What's in this release

What's new

Virtualization

• Virtual machine high availability (HA)
  • Supports enabling HA for virtual machines that have SR-IOV passthrough NICs, or vGPUs mounted.
  • Supports triggering alarms on virtual machine HA success or failure.
• Supports enabling IO Threads for virtual machines via the command line.
• Supports batch selecting virtual machines and upgrading their Arcfra VMTools in one click.
• Supports configuring full-copy cloning speed limit for a cluster.
• Supports configuring the queue count for virtual NICs.

Networking

• A virtual distributed switch supports IGMP/MLD snooping. When enabled, the virtual distributed switch listens to IGMP/MLD packets and dynamically maintains multicast group members. This ensures precise forwarding of multicast traffic, reducing network load.
• The storage network supports enabling or disabling RDMA online.
• When the storage network enables RDMA, it supports automatic flow control configuration after restarting the node's Chunk service or performing an ifdown operation on the relevant physical network port.
• When RDMA is enabled for the storage network, you can select and bond multiple network ports across NICs within the associated virtual distributed switch.
• Supports configuring management network, DNS, hostname, and other settings for nodes via the TUI (Terminal User Interface) before cluster deployment.
• Supports configuring the vnet NIC queue length via the command line.

Operations and management

• Supports configuring the ACOS network service to either exclusively use CPU resources or share them with system services through the command line.
• Boost mode is enabled by default during ACOS (AVE) cluster installation and deployment.
• Adds an alert for excessive load of network port ingress and egress traffic.
• Adds monitoring and display of the status of member physical disks in hardware RAID.
• Supports detecting the Write Cache mode of hardware RAID 1 before deployment.
• Supports SR-IOV passthrough for Intel E810 NIC.

Improvements

Virtualization

• Optimizes the snapshot rebuild and virtual machine cloning mechanism so that the PCI addresses of virtual NICs remain unchanged after snapshot rebuild or cloning.
• Optimizes the cross-cluster hot migration mechanism in non-Boost mode to transfer only valid data during migration, reducing migration time.
• Optimizes the service leader election mechanism to no longer rely on the Meta Leader, improving virtual machine HA service handling in abnormal scenarios.

Block storage

• Reduces the data locality detection threshold from 6 hours of continuous access to 3 hours, accelerating data locality after virtual machine migration.
• Adds a backup throttling mechanism that automatically limits the backup rate based on cluster load when third-party backup vendors submit a large number of concurrent backup requests.
• Increases the number of data channels between nodes. When using balance-tcp mode for network port bonding on a VDS associated with the storage network, multiple network ports associated with the storage network can be fully utilized.

Networking

• Supports migrating the default VM network to another virtual distributed switch via the network-tool.
• Adds a pre-alert for high usage of the host conntrack table and a real-time alert for a full conntrack table, notifying users to handle situations with an excessive number of connections to avoid impacts on network communication within the host.

Operations and management

• Modifies UDEV rules to ensure the uniqueness of physical disk ID links.
• Adds an upgrade pre-check item to prevent clusters with expired licenses from being upgraded.
• Optimizes the process for enabling the SR-IOV feature on Mellanox NICs, reducing the number of times the host needs to restart.
• Optimizes the fluent-bit service status check logic to prevent cluster upgrade failures caused by failed checks.
• Optimizes the license validation logic for converting a standard cluster to an active-active cluster, allowing Standard Edition clusters to perform the active-active conversion.
• Lowers the default alert level of some low-impact cluster alerts to reduce operational noise.

Kernel

• Optimizes Open vSwitch performance.
• Upgrades the Intel ixgbe driver to version 6.2.5.

