I have cluster consists of 2 physical nodes (2x Intel Xeon E5-2670 2.6Ghz , 192Gb, RAID1 SSD, ОС Debian 8, mariadb-server-10.2) + arbitrator (on Proxmox). I write to only one node at time. My cluster works fine from 1 hour to several days and stalls. It doesn't show any errors in syslog or mysql-error.log (sometime but not always shows something like)
Aug 3 10:43:32 ru-tul-dc01-db-galera-node1 mysqld: 2017-08-03 10:43:32 139707425306368 [Warning] Aborted connection 2905 to db: 'basename' user: 'username' host: 'ip' (Got an error reading communication packets)
Aug 3 10:43:32 ru-tul-dc01-db-galera-node1 mysqld: 2017-08-03 10:43:32 139707424700160 [Warning] Aborted connection 2919 to db: 'basename' user: 'username' host' host: 'ip' (Got an error reading communication packets) or lots of
Aug 3 14:35:08 ru-tul-dc01-db-galera-node1 mysqld: 2017-08-03 14:35:08 140310822119168 [Warning] WSREP: certification interval for trx source: dcdccd42-7837-17837-11e7-a045-2a82ddbaac7a version: 3 local: 1 state: CERTIFYING flags: 1 conn_id: 1010 trx_id: 10244940 seqnos (l: 152638, g: 5113465, s: 5088151, d: -1, ts: 3445: 3445393337346) exceeds the limit of 16384) PROCESSLIST shows some queries in "Query End". It cannot be killed (it just changes status from Query End to Killed. mysql -u root -e "SHOW STATUS LIKE 'wsrep_%' " shows that nodes are Synced, wsrep_local_recv_queue = 0 and so on.
The only way back it to work is to restart node.
Please help me to realise what's the problem and how to fix it.