Installation Notes

Auto Reboot

A new kernel is going to be installed and a reboot of the system is necessary. The output of the progress bar may stop before rebooting.

Note: Please wait, until all software packages are installed. The reboot of the system will then be initialized automatically and the server is going to be available after a few minutes.

Install Update within Cluster

Please read the following release notes carefully, before continuing. When installing this update on a Collax V-Cube+ cluster the following steps have to be performed:

Note: If starting the update from 6.1.8 to 6.1.10 no down time of the running virtual machines or applications is required.

Procedure

1. If the nodes have the version 6.1.8 or prior: Please read the appropriate notes for that release including the documented procedure first.
2. The software packages can be downloaded whithin a running Cluster to prepare the update and minimize the planned downtime. Go to the local administration of each node and access the form System Operation → Software → Systemupdate. Then click Get Package List. This will download the list of available update packages. The successful update of the package list is indicated by the message Done!. Click Get Packages. If the packages are downloaded successfully then the message Done! will be displayed.
3. Put one node into standby mode. Go to Status/Maintenance → Cluster Nodes and execute the action Standby. Please wait until the cluster resources are stopped or the VMs are migrated.
4. Now, put this node into maintenance mode. Go to the dialog Status/Maintenance → Cluster Services and stop the service HA Cluster for this node.
5. Click Install within the dialog Systembetrieb → Software → Systemupdate to install all update packages. The end of this process is indicated by the message Done!.
6. If a new kernel is going to be installed the system will be rebooted automatically after installing the update. An appropriate note will be shown. Please wait for the reboot in that case.
7. Start the service HA Cluster on the node that has been updated. Please wait until the service ist started and information about Cluster Resources are available before you proceed with the next step.
8. Once the service HA Cluster has been started resume the node by using the action Set active within the dialog Status/maintenance → Cluster Nodes .
9. Afterwards check the status of the eSAN-disk synchronization in the form Cluster Administration → Virtual Hard Disks. Do not continue with the next step until the column Status shows 100% or OK for all eSAN disks.
10. Please repeat this procedure from step 2 on the next node that needs to be updated.

Install Update from Version 6.0.10 within Cluster

Please read the following release notes carefully, before continuing. Note that when installing this update on a Collax Inter-V cluster the following steps have to be performed:

Procedure

1. This version is deprecated and does not receive further support.
2. Please shut down all virtual machines within the cluster management gui before continuing with the following steps. Please consider to disable the shutdown event tracker for Windows operating systems before shutting down the virtual machine.
3. Put all nodes into standby mode. Go to Status/Maintenance → Cluster Nodes and execute the action Standby. Please wait until the cluster resources are stopped.
4. Put one node into maintenance mode. Go to the dialog Status/Maintenance → Cluster Services and stop the service HA Cluster for this node.
5. Go to the local administration of this node and access the form System Operation → Software → Systemupdate. Then click Get Package List. This will download the list of available update packages. The successful update of the package list is indicated by the message Done!.
6. Click Get Packages. If the packages are downloaded successfully then the message Done! will be displayed.
7. Click Install. This action installs all update packages. The end of this process is indicated by the message Done!.
8. A new kernel is going to be installed. The system will be rebooted automatically after installing the update. An appropriate note will be shown if the update process is completed.
9. After the reboot start the service HA Cluster of the node again: Go to the cluster administration form Status/Maintenance → Cluster services. Please wait until information about Cluster Resources are available.
10. Repeat the steps 3 to 8 for the next cluster node.
11. Once all cluster nodes are updated resume them one by one by using the action Set active within the dialog Status/maintenance → Cluster Nodes .
12. Next check the status of the eSAN disk synchronization in the form Cluster Administration → Virtual Hard Disks. Do not continue with the next step until the column Status shows 100% or OK for all eSAN disks.
13. If all nodes within the cluster are updated and active the virtual machines can be started again.

