What consensus protocol does Qdrant use for distributed clusters, and how does it work?

Qdrant uses the Raft consensus algorithm to keep cluster-wide metadata — things like collection configuration, shard placement, and cluster membership — consistent across all nodes in a distributed deployment, even as nodes join, leave, or fail.

Raft works by electing one node as leader at any given time; the leader is the only node that accepts new metadata changes, which it replicates to follower nodes as log entries, and a change is only considered committed once a majority of nodes have acknowledged it — a design that guarantees consistency even if some nodes are temporarily unreachable, as long as a majority remains available.

If the leader node fails, the remaining nodes hold a new leader election among themselves, and once a new leader is chosen, cluster metadata operations resume without requiring manual intervention — this is specifically why a Qdrant cluster needs a minimum of three nodes for meaningful fault tolerance, since a two-node cluster can't establish a true majority if one node is lost.