Galera pre-deployment check

Network, SST, PKs, and failure drills before trusting wsrep.

Galera gives synchronous multi-master replication for InnoDB — and new failure modes if you skip homework.

Network

  • Low-latency, non-lossy links between nodes
  • Explicit membership (wsrep_cluster_address)
  • Open SST/IST ports, not only 3306
  • Avoid flaky WAN between nodes unless you have measured RTT and packet loss

State transfer

  • Know whether you use rsync, xtrabackup, or mariabackup SST
  • Donor nodes need IO and network headroom during SST
  • Rehearse a full SST in staging until it is boring
  • Confirm IST covers the common restart case

Application compatibility

  • InnoDB (or compatible) tables only for replicated writes
  • Explicit primary keys on every table
  • Careful with LOCK TABLES, GET_LOCK, and long XA dreams
  • Retry deadlocks and certification failures

Smoke before production

  1. Bootstrap node 1; join 2 and 3
  2. Kill a node under load; confirm IST recovery
  3. Document fencing policy for split-brain
  4. Run a schema change under the wsrep_OSU_method you actually chose

Not production until you have recovered a node while tired.