NewYour coding agent can read the release notes before it upgrades.Set up the MCP server →
Go modules · #1938 by repository stars
Last release 1 months ago
23 Aug 2026
Ships fairly regularly
a new release about every 5 weeks
Some releases are documented
notes for 10 of 28 stable releases
Nothing withdrawn
no release was ever pulled
3 years old
45 releases · first in 2023
One column per quarter.
There are two big new features in this release, both of which are not yet enabled in the default configuration. They goal is to promote both of them a
There are two big new features in this release, both of which are not yet enabled in the default configuration. They goal is to promote both of them as a sensible default once they have been validated at a larger scale. If you do test them and have any issues or other feedback, please do not hesitate to open an issue.
Capacity Eviction: prevents node out-of-memory scenarios by ensuring the node has enough resources to restore a container before allowing it to happen. If the node does not have enough capacity, it will evict the requesting pod and forward requests to the new one as soon as it's available. This feature can be toggled on using a kustomize component.
# config/production/kustomization.yaml
components:
[...]
- ../capacity-requestReuseport Activator (Experimental, requires Linux 6.6 or later): This enables a completely new activator which handles traffic to the container purely in the kernel and requires no userspace proxying. See the architecture document for more details on how this works. This feature can be toggled on using a kustomize component.
# config/production/kustomization.yaml
components:
[...]
- ../reuseport-activatorApart from that this new release contains several bug fixes and dependency upgrades like CRIU v4.2.1 and Go 1.26.
Full Changelog: v0.12.1...v0.13.0
fix: ignore "wait: no child processes" error when attaching by @ctrox in #187
Full Changelog: v0.12.0...v0.12.1
The main change for this release is that the eBPF attachment is now being done by the shim again instead of the node daemonset. This makes the shim le
The main change for this release is that the eBPF attachment is now being done by the shim again instead of the node daemonset. This makes the shim less dependent on the node component. It was initially delegated to the node component since loading eBPF programs caused significant memory allocations in the shim but with this new implementation this was addressed by doing the attachment in a separate short-lived shim process. For upgrading to this release, the order does not matter since the node component can still take over the attachment if older shim versions are detected that don't support it.
Full Changelog: v0.11.3...v0.12.0
ci: pin action versions by @ctrox in #176
Full Changelog: v0.11.2...v0.11.3
fix: metrics socket connection leak by @ctrox in #167
Full Changelog: v0.11.1...v0.11.2
fix: close bpf maps/netns on stop by @ctrox in #164
added annotations for configuring proxy timeouts
added annotations for configuring proxy timeouts
zeropod.ctrox.dev/connect-timeout: "5s"
zeropod.ctrox.dev/proxy-timeout: "5s"much improved checkpoint/restore failure handling
local checkpoint images are now deleted on restore, not just on pod deletion
Full Changelog: v0.10.0...v0.11.0
cleanup of old migration resources: migration resources are now owned by the source/destination pod, resulting in automatic cleanup by the API Server
-auto-gc-migrations=false on the manager.Full Changelog: v0.9.2...v0.10.0
Nothing published for this version
feat: add flag to ignore localhost in socket tracker by @ctrox in #129
While validating the latest release on different systems, an issue was identified where the socket_tracker map would be emptied by the kernel sporadic
While validating the latest release on different systems, an issue was identified where the socket_tracker map would be emptied by the kernel sporadically. This includes a hotfix for this issue by changing the BPF map type.
If you are coming from release v0.8.0 or earlier and make use of scale to zero, ensure to re-schedule all pods using zeropod after updating the zeropod-node DaemonSet. This is due to an architectural change in the socket tracker. Pods running pre-v0.9.0 zeropod shims might not be able to detect the last TCP connection and will simply scale down after the configured duration is up.
Full Changelog: v0.9.0...v0.9.1
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Nothing published for this version
Your coding agent can read these notes before it upgrades. Set up the MCP server →