Databases — ozpanel operator reference

Three engines

Engine	Topology	Failover decided by	Holds
MariaDB + Galera	Synchronous multi-primary; `garbd` at even node counts	Galera weighted quorum	Panel state, tenant MySQL
PostgreSQL + repmgr + PgBouncer	One primary, hot standby per node, PgBouncer per node	Signed declaration (repmgr `failover=manual`)	Tenant PostgreSQL, CrowdSec, JuiceFS metadata
Redis + Sentinel	One master, replicas, Sentinel per node	Signed declaration	Rspamd, sessions, cache, Caddy certs, counters

Operator · Databases. — ozpanel screenshot — **Operator · Databases.**

MariaDB and Galera

Ports3306 (peers + tenant container bridge), 4567, 4568 IST, 4444 SST, 4569 garbd. Peers only.
TLSCluster CA certs. GET/POST /galera/ssl.
Ready gateozpanel-galera-ready.service: Primary and Synced before consumers start.
AccountsPer-node service accounts (ozpanel_<tag>). POST /mariadb/panel-credential/rotate.
Crashozpanel galera-recover: rejoin, join-only.
Lone survivorBootstraps alone only if every guard passes; else stays down and says why.
Start safetygcache.recover=no; MemoryMax = half of RAM.
All downPOST /galera/recover-position everywhere → POST /galera/bootstrap on the highest → POST /galera/join others. See Cluster.

PostgreSQL, repmgr, PgBouncer

ConnectAlways local PgBouncer 127.0.0.1:6432, pool_mode=transaction. Never 5432.
Replicationwal_keep_size=1GB, wal_log_hints=on. No slots, no WAL archive, no PITR.
SynchronousDefault synchronous_commit=local; JuiceFS role uses on. FIRST 1 over the succession. Lone survivor releases it, alerts.
PromotionLag ≤ max_lag_bytes (1 MiB). Recorded sync standby only.
FencesSelf-fence · boot gate · rejoin marker · pooler repointed every pass.
PgBouncer authauth_query via a single-function role. /pgbouncer/auth-query/enable, …/rotate.
ReadoutsGET /pgbouncer/failover, /postgresql/verify, /repmgrd/sync, /cluster/sync-replication.
MovesUse POST /cluster/handover-to. /repmgrd/rejoin rebuilds a standby.

Redis and Sentinel

One deciderreplica-priority 0 everywhere; the declared primary commands the move