Maintenance window is a time window during which we'll switch to new upgraded virtual machines and point the connection URI's DNS name at the new server.

Aiven service upgrades are performed in rolling forward style, which means that new service nodes are first created alongside with the older nodes one at a time, after which the old nodes are retired.

In case of PostgreSQL the maintenance window usually lasts around 10 seconds. The downtime comes from the time it takes for the changed DNS entry that points to the location of the new master server to get propagated globally.

In case of Kafka and Elasticsearch the service DNS address resolves to all the available service nodes. During an upgrade the DNS address changes to reflect the added and removed nodes. For example, a three node plan will have a minimum of three nodes available at all times during the whole upgrade operation.

Note: while the DNS name remains the same, the IP address it points to, will change during a maintenance break.

Did this answer your question?