Resolved issues

Virtualization

• During virtual machine hot migration, if the source host restarted the libvirtd service due to maintenance or a failure, it might have caused the hot migration and subsequent virtual machine operations to fail. The issue has been resolved in this release.
• During virtual machine hot migration, continuously imposing high-level limits on vCPU performance might have caused virtual machine service abnormalities. The issue has been resolved in this release.
• Editing the NIC IP of Ubuntu or Debian virtual machines via Arcfra VMTools might have altered the virtual machine's existing DNS configuration. The issue has been resolved in this release.
• When Boost mode was not enabled, cross-cluster hot migration might have reported inaccurate progress. The issue has been resolved in this release.
• After configuring DNS servers for a virtual machine via Arcfra VMTools, a virtual machine reboot might have caused the gateway configuration not to take effect. The issue has been resolved in this release.
• When a virtual machine in a cluster failed to meet placement group policies and triggered an alert, even migrating that virtual machine to meet the requirements might not have automatically resolved that alert, or could trigger repeated false alerts. The issue has been resolved in this release.
• Configuring a static IP for a virtual machine via Arcfra VMTools might have failed if the operating system's nmcli tool version was outdated. The issue has been resolved in this release.
• In tiered storage mode, the logical disk space displayed in virtual machine monitoring was incorrect. The issue has been resolved in this release.
• After performing a one-click replica factor increase on a cluster, creating virtual machines via fast copy from templates cloned from the virtual machine within that cluster might have failed. The issue has been resolved in this release.
• When editing the cluster HA fault detection sensitivity, the operation might have failed because incorrect Leader node information was retrieved unexpectedly. The issue has been resolved in this release.

Block storage

• When the number of cores in a single CPU exceeded 192, the storage service could not start properly. The issue has been resolved in this release.
• In rare cases, the iSCSI abort command failed to clear all in-flight requests in time. The issue has been resolved in this release.
• When network failures occurred concurrently with other failures, there was a small chance of limited-scope data anomalies. The issue has been resolved in this release.
• The iSCSI PR command might have caused iSCSI link reconnection. The issue has been resolved in this release.
• When the iscsi inquiry command specified a particular length, the return value might have been abnormal. The issue has been resolved in this release.
• During snapshot creation, deletion, or rollback, as well as volume cloning and volume rebuild from snapshots, compounding failures might have caused data inconsistency. The issue has been resolved in this release.
• LSM metadata heat might have been inaccurate. The issue has been resolved in this release.
• A physical disk might have failed to be removed or remained in an isolated state due to an erroneous data migration command. The issue has been resolved in this release.
• RPC requests might have become stuck when the network was unstable. The issue has been resolved in this release.
• After canceling volume pinning on a pinned cache volume, capacity rebalancing might not have been triggered properly. The issue has been resolved in this release.
• When the load difference between nodes in a cluster was small, capacity rebalancing was triggered too frequently. The issue has been resolved in this release.
• The first UNMAP I/O after a volume snapshot or cloning operation was slow. The issue has been resolved in this release.
• After a volume had continuous I/O for 6 hours, a performance degradation spike might have occurred. The issue has been resolved in this release.
• After a network anomaly, a rare zbs-metad service exception might have caused I/O anomalies on some volumes. The issue has been resolved in this release.
• A failure in the zbs-chunkd service might have caused I/O interruptions. The issue has been resolved in this release.
• A highly rare data anomaly might occur after Meta member failures. The issue has been resolved in this release.
• During a node storage network failure, a segmentation fault in the spdk-master-thread caused the zbs-chunkd service to crash. The issue has been resolved in this release.
• A potential use-after-free issue was found in iSCSI. The issue has been resolved in this release.
• After adding a node, a false alert about excessive host storage space usage might have been triggered. The issue has been resolved in this release.
• The repeated failure and recovery of the storage network could cause the zbs-chunkd service to crash. The issue has been resolved in this release.
• Erasure code consistency repair failures prevented nodes from being removed. The issue has been resolved in this release.
• The zbs-chunkd service might have a rare chance of crashing during garbage collection. The issue has been resolved in this release.
• The zbs-metad service might have a rare chance of crashing when it exits. The issue has been resolved in this release.
• When a network anomaly occurred, untimely cleanup of iSCSI link I/O could in very rare cases cause data anomalies. The issue has been resolved in this release.
• The latency data were mistakenly displayed as zero during volume performance tracking. The issue has been resolved in this release.
• After the rack configuration was changed to a single rack, the rack space imbalance alert could not be automatically dismissed. The issue has been resolved in this release.
• Before heavy write traffic causes the performance tier load to exceed the medium threshold, the direct-write capacity tier logic is now triggered in advance, making backup and other workloads run more smoothly.
• Under conditions of sustained high pressure and frequent HDD failures on nodes, erasure-coded volume data might not have been properly sunk. The issue has been resolved in this release.
• Due to the Meta proxy not setting the timeout correctly, the Meta connection could not switch. The issue has been resolved in this release.
• In failure scenarios, the inspection service failed to detect generation inconsistencies in multi-replica data on the performance tier in advance. The issue has been resolved in this release.
• In erasure code mode, UNMAP requests might have been processed abnormally when a cluster failure occurred. The issue has been resolved in this release.
• The timemachine service failed to complete initialization on startup, causing license import failures. The issue has been resolved in this release.
• Data anomalies could occur in Boost mode when Windows virtual machines used virtio-win driver versions 0.1.190 through 0.1.221. The issue has been resolved in this release.