Install Update on a Collax V-Cube

To install this update on a Collax V-Cube without a Cluster interconnection please follow these steps:

Procedure

1. Please read this document before proceeding to the next step.
2. In the administration interface go to System → System Operation → Software → System Update and then click Get Package List. This will then download the listed update packages. The successful update of the package list is indicated by the message “Done!”.
3. Click Get Packages to download the software packages.
4. Click Install. This action installs the update. The end of this process is indicated by the message Done!.
5. A new kernel is going to be installed. The system will be rebooted automatically after installing the update. An appropriate note will be shown once the update process is completed.
6. If the server is updated and booted the virtual machines can be started again.

New in this Version

Virtualization: New virtual BIOS and Windows Virtio Driver for virtual Machines

The BIOS for the virtual machines (SeaBIOS) and the Virtio drivers for Windows will be renewed with this update. To receive optimum performance and stability, use the SeaBIOS 0.6.2, the new Virtio hard disk drivers and new Virtio network drivers. To update perform the following this procedure:

• Put Windows driver ISO in the VM
• Update driver with device manager inside the Windows VM. Click Update Driver Software… at the Virtio device under Storage controllers and Network adapters
• Choose the driver within the directory for the used operating system
• Shutdown VM after updating the drivers
• Set BIOS to Seabios 0.6.2 and activate the configuration
• Boot VM

Virtualization: Virtual Floppy Disk with Virtio Drivers

From this version on a floppy disk drive can be selected for all virtual machines. The floppy disk drive also contains new Virtio drivers for Windows similar to the CD-ROM driver. The floppy drive is suitable for new installations of Windows VMs which can not load paravirtual hard disk drivers via the CD-ROM drive.

V-Cube+: Configure Max Time for Shutdown

If the cluster resource for a virtual machine stops, an option can be set to shut down the VM cleanly. Because the time required to shut down a virtual guest varies, this time can be set for the Cluster management. The default is set to six minutes. If a VM takes longer to power off, this value can be changed accordingly.

V-Cube+: Create snapshots with less disk space left than required

Virtual machine Snapshots enable you to return to a specific state of a vm at a certain point in time. For example to a point in time before a failed update or a patch was applied. Snapshots in Collax virtualization create a differential image as separate file. This file can only become as big as VM disk itself. With this update it is possible to create a snapshot whose maximum size exceeds the size of the storage available. A notice will be issued. It is therefore possible to size the cluster share smaller than the maximum space required, assuming that the administrator monitors the available storage space carefully.

System Management: Monitoring of Disk Space Usage within Cluster Share

The cluster share is used to share different files within the cluster. During normal cluster operation the space available on the hard disk of the cluster share is utilized. If the disk space of the share is completely used up this can lead to failure of a virtual machine that run in snapshot To avoid this, from this update, a check for the active monitoring is going to be implemented. At 80% of disk space usage a warning is dispatched by e-mail, while a ‘CRITICAL’ state message is sent if more than 90% is used.

System Management: Installation via USB-Stick

Installing Collax Server from USB flash drive is possible from this update on. The How-to can be downloaded from www.collax.com in every Server Product section within the menu Howto/White Paper.

Hardware: Additional hardware support

This update brings support for two additional 10 gigabit network interface cards: Intel 10 Gigabit XF Dual Port Server (glass fiber, EXPX9502AFXSR) and Intel Ethernet Server Adapter X520-T2 (copper, E10G42BT).

Misc: Important System Components

The following important system components will be installed with this update.

• libvirt 0.9.1-ak59
• kvm 88-ak32
• pacemaker 1.0.11-ak25
• openais 1.1.4-ak21
• heimdal 1.4-ak83
• kernel 2.6.32.39-ak77

Issues Fixed in this Version

Virtualization: I/O Scheduler

I/O scheduling is a term used to describe the method operating systems decide the order I/O submission to storage volumes. So far the method used was CFQ (Completely Fair Queing). During high I/O load this occasionally resulted in processes hanging. From this update on the I/O scheduler Deadline is used. One feature of this scheduler is a property that sets time limits for all I/O operations. This prevents starvation of requests.

V-Cube+: Wrong Start Order of DNS Zones

The reboot of a cluster node triggered the DNS zones to be loaded before the LDAP directory was started. Because the DNS zones are located in the LDAP the DNS server was not able to resolve hosts from the master zone after a reboot. With this update the DNS zones are read after the LDAP server is started, the name resolution of hosts (e.g. mail server) is working after a reboot.

