ack-fluid is a component provided by the cloud-native AI suite to support Fluid, a distributed dataset orchestration and acceleration engine. ack-fluid serves data-intensive applications on the cloud with dataset abstraction, data access and acceleration, data orchestration, and data-aware scheduling. You can use ack-fluid to simplify data management during AI development and accelerate data access in AI model training and model inference scenarios. This topic introduces ack-fluid and describes the usage notes and release notes for ack-fluid.
Introduction
ack-fluid defines Kubernetes CustomResources (CRs), such as datasets and runtimes, to allow you to query the data stored in Kubernetes clusters and manage the lifecycle of the data by using Kubernetes-native methods.
ack-fluid accelerates access to application data stored in Kubernetes clusters through distributed cache system orchestration and management. Fluid supports various distributed cache systems serving as runtimes, including Alluxio, JuiceFS, JindoFS, and EFC. Fluid also supports the ThinRuntime, which allows you to access self-managed storage systems in a low-code way and reuse the key capabilities of Fluid, such as data orchestration and data access through the runtime platform.
ack-fluid provides standard data access APIs, which allow you to mount persistent volume claims (PVCs) to pods to access the data stored in containers. Fluid allows you to use the Fluid CSI plug-in to mount storage systems in standard Kubernetes environments and use the Fluid FUSE Sidecar plug-in to mount storage systems in serverless Kubernetes environments.
For more information about ack-fluid, such as the use scenarios and features, see Overview of Fluid.
Usage notes
ack-fluid supports only Container Service for Kubernetes (ACK) Pro clusters, ACK Serverless Pro clusters, and ACK Edge Pro clusters that run Kubernetes 1.18 or later. For more information about how to install and use ack-fluid, see Accelerate data access.
Release notes
April 2024
Version | Description | Release date | Impact |
v1.0.8 | The dynamic dataset mounting feature is supported by Fluid. The dataset mounting targets of all persistent volumes (PVs) and persistent volume claim (PVCs) referenced in a container can be automatically updated and dynamically mounted.
Note To use this feature, submit a ticket to apply to be added to a whitelist. | 2024-04-25 | No impact on workloads. |
February 2024
Version | Description | Release date | Impact |
v1.0.7 | JindoRuntime is updated to JindoCache and serves as the default caching engine. Affinity-based scheduling is supported by FUSE clients. The issue that the default read-only permissions on datasets are not propagated to ThinRuntime clients is fixed. Control plane components can be automatically enabled based on the types of cluster resources. The security of the Fluid components is enhanced.
| 2024-02-04 | No impact on workloads. |
December 2023
Version | Description | Release date | Impact |
v1.0.6 | Resource requests and limits can be specified when you perform data operations. Scheduling policies based on multiple node affinity levels can be customized. The JindoRuntime can be used to accelerate access to persistent volumes (PVs) and hostPath volumes.
| 2023-12-14 | No impact on workloads. |
November 2023
Version | Description | Release date | Impact |
v1.0.5 | The issue that mount points are propagated is fixed for the Fluid CSI plug-in. Scheduling policies based on multiple node affinity levels are supported. Master component metadata persistence is supported by the JindoRuntime. Dynamic update of FUSE components is supported by the JuiceFSRuntime. The scheduled reclaim feature is supported for data operations. Pods that are used for auto CustomResourceDefinition (CRD) updates can be reclaimed at a scheduled time. The issue that redundant AccessKey pairs and tokens are mounted is fixed for JindoFSx. The issue that Alluxio fails to parse sharedOptions is fixed.
| 2023-11-17 | No impact on workloads. |
September 2023
Version | Description | Release date | Impact |
v1.0.4 | Bugs related to the Fluid CSI plug-in are fixed. The FUSE Sidecar plug-in can be launched in asynchronous mode. JuiceFSRuntime configurations can be dynamically updated. ACK Lingjun clusters are supported.
| 2023-09-18 | No impact on workloads. |
August 2023
Version | Description | Release date | Impact |
v1.0.3 | The privileged mode is supported to mount storage systems with the FUSE Sidecar plug-in. The out-of-memory (OOM) issue is fixed for the JuiceFSRuntime controller. Issues related to FUSE auto recovery are fixed. The residual runtime resource issue that occurs after an update is fixed.
| 2023-08-18 | No impact on workloads. |
July 2023
Version | Description | Release date | Impact |
v1.0.2 | The Helm backend storage driver (HELM_DRIVER) is switched to the ConfigMap mode to limit the Role-Based Access Control (RBAC) permissions of Fluid control plane components. ThinRuntime can be interfaced with third-party storage features. Parameters can be stored in ConfigMaps. Local disks attached to elastic container instances can be used to cache data. Data preloading at a scheduled time with DataLoad is supported.
| 2023-07-18 | Residual Secret resources related to Helm may exist if the Helm backend storage driver is changed. If a cluster uses ThinRuntime to interface with third-party storage systems, newly created ThinRuntime clients may fail to run as normal due to compatibility issues after ack-fluid is updated to 1.0.2 or later. For more information, see Fluid.
Note In other scenarios, ThinRuntime clients are not affected by the updates of ack-fluid.
|
June 2023
Version | Description | Release date | Impact |
v1.0.1 | Anti-affinity and scheduling constraints are supported for ARM64-based nodes. Auto CRD update can be enabled or disabled. The residual HostPath mount target issue of the Fluid CSI plug-in is fixed.
The issue that ack-fluid is incompatible with Kubernetes versions earlier than 1.21 is fixed.
| 2023-06-19 | No impact on workloads. |
v1.0.0 | Scheduled DataMigrate jobs are supported by the JuiceFSRuntime. The FUSE sidecar volume injection issue of the EFCRuntime is fixed. The logic for updating dataset status is optimized to accelerate application pod startups. Detailed error information is displayed when runtime deployment failures occur.
| 2023-06-07 | No impact on workloads. |
May 2023
Version | Description | Release date | Impact |
v0.9.10 | | 2023-05-09 | No impact on workloads. |
April 2023
Version | Description | Release date | Impact |
v0.9.9 | Environment variables can be configured for the JindoRuntime. The degree of concurrency can be configured for the JuiceFSRuntime. The subdirectory quota can be configured for the JuiceFSRuntime. The issue that the DataMigrate and DataLoad paths are different is fixed. The association time of PVCs and PVs is optimized.
| 2023-04-25 | No impact on workloads. |
v0.9.8 | PVCs can be specified as data sources for DataMigrate jobs. Container images can be specified for the JindoRuntime. PVs and PVCs can be used to mount the subpaths of datasets. Fluid control plane component monitoring is supported.
| 2023-04-18 | No impact on workloads. |
v0.9.7 | | 2023-04-06 | |
March 2023
Version | Description | Release date | Impact |
v0.9.6 | DataMigrate jobs are supported. Node affinity can be configured for DataLoad jobs. The mounting logic of the Fluid CSI plug-in is optimized.
| 2023-03-22 | No impact on workloads. |