Skip to content

v 0.3.0-alpha

Pre-release
Pre-release
Compare
Choose a tag to compare
@GlennBullingham GlennBullingham released this 18 Sep 18:55
v0.3.0
1cafdbe

Release Summary

Release Date: 12th August 2020

Mayastor is currently in an early stage of development. This release is considered alpha software. As such, it is not intended for use in production.

New Features:

  • Automatic Replica Replacement - A Replica which has entered the "Faulted" state will be automatically replaced by the control plane, subject to there being available spare capacity in an appropriate Pool. The content of the replacement Replica will be 'rebuilt', i.e. made consistent with the other Replicas of that Nexus. This occurs without significant impact upon workload I/O.
  • NVMe-oF Support - The Mayastor CSI plugin can provision Persistent Volumes which are mounted on worker nodes over NVMe-oF, by specifying this transport type option in the corresponding StorageClass. (This functionality is made conditional upon kernel mode NVMe-oF initiator support, necessitating the use of kernel version 5.1 or greater).
  • Block Mode - Persistent Volumes now support raw block access: volumeMode: block
  • Direct Device Access - Mayastor Pools now support direct (host kernel independent) access to block device members, reducing abstraction overhead for high performance devices. The Mayastor Pool CRD accepts the previous form e.g. disks: ["/dev/sdx"], in addition to a new URI format e.g. disks: ["nvmf://10.240.0.4:8420/...etc"]

Enhancements:

  • Plugin Separation - The Mayastor CSI driver and plugins have been moved from within the Mayastor DaemonSet to their own, dedicated DaemonSet. This allows them to be independently scheduled by k8s and as a result, Mayastor Persistent Volumes can now be mounted and consumed on any worker node in the cluster, not just on those which are also running an instance of the Mayastor storage engine.
  • Rebuild Performance - The previous limitation on the maximum outstanding I/O permitted by a rebuild process has been raised, allowing faster Replica rebuilds but without significant impact on workload I/O to the affected Nexus. Each rebuild now supports up to 16 concurrent copy tasks, with the limitation on total rebuild I/O outstanding increased to 1MiB.
  • Durability - Mayastor pods persist their configuration state to local host storage each time they they process an API call which results in a state change. This allows a Mayastor pod to be correctly restarted in the event that it or its host node stop unexpectedly. At this time, such a restart will permit access to Replicas shared with other Nexus' but will not restore access to workload I/O for any PVC exported by the affected Mayastor process. Automated Failover and attendant continuous access to data whilst in a degraded state is the subject of ongoing development and will be introduced in future releases.
  • SPDK - Mayastor now uses SPDK version 20.07
  • Replica Error Threshold - A Nexus maintains counts of I/O error responses arising from its child bdevs ('Replicas'). In this version, if the error count held for a Replica exceeds a (default) value of 64, that Replica will be marked as "Faulted" by the Nexus and will be removed from the I/O path. In this release, the threshold value may only be changed by recompilation.
  • Flush - The Nexus now implements support for the flush command

Getting Started

The Mayastor quick deployment guide can be found here

Upgrade

Upgrades from previous versions of Mayastor are not supported. Any earlier release should be removed prior to installing this version. For testing purposes, it is preferable to install to a new, clean cluster.

Support

If you are having issues in setting up or upgrade, you can contact us via:

Assets 2
Loading