OpenSolaris

You are not signed in. Sign in or register.

OpenSolaris Project: Cluster Agent: OpenSolaris xVM

View the leaders for this project
Project Observers

Endorsing communities

HA Clusters
Xen

Overview

The aim of this project is to provide a failover agent for OpenSolaris x86 Virtual Manager (xVM) and Logical Domains (LDoms) guest domains as part of the open-source code base for Solaris Cluster, namely Open High Availability Cluster (OHAC). Within this project the term xVM refers to the OpenSolaris xVM project and LDoms. It is intended that an OHAC failover agent for xVM guest domains will perform the following:

  • Manage the start/stop and restart of xVM guest domains within an OHAC environment.
  • Failover a xVM guest domain between OHAC nodes.
  • Allow for strong positive and negative affinities between xVM guest domains across OHAC nodes.
  • Allow for different failover techniques between OHAC nodes, i.e. Stop/Failover/Start or Live Migration.

Project Goals

  • Provide high availability for xVM guest domains in the event of an OHAC node failure. For example, if a node fails, the OHAC failover agent for xVM guest domains will relocate the guest domain onto another OHAC node. Note that OHAC supports upto 16 nodes, campus clusters (upto 40km separation), metro clusters (upto 400km separation) and Geographic Edition for Disaster Recovery (unlimited distance), Dual partition software upgrade / Live upgrade, to list just a few features of OHAC.
  • Provide the ability to failover a xVM guest domain across OHAC nodes using live migration. In this scenario, both nodes are available and it is preferred to live migrate a xVM guest domain, using a cluster file system, between OHAC nodes.
  • Provide the ability to favour one xVM guest domain over another. For example, assuming a 2-node cluster, under normal operations both nodes could be running several xVM guest domains. Some guest domains maybe more important than others and in the event of a node failure it maybe desirable to favour one guest domain over other guest domains. In this regard, before the favoured guest domain is started the less favoured guest domain would be shutdown thereby freeing system resources for the favoured guest domain.

Status

  • 26th June 2009

    HA-xVM support for LDoms webrev and HA-LDoms design document following the open CLARC commitment review.

    Here is a revised edition of man page for ccradm to aid in the code review

  • 5th Dec 2008

    Addendum to HA-xVM to include LDoms guest domains.

  • 2nd Dec 2008

    HA-xVM demo. Hit the download button from that link to stream the demo.

    The demo is based on snv_86 and SCX 2/08 simply to reflect the cheat sheet. The purpose of the demo was to show the following,

    • Show that an OHAC resource group managing an xVM guest domain via an OHAC resource can switch an xVM guest domain from one node to another using live migration. What's not shown is that the OHAC interconnects are used for the live migration.
    • Show that an OHAC managed xVM guest domain can survive a node crash.

    In particular within the demo, RG1 manages xVM domain domu1. RG1 is then switched from node podio2 to node podio1 and domu1 is subsequently live migrated between the two nodes. While RG1/domu1 is online on podio1 that node podio1 is crashed via uadmin 2 1. OHAC automatically detects that failure and restarts domu1 on podio2.

  • 19th Nov 2008

    HA-xVM webrev and requirements specification following the open CLARC inception review.

  • 19th Mar 2008

    HA-xVM webrev, agent and cheat sheet.

  • 15th Feb 2008

    OHAC agent for HA-xVM Guest Domains webrev.

  • 26th Sept 2007

    Waiting for Nevada b75 to appear in the Solaris Express Community DVD so that we can provide a documented proof of concept using OHAC with xVM.

  • 23rd Jul 2007 (Initial proof of concept).

    Here we installed Nevada b66 with OHAC, then BFU'd the latest xVM bits. This allowed us to boot both nodes running OHAC and xVM in Domain-0. We then created a cluster file system and an xVM guest domain (running Nevada b63) which was manually failed over between the two OHAC nodes, afterwards a live migration was performed. A xVM guest domain was also manually failed over using a ZFS as a Virtual Block Device.

Documentation

Getting involved