Confluence down (Hazelcast Split Brain)

Incident Report for ByteSource Technology Consulting GmbH

Resolved

Incident was triggered by Delete Requests, the Delete Events made the DB slower what lead to slow responses and to many stuck threads (>6500). As no execution threads were available the Hazelcast declared a "split brain" status (which practically switched off all other nodes). After such a split brain issue all nodes must be restarted.

Posted Aug 13, 2025 - 15:23 CEST