers). MaxPerPodContainer would be adjusted in this situation: A worst case scenario would be to downgrade MaxPerPodContainer to 1 and evict the oldest containers. Additionally, containers owned by pods that have been deleted are removed once they are older than MinAge.

Containers that are not managed by kubelet are not subject to container garbage collection.

User Configuration

Users can adjust the following thresholds to tune image garbage collection with the following kubelet flags :

  1. image-gc-high-threshold, the percent of disk usage which triggers image garbage collection. Default is 85%.
  2. image-gc-low-threshold, the percent of disk usage to which image garbage collection attempts to free. Default is 80%.

We also allow users to customize garbage collection policy through the following kubelet flags:

  1. minimum-container-ttl-duration, minimum age for a finished container before it is garbage collected. Default is 0 minute, which means every finished container will be garbage collected.
  2. maximum-dead-containers-per-container, maximum number of old instances to be retained per container. Default is 1.
  3. maximum-dead-containers, maximum number of old instances of containers to retain globally. Default is -1, which means there is no global limit.

Containers can potentially be garbage collected before their usefulness has expired. These containers can contain logs and other data that can be useful for troubleshooting. A sufficiently large value for maximum-dead-containers-per-container is highly recommended to allow at least 1 dead container to be retained per expected container. A larger value for maximum-dead-containers is also recommended for a similar reason. See this issue for more details.

Deprecation

Some kubelet Garbage Collection features in this doc will be replaced by kubelet eviction in the future.

Including:

Existing FlagNew FlagRationale
--image-gc-high-threshold--eviction-hard or --eviction-softexisting eviction signals can trigger image garbage collection
--image-gc-low-threshold--eviction-minimum-reclaimeviction reclaims achieve the same behavior
--maximum-dead-containersdeprecated once old logs are stored outside of container's context
--maximum-dead-containers-per-containerdeprecated once old logs are stored outside of container's context
--minimum-container-ttl-durationdeprecated once old logs are stored outside of container's context
--low-diskspace-threshold-mb--eviction-hard or eviction-softeviction generalizes disk thresholds to other resources
--outofdisk-transition-frequency--eviction-pressure-transition-periodeviction generalizes disk pressure transition to other resources

What's next

See Configuring Out Of Resource Handling for more details.

Last modified May 30, 2020 at 3:10 PM PST: add en pages (ecc27bbbe7)