Quorum loss occurs when enough replicas are unavailable that ScyllaDB cannot satisfy a requested consistency level. A sizable share of requests may then fail until quorum is restored. section 11.3.3 section 11.4
Reduce the chance of quorum loss during cluster operations:
- Automate procedures so safety checks replace manual decisions. section 11.3.3 section 11.4
- Take one node offline at a time; removing more nodes increases the risk. section 11.3.3
- Validate each node’s health before continuing with a cluster-wide operation. section 11.3.3
Restoring service
If infrastructure failure takes enough nodes offline, restore those nodes or otherwise bring the cluster back to quorum. Requests can then succeed at their desired consistency level again. If unsafe maintenance caused the outage, improve automation and perform subsequent changes sequentially. section 11.3.3 section 11.4
