# Volcano v1.13.1

2025-12-23 · [Original release notes](https://github.com/volcano-sh/volcano/releases/tag/v1.13.1) · [Full summary](https://ratatosk.io/en/releases/volcano/v1.13.1)

Breaking changes 1 · Important fixes 6 · Other changes 7

## Breaking changes (1)

### Capacity plugin reclaim on any sufficient resource dimension

- Affects: capacity plugin setups with multi-dimensional resource requests (Inferred)

In the tested case, the other queue's pod `ns1/p18` is evicted, and the queue-overused metric is updated to reflect the new preemption decision.

Evidence: “case5: Can reclaim from other queues when allocated + req <= deserved on one dimension” https://github.com/volcano-sh/volcano/pull/4830

> * [release-1.13] fix: capacity plugin's preemptivefn logic by @hajnalmt in https://github.com/volcano-sh/volcano/pull/4830

## Important fixes (6)

### ReplicaSet updates now synchronize related PodGroup information

- Affects: ReplicaSets associated with PodGroups

Previously, updating Deployment or ReplicaSet information could leave related PodGroup fields stale when synchronization was needed. For example, changing `scheduling.volcano.sh/group-min-member` to `"2"` now updates the PodGroup's `minMember` to 2.

> * Cherry-pick PR #4786 to release-1.13: Fix replicaSet podgroup update synchronization by @jiahuat in https://github.com/volcano-sh/volcano/pull/4799

### Capacity plugin queue cloning panic fixed

- Affects: queue resources handled by the capacity plugin

When parent and child queues had inconsistent resource specifications, `realCapability` could remain nil and `queueAttr.Clone()` could panic. Initializing `realCapability` avoids that nil-pointer panic.

> * [Cherry-pick 1.13] fix: Initialize realCapability field in newQueueAttr by @dafu-wu in https://github.com/volcano-sh/volcano/pull/4694

### Scheduler PreBind lock blocking and timeout fixed

- Affects: scheduler setups with registered `PreBinder`s (Inferred)

The scheduling main loop could block and time out when a PreBind lock was not released. The PreBind execution and binding paths now use a snapshot of registered `PreBinder`s without holding the registry lock.

> * [cherry-pick 1.13]Scheduling main loop blocked and timeout due to un-released PreBind lock in Volcano by @guoqinwill in https://github.com/volcano-sh/volcano/pull/4700

### Scheduler nil-pointer panic from uninitialized metrics fixed

- Affects: scheduler installations started without `--enable-metrics=true`

Framework plugins could trigger a nil-pointer panic by accessing the uninitialized `k8smetrics.Goroutines`. That metric is now initialized regardless of whether metrics export is enabled.

> * [release-1.13] Fix scheduler panic when metrics are disabled by @Copilot in https://github.com/volcano-sh/volcano/pull/4770

### PodGroup creation skipped for annotated ReplicaSet pods

- Affects: ReplicaSet pods with a non-empty `scheduling.KubeGroupNameAnnotationKey` annotation

The handler returns instead of continuing toward PodGroup creation or update for these pods.

> * [release-1.13] fix: replicaset KubeGroupNameAnnotation handling by @hajnalmt in https://github.com/volcano-sh/volcano/pull/4826

### Volcano PodGroup parsing clears stale task-member entries

- Affects: PodGroups using `Spec.MinTaskMember` (Inferred)

When task members in a PodGroup changed, task validity checks could fail during scheduling because removed entries remained in the parsed map. The map now reflects the current entries, including when it becomes empty.

> * [release-1.13] Fix: Changes to task members in a PodGroup caused task validity checks to fail during scheduling by @ouyangshengjia in https://github.com/volcano-sh/volcano/pull/4852

## Other changes (7)

- Ray end-to-end test default image updated to `rayproject/ray:2.49.0`
- Ray plugin e2e test prunes unused images on all nodes
- PodGroup controller worker startup loop updated
- Node-cache removal skipped for terminated tasks
- Null `jobID` during scheduler-cache deletion logged instead of treated as an error
- Scheduler task details and cache diagnostic wording updated
- Disk-space cleanup added to E2E workflows

---

Summaries are produced by AI, so they can be wrong, and they are provided without warranty.
