Synnefo » snf-ganeti » ganeti-local

  <!ENTITY dhdate      "<date>February 12, 2009</date>">

      <year>2008</year>

      <year>2009</year>

    <refmiscinfo>ganeti 2.0</refmiscinfo>

# gnt-instance add -n node2.example.com \

> -o debootstrap --disk 0:size=30g \

      version (2.0) supports Xen 3.0 (also tested with 3.1) and KVM

      hypervisors.

      <userinput>gnt-cluster init</userinput>.

  <refsect1>

    <title>Cluster architecture</title>

    <para>

      In Ganeti 2.0, the architecture of the cluster is a little more

      complicated than in 1.2. The cluster is coordinated by a master

      daemon (<citerefentry>

      <refentrytitle>ganeti-masterd</refentrytitle>

      <manvolnum>8</manvolnum> </citerefentry>), running on the master

      node. Each node runs (as before) a node daemon, and the master

      has the <acronym>RAPI</acronym> daemon running too.

    </para>

    <refsect2>

      <title>Node roles</title>

      <para>Each node can be in one of the following states:

        <variablelist>

          <varlistentry>

            <term>master_candidate</term>

            <listitem>

              <para>The node receives the full cluster configuration

              (configuration file and jobs) and can become a master

              via the <command>gnt-cluster masterfailover</command>

              command. Nodes that are not in this state cannot

              transition into the master role due to missing

              state.</para>

            </listitem>

          </varlistentry>

          <varlistentry>

            <term>regular</term>

            <listitem>

              <para>This the normal state of a node.</para>

            </listitem>

          </varlistentry>

          <varlistentry>

            <term>drained</term>

            <listitem>

              <para>Nodes in this state are functioning normally but

              cannot receive new instance, because the intention is to

              set them to <emphasis>offline</emphasis> or remove them

              from the cluster.</para>

            </listitem>

          </varlistentry>

          <varlistentry>

            <term>offline</term>

            <listitem>

              <para>These nodes are still recorder in the ganeti

              configuration, but except for the master daemon startup

              voting procedure, they are not actually contacted by the

              master. This state was added in order to allow broken

              machines (that are being repaired) to remain in the

              cluster but without creating problems.</para>

            </listitem>

          </varlistentry>

        </variablelist>

      </para>

    </refsect2>

    <refsect2>

      <title>Cluster configuration</title>

      <para>The master node keeps and is responsible for the cluster

      configuration. On the filesystem, this is stored under the

      <filename

      class="directory">@LOCALSTATEDIR@/ganeti/lib</filename>

      directory, and if the master daemon is stopped it can be backed

      up normally.</para>

      <para>The master daemon will replicate the configuration

      database called <filename>config.data</filename> and the job

      files to all the nodes in the master candidate role. It will

      also distribute a copy of some configuration values via the

      <emphasis>ssconf</emphasis> files, which are stored in the same

      directory and start with <filename>ssconf_</filename> prefix, to

      all nodes.</para>

    </refsect2>

    <refsect2>

      <title>Jobs</title>

      <para>

        All cluster modification are done via jobs. A job consists of

        one or more opcodes, and the list of opcodes is processed

        serially. If an opcode fails, the entire job is failed and

        later opcodes are no longer processed. A job can be in one of

        the following states:

        <variablelist>

          <varlistentry>

            <term>queued</term>

            <listitem>

              <simpara>The job has been submitted but not yet

              processed by the master daemon.</simpara>

            </listitem>

          </varlistentry>

          <varlistentry>

            <term>waiting</term>

            <listitem>

              <simpara>The job is waiting for for locks before the

              first of its opcodes.</simpara>

            </listitem>

          </varlistentry>

          <varlistentry>

            <term>canceling</term>

            <listitem>

              <para>The jos is waiting for locks, but is has been

              marked for cancelation. It will not transition to

              <emphasis>running</emphasis>, but to

              <emphasis>canceled</emphasis>.

              </para>

            </listitem>

          </varlistentry>

          <varlistentry>

            <term>running</term>

            <listitem>

              <simpara>The job is currently being executed.</simpara>

            </listitem>

          </varlistentry>

          <varlistentry>

            <term>canceled</term>

            <listitem>

              <para>The job has been canceled before starting

              execution.</para>

            </listitem>

          </varlistentry>

          <varlistentry>

            <term>success</term>

            <listitem>

              <para>The job has finished successfully.</para>

            </listitem>

          </varlistentry>

          <varlistentry>

            <term>error</term>

            <listitem>

              <para>The job has failed during runtime, or the master

              daemon has been stopped during the job execution.</para>

            </listitem>

          </varlistentry>

        </variablelist>

      </para>

    </refsect2>

  </refsect1>