Query Performance with Failed Nodes
When a node fails within a Vertica cluster, the cluster detects the failure within milliseconds. The impact of a node failure on currently running queries depends on the K-safety level of your cluster and the location of your data within the cluster.
K-Safety 0 Cluster
If a node fails in a cluster with a K-safety of 0, any currently running queries roll back and the cluster shuts down. You cannot run further queries until you restore the failed node.
K-Safety 1 or Greater Cluster
The failure of a node can invalidate the query plan for a currently running query. If the running query does not require the failed node, the query continues to run. If a currently running query does require the failed node, the query rolls back. Vertica then automatically reconfigures the query plan to use alternate nodes and automatically retries the query. By default, Vertica makes three attempts to rerun the query before canceling it.
In clusters with a failed node, the database optimizer chooses different segmentations of buddy projections to distribute the workload throughout the cluster. Query performance within a cluster containing a failed node can, however, deteriorate, as one node must now do the work of two nodes. The extent of the performance impact varies depending on the queries that you run and the location of the data that you are querying. Query performance remains affected until the failed node has been completely restored.