V-Cube+: Resources stopped after Backup or Activation of Configuration

If virtual machines are backed up, or when settings were activated in the cluster, cluster resources then could be stopped accidently. This behavior is corrected by using an improved parser for attributes in the CIB (Cluster Information Base). Resources keep their original state if a backup or an activation is executed.

V-Cube+: Restart of VM after Snapshot

Occasionally if a snapshot of a virtual machine was created that VM was restarted by the cluster manager. From this update every snapshot creation triggers an update of the resource state of the VM before it is set to * managed* within the cluster. This results in the VM not being restarted after a snapshot.

V-Cube+: VNC Console shows content of a deleted Virtual Machine

Under certain circumstances if a virtual machine in a cluster was deleted and then a new virtual machine added and started, the VNC console displayed content of the deleted VM Also the login on the VNC display failed with the message “Password wrong”. With this update all corresponding VNC Proxy processes are purged after the virtual machine is deleted. The correct screen is now displayed by the VNC console.

V-Cube+: Live Migration of a Virtual Machine

    Live migrations of virtual machines could fail in different
    situations. For a live migration network sockets are
    opened. If one of these sockets was already allocated, the live migration failed
    and a fencing request was requested. With this update a range of TCP sockets
    are reseved exclusively for live migrations within the kernel.
    So every migration can be executed successfully.




    If a virtual machine had been migrated, it could happen,
    that

libvirtd linked virtual network interfaces to physical network devices in wrong order. Then the migration failed. With this update the virtualization back-end is improved, so that the existing network interfaces of the VMs are disregarded during a migration and so the migration is executed correctly.

If it was not possible to connect to the VNC Socket during a live migration, the migration of the VM failed. The message in the Syslog was * libvirtd: 00:50:40.267: error: qemudNetworkIfaceConnect:1624: Failed to add tap interface cvm0.0 to bridge br0: Device or resource busy *. With this update this issue is solved.

If the live migration lasted longer than 900 seconds the VM was stopped. From this version on this time out is going to be calculated dynamically depending on the size of the RAM of the virtual machine. With it the Live-Migration is executed without stopping the VM.

V-Cube+: Starting a VM failed after creating a snapshot

For the action Start the name of the selected virtual machine was determined for the subsequent start. The start of the VM could fail if a snapshot and a VM with a similar name existed (Eg: Win and Win2k8). As of this update it is determined by an unique ID which VM should be started.

V-Cube+: Creation of offline snapshots while a Cluster Node is in stand-by

The action Create Snapshot was not possible within the administration GUI of a cluster node that had been switched into the standby mode. This is corrected from this update on.

V-Cube+: Merge snapshot if hardware of virtual machine had changed

The snapshot operation commit could fail, if the hardware of the virtual machine had changed since generating the present snapshots. This is resolved with this update.

V-Cube+: Complete synchronization of Embedded SAN after first Node-Standby

If a cluster node was put into standby mode for the first time and activated again within the cluster afterwards, all Embedded SAN disk drives were completely synchronized again. This behavior is corrected with this update. Once the Embedded SAN disk drives have been synchronized completely, only the differences need to be transferred in the future.

V-Cube+: Keywords in names of Embedded SAN Disks

When adding an Embedded SAN (DRBD) hard drive, there has been no restriction for the definition of a name. For example, if disk was set as name, a synchronization of this DRBD disk was not possible since disk is processed as a keyword of the DRBD backend. Therefore the added DRBD disk could not be used in the cluster. With this update a check is executed when adding a DRBD disk, so the functionality of the disks within the disks within the cluster can be guaranteed.

GUI: Firefox 4 and Cluster GUI

With the version 4 of the browser Mozilla Firefox 4 it was not possible to administrate the the cluster GUI correctly. This is repaired with this update.

System Management: Notification via Email of Active Monitoring

Active monitoring scripts check Cluster services and send appropriate notifications via email. If no group permission have been set it was possible that alerts were not sent to the administrator. This update corrects this. Notifications are sent to members of the group with access to the active monitoring or to the administrator.