Multinode systems

A Vespa system consists of one or more stateless and stateful clusters configured by one application package. A Vespa system is configured and managed through an admin cluster as shown below.

All nodes of a Vespa system has the same software installed. Which processes are started on each node and how they are configured is determined by the admin cluster from the specification given in the services.xml file in the application package.

Making changes to live systems

To change the system, deploy the changed application to the admin cluster. The admin cluster will automatically change the participating nodes as necessary. It is safe to do this while serving live query and write traffic. In rare cases the admin cluster will tell you that some processes must be restarted to make the change effective. To avoid query or write traffic disruption, such restarts must be done on one node at the time, and you must wait until the node is fully up before restarting the next node.

Creating a multinode system from a sample application

To create a fully functional production ready multinode system from a single-node sample application, follow these steps:

  • Add an explicit admin cluster to your services.xml file:
    <services version="1.0">
      <admin version="2.0">
        <adminserver hostalias="admin0"/>
          <configserver hostalias="admin0"/>
    This cluster can have multiple nodes as config servers, but it is not strictly necessary as the other nodes will continue to run when the admin cluster is unreachable.
  • Install the Vespa image or Docker image on all the nodes and start Vespa on the nodes.
  • Run
    echo "override VESPA_CONFIGSERVERS [adminserver-hostname]" >> /opt/vespa/conf/vespa/default-env.txt
    where [adminserver-hostname] is replaced by the full hostname of the admin server (or a comma-separated list if you configured multiple ones).
  • Add each node of your system to the src/main/application/hosts.xml file (use any aliases you like).
  • Add these nodes to your container and content clusters as desired by adding more node tags in services.xml (to avoid traffic disruption on restarts or node failure you need at least two nodes in each cluster). You can add the same nodes to multiple clusters if you like.
With this, your application is fully production ready!

Log collection

Logs are automatically collected from all nodes in real time to the admin node listed as "adminserver". To view log messages from the system, run vespa-logfmt on an admin node.