- Native Sidecar support – Enable Kubernetes Native Sidecar injection for FUSE deployments to resolve lifecycle management, startup order, and resource isolation issues.
- Dynamic PodMetadata configuration – Customize labels/annotations for ThinRuntime FUSE Pods during deployment.
- Optimized node scheduling – Disable non-essential node sync to accelerate cache engine scheduling.
- Realtime resource updates – Set default ThinRuntime Reconcile RateLimit to 0 for instant resource status reflection.
- Reliable mount detection – Reinforced postStart redirection logic to eliminate probabilistic mount-check failures.
- Extended storage support – Add 3FS and Curvine client compatibility through ThinRuntime.
- Privilege reduction – Strip non-essential administrative permissions from FUSE containers.
- Minimized control-plane permissions – Tighten ServiceAccount RBAC rules for core controllers by removing redundant authorizations.
- JindoFS 6.9.1 by default – higher read-bandwidth utilisation; file-prefetcher tag now auto-matches JindoFS image.
- Sidecar coexistence – multiple sidecar Pods can share one node without mount conflict; webhook allocates unique hostPath automatically.
- JuiceFSRuntime can disable metadata-sync & dataset-size calculation to cut control-plane load.
- ThinRuntimeProfile supports custom Pod lifecycle hooks.
- Fix JuiceFSRuntime fields unexpectedly overwritten after update.
- Fix missing tolerations when JindoRuntime sets multiple ones.
- Fix Fuse-Pod creation failure on very long Dataset names.
- Suppress excessive Runtime status updates.
- Correct volume generation when TieredStore contains no data.
- Prefetch Sidecar – accelerates model-serving cold start and improves bandwidth usage.
- JuiceFSRuntime Fuse seamless upgrade – idle Pods auto-cleaned via
cleanPolicy: onFuseChange. - JuiceFSRuntime preheat – exposes advanced warming parameters in DataLoad.
- Control-plane tunables – runtimeWorkers, kubeClient QPS/Burst, work-queue, env all user-configurable.
- Security: upgrade Helm → 3.17.3, Alpine → 3.19.1.
- ThinRuntime controller optimised to 35 instances/s.
- Fuse mount-check ConfigMap shared at namespace level to cut webhook CPU.
- Batch API detection added to lower DatasetController APIServer pressure.
- ThinRuntime
containerNetworksetting now takes effect. - JuiceFSRuntime no longer triggers spurious updates during sync.
- Enhanced Binding Efficiency in Kubernetes Clusters - Optimized ThinRuntimeController’s binding performance for large-scale Dataset deployments, reducing resource scheduling latency in Kubernetes clusters.
- Integration with Third-Party Storage via HostPath - Enabled Dataset to mount third-party storage systems (e.g., NFS, Ceph) through node-local HostPath volumes, extending compatibility with heterogeneous data sources.
- Distributed Fault Tolerance for High Availability - Isolated single-node FUSE failures from impacting Dataset-wide availability, ensuring cluster-level stability in distributed environments.
- Extended Support for Long Naming Conventions - Permitted creation of Datasets with lengthy names and namespaces, avoiding Kubernetes label length constraints (63-character limit).
- Webhook development-mode log panic removed.
- Added support to configure the
maxConcurrentReconcilesparameter for theThinRuntimeController, allowing for improved concurrency management. - Introduced support for resource configuration within the Fluid webhook, enabling more precise control over resource allocation.
- Added support for configuring
Burst(default 30) andQps(default 20) settings for client interactions with the APIServer through the Fluid Webhook, enhancing request handling capabilities. - Implemented
mountModeconfiguration to allow skipping the check for mount readiness, streamlining deployment processes.
- JindoRuntime master StatefulSet anti-affinity added for HA.
- Dynamic mount status – exposes real-time mount progress to users.
- Dataflow affinity – new operations can inherit node affinity from previous ones.
- JindoCache Prometheus – native data-plane metrics scraping.
- imagePullSecret – supported at Runtime level (Thin, Jindo, JuiceFS).
- CronDataLoad no longer fails when affinity / nodeSelector / schedulerName / imagePullSecrets present.
- Upgrade-safe – DataOps with same Dataset name keep working after Fluid version bump.
- DataProcess resources – requests/limits now configurable.
- Dataset-in-use info – injected into mutated app Pods for observability.
- Hierarchical imagePullSecrets – AlluxioRuntime supports repo secrets at multiple scopes.
- Sub-path dynamic mount – JuiceFS and others can mount
subPathon the fly.
- JindoRuntime syncs AK/SK secrets correctly.
- JuiceFS worker port conflict resolved.
- DataLoad cronJob template fixed for JindoCache.