Networking

• When port bonding mode was set to balance-tcp, the port connectivity test might have reported abnormal results. The issue has been resolved in this release.
• Frequent network port abnormal alerts occurred due to instability in the health check service. The issue has been resolved in this release.
• When the storage network had RDMA enabled and used Linux Bond, after the cluster was upgraded to version 6.2.0 P4 and the host was restarted, one NIC in the host might have triggered an abnormal alert. The issue has been resolved in this release.
• After losing connection to zookeeper, the vipservice service's memory usage continued to increase, eventually affecting VM network connectivity. The issue has been resolved in this release.
• Due to a deadlock between two processes in the l2ping service, excessive CPU resources were consumed, preventing OVS from forwarding traffic and triggering network reachability alerts between the two hosts. The issue has been resolved in this release.
• The exit logic of the l2ping service has been optimzed in this release, preventing the service from remaining running while no longer functioning properly.

Operations and management

• The ntpd service might not have started due to time rollback after the systemd configuration file was generated or the host was restarted. The issue has been resolved in this release.
• After the cluster was expanded, subsequent upgrades failed if the upgrade tool version on the new nodes differed from that on the existing nodes. The issue has been resolved in this release.
• During port scanning, a surge in ntpm HTTP metrics might have caused monitoring and alert service failures; the collection of relevant metrics was disabled by default. The issue has been resolved in this release.
• Cached data in the browser might have caused unexpected behavior during cluster deployment. The issue has been resolved in this release.
• Some fans on the node had a significantly different speed compared to others. The issue has been resolved in this release.
• Cluster upgrade might have failed after rebuilding the witness node in an active-active cluster. The issue has been resolved in this release.
• Scheduled tasks failed due to root account expiration, leading to log rotation failures and disk full issues. The issue has been resolved in this release.
• The log rotation configuration was written to an incorrect path, causing the logs to fail to rotate. The issue has been resolved in this release.
• When a cluster NTP server used an unresolvable domain name, an exception alert might not have been triggered correctly. The issue has been resolved in this release.
• MongoDB log files might have consumed abnormal disk space due to system time jumps on a node. The issue has been resolved in this release.
• Stopping or restarting the network service under unexpected network port configurations might have cleared the DNS settings. The issue has been resolved in this release.
• Optimizes the handling logic for large files during log rotation to prevent rotation failures caused by excessively large individual log files.
• During cluster upgrade, a conflict between resource restarts and upgrade control could cause the upgrade to fail. The issue has been resolved in this release.
• The MongoDB CVE-2025-14847 vulnerability has been resolved in this release.

Kernel

• Guest OS-initiated SMI interrupts caused virtual machine crashes. The issue has been resolved in this release.
• Printing kernel scheduler logs caused the node to become unresponsive. The issue has been resolved in this release.
• The nouveau driver was incompatible with certain graphics cards, causing host crashes. The issue has been resolved in this release.