> For the complete documentation index, see [llms.txt](https://documentation.alluxio.io/k8s-operator-en/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://documentation.alluxio.io/k8s-operator-en/release-notes/operator-3-7-5.md). # Alluxio Kubernetes Operator 3.7.5 Release Notes ## Summary ### New Features * **Write-cache dashboard** ships with the operator chart and is discovered by the bundled Grafana without extra configuration. It opens on an operations view — Cluster Overview, Risk Signals, and Aggregate Performance — with further operations and engineering rows collapsed below. * **Job monitoring** gains panels on the Job row of the bundled Alluxio dashboard for two previously unobservable areas: the indexing phase of `job load` (progress of listing-index rebuilds by status, in-flight index subtasks against the concurrency quota, pending index directories per job, and worker-side index build duration and results) and free-job outcomes (files and bytes freed versus failed per second). A failing free job and a stalled one are now distinguishable. The indexing panels show no data on clusters that never submit index-enabled loads. * **Automatic cleanup of stale FUSE attachments** lets you enable a per-node audit in the CSI node plugin that reaps leftover FUSE pods and `FuseAttachment` state whose consuming pods are gone, including stale target-path records and idle attachments left behind by a missed unpublish. These no longer require manual cleanup. Set `csi.nodePlugin.nodeserver.auditInterval` to a Go duration such as `10m` to enable it; it is disabled by default. * **Named worker groups** let you set an optional `name` on each entry of `workerGroups` so the group's StatefulSet, ConfigMap, and worker-group pod label derive from a stable name (`-worker-`) rather than the entry's position in the list. Inserting, removing, or reordering entries then no longer renames another group's resources. A name must match `^[a-z0-9]([-a-z0-9]*[a-z0-9])?$` and be at most 24 characters. Groups left unnamed keep the existing index-based naming (`-worker-group-`). When adopting names on an existing cluster, note that renaming a group recreates its resources, while giving a group the name `group-` matching its current index keeps resource names identical, so only the changed pod label triggers a single rolling restart. ### Bug Fixes * **UFS monitoring** no longer splits each instance into two identically labeled lines on the per-instance panels when the server reports UFS data access by source. The cluster read-throughput panel also gains `UFS (cold read)` and `UFS (load)` lines showing the distributed-load share of UFS read bandwidth; both hide on servers that do not report the source breakdown. * **Distributed load retry monitoring** fixes the `Retry DLQ Depth` panel drifting permanently upward and reporting a large depth with no running jobs. The panel now reads a per-job value that appears while a load job runs and disappears when it ends. Coordinators older than AI-3.8-15.1.12 show no data on this panel. * **Request success-rate monitoring** fixes the `Request Call - Success Rate` panel reading 20–80% on a healthy cluster, which happened because expected negative lookups — a `stat` on a missing path, or a `HeadObject` on a missing key — were counted as failures. Not-found outcomes now count as success, and a series with no recorded failures plots 100% instead of disappearing from the panel. The adjacent `Request Call - Failures` and `FUSE Request Failure` panels use the matching definition, so they can no longer contradict the success rate side by side. * **FDB Operator tolerations** can be configured as a list without triggering a Helm coalesce warning, because the chart default for `fdb-operator.tolerations` is now an empty list. ## Breaking Changes No breaking changes in this release. ## Upgrade Notes No migration steps are required for this release.