- Data-flow affinity scheduling – operations stick to cache nodes where previous jobs ran.
- Dynamic mount-point updates – Dataset mount list editable without restart.
- Sidecar PVC acceleration – Fluid PVC can be accelerated in sidecar mode.
- JuiceFS community → v1.1.0, enterprise → v4.9.16
- JindoRuntime → v6.2.0
- JuiceFS attr-cache / entry-cache options can be overridden.
- JindoRuntime name no longer restricted by “jindofs” substring.
-
Configurable tiered data locality scheduling capability to optimize affinity between application pod and the data cache.
-
Support three kinds of data operations with different modes: once, onEvent, Cron, including: DataLoad, DataMigrate and DataProcess.
-
Support defining dataflow for data operations.
-
Support Python SDK for data scientists and operators to interact with Fluid control plane.
-
A new runtime for sharing in-memory immutable data: VineyardRuntime is supported in Fluid.
-
Security Hardening: Define a more restricted minimum necessary cluster role permissions for Fluid components, including eliminating all the secret-related and some create/update/delete privileges.
-
Enhance CSI Plugin and the FUSE Recovery feature for production usage.
-
Fluid drop all the Secret-related privileges to enhance its security.
-
Support cron for DataMigrate（#3224）
-
Support cron for DataLoad (#3309)
-
Make toleration configurable in helm charts (#3354)
-
Support Data Process CRD (#3345) (#3360) (#3367)
-
Support dataflow (#3384)
-
Data operations support resources (#3571)
-
Support ssh password free pod for parallel data migrate (#3645)
-
Add the vineyard runtime CRD definitions. (#3555)
-
Add the helm chart for vineyard runtime. (#3624)
-
Add the controller and RBAC yaml for vineyard runtime. (#3659)
-
Implement vineyard runtime engine and controller (#3649)
-
Add a replicas field to the vineyard Runtime and delete the svc suffix in the vineyard helm chart. (#3700)
-
Allow users to skip inject post start hook when using Fuse Sidecar (#3423)
-
Support worker of juicefs can be update by runtime (#3422)
-
Make juicefsruntime.spec.fuse configurale (#3539)
-
Feature: support symlink for NodePublish (#3440)
-
Support tiered locality scheduling for app pod (#3461) (#3489)
-
Allow setting kube client QPS and Burst (#3736)
-
Support configure rate limiter for controllers (#3758)
-
Check path existence and lock in NodeUnpublishVolume (#3284)
-
UmountDuplicate larger than the threshold (#3429)
-
Fix duplicate mount issue, To #48332603 (#3433)
-
Bugfix: remove umountDuplicate and add warning event (#3403)
-
Bugfix: fix csi plugin concurrency issue on FuseRecovery and NodeUnpublishVolume (#3448)
-
Enhancement: recover fuse according to multiple peer group options (#3491)
-
Support symlink for NodePublish (#3440)
-
Upgrade Docker base image from Alpine 3.17 to 3.18（#3254）
-
Use configmap helm driver instead of secrets (#3272)
-
Del secret get of juicefs runtime clusterrole (#3305)
-
Enhance: optimize secret-related code logic in ThinRuntime controller (#3323)
-
Support alluxioruntime with no secret and generate idempotent mount scripts (#3378)
-
Refactor: safe exec commands (#3699)
-
Support pipe command (#3692)
-
Use utils.command to replace exec.command (#3686)
-
Fix JuicefsRuntime: escape customized string before constructing commands (#3761)
-
Enhance: remove jindoruntime's fsGroup (#3632) (#3634) (#3635)
-
Support configure hostpid for runtime fuse. (#3755)
-
Update runtime info in dataset's status during binding (#3270)
-
Print error message when failing to helm install (#3271)
-
Fail fast with wrong kubelet rootdir (#3331)
-
Auto clean up crd-upgrade pod (#3500)
-
Change some fields from optional to required in CRD YAML (#3684)
-
Change enableServiceLinks from true to false (#3701)
-
Enhancement: auto discover runtime crds and dynamically enable/disable reconcilers (#3708)
-
Enhancement: add common labels to resources that managed by fluid (#3720)
-
Enhancement: Avoid heavy pod listing when worker statefulset has no replicas (#3771)
-
JuiceFSRuntime's default version upgrades to v1.1.0(community) and v4.9.16(enterprise)
-
JindoRuntime's default version upgrades to v6.2.0
-
Bugfix: Support batch/v1beta1 cronjobs for compatibility before Kubernetes v1.21 (#3280)
-
Bugfix: fix csi plugin loop mount bug (#3287)
-
Bugfix: reconcile thinruntime failed when dataset is deleted (#3300)
-
Bugfix: fix thinruntime stuck bug when deleting it (#3335)
-
Fix worker&fuse options & worker tiredstore (#3344)
-
Del metadata sync && del duplicate metrics (#3380)
-
Bugfix: clean up orphaned thinruntime resources (#3393)
-
Fix Alluxio master in HA mode start error. (#3658)
-
Pass AccessModes to thinruntime fuse container (#3696)
-
Fix fatal error: concurrent map writes for runtime controllers (#3757)
-
Fix incorrect conversion between integer types (#3688)
-
Refactor fuse sidecar mutation with mutator (#3477) (#3487) (#3492)
-
Auto infer engine implementation to support multi-engine Runtimes (#3672)
- Integrate Fluid with Kubeflow Pipline (#3694)
- Change matching pod requests mode of webhook from namespaceSelector to objectSelector
-
Add thinRuntime to simplify integration with third-party storage systems
-
Addon component for Fluid's open source CubeFS, NFS
-
Support for accessing data across namespaces
-
Support for subDataset
-
Native acceleration system EFCRuntime for distributed file systems NFS, GPFS
-
Support dataMigrate for data migration operations (currently only supported by JuiceFSRuntime)
-
Add customizable configuration for cache cleanup timeout and maximum retry times, Webhook timeout limits
-
Add Dataload configuration for ImagePullSecrets, node affinity
-
RBAC permission reduction
-
Upgrade to golang 1.18
-
Support for installing Fluid via Helm Repo
- Use data operation framework to construct data migrate, load, backup behaviors
-
[JuiceFS] FUSE pod scheduled failed because of conflict port
-
Fluid csi on rke2 k8s(1.22) use mount output empty,it caused app pod not work
-
AlluxioRuntime is upgrade from v2.8.2 to v2.9.1
-
JindoRuntime is upgraded from from 4.5.1 to 4.6.7
-
JuicefsRuntime is upgraded from v1.0.0 to v1.0.4
-
Lifecycle management of Serverless Job with fluid sidecar support
-
Enabling Runtime Controller on demand
-
Arm64 support with JuicefsRuntime
-
Container Network with short-circuit read support
-
Leader election support for Controllers and Webhook
-
Automatic CRD upgrader
-
Restrict Pod scheduling to dataset cache nodes
-
Tens of thousands of nodes support
-
Image pull secrets support
-
GCS support for Alluxio Runtime
- Port Allocation with different strategies: bitmap and random
-
AlluxioRuntime is upgrade from v2.7.2 to v2.8
-
JindoRuntime is upgraded from Jindo Engine to JindoFSX Engine, and the version is from 3.8 to 4.5.1
-
JuiceRuntime is upgraded from v0.11.0 to v1.0.0
-
Update Kubernetes v1.20.12 dependencies and use Go 1.16
-
Update CustomResourceDefinition to apiextensions.k8s.io/v1
-
Update MutatingWebhookConfiguration to admissionregistration.k8s.io/v1
-
Update CSIDriver to storage.k8s.io/v1
-
Support fuse sidecar auto injection for all the runtimes，it’s helpful for no CSI environment
-
Support fuse auto recovery and upgrade
-
Support lazy fuse mount mode
-
Support New cache runtime: JuiceFS
- Change cache worker deployment mode from DaemonSet to StatefulSet to use K8s Native schedule mechanism
-
Fix “Failed to delete dataload when target dataset is removed”
-
Fix “node-driver-registrar will not receive any volume umount in subdirectories of kubelet-dir”
-
Support dataset cache autoscaling and cronscaling
-
Add dataset mount point dynamically update feature
-
Enhance dataset cache aware Pod scheduling
-
Enhance HA support for cache Runtime
-
Support new cache Runtime：GooseFS
-
Support on-the-fly dataset cache scale out/in
-
Add Metadata backup and restore operation
-
Support Fuse global mode and toleration
-
Enhance Prometheus monitoring support for AlluxioRuntime
-
Support new Runtime：JindoFS
-
Support HDFS configuration
-
Warm up Dataset automatically before using it
-
Support managing 4,000,000 files
-
Support deploying multiple dataset in the same node
-
Support showing the HCFS Access Endpoint in Dataset status
-
Accelerate data access in Host-path mode in K8s
-
Accelerate data access in Persistent Volume mode in K8s
-
Support the underlay storage(NFS, Lustre) which is configured only with non-root
-
Make the Alluxio Runtime’s settings optimized by default
-
Add code coverage badge
-
Update chart v0.2.0
-
Update docs
-
Refactor the package name
-
Add 'Dataset', 'AlluxioRuntime' CRD for managing the data
-
Add PV and PVC for Posix interface
-
Documentation
-
Helm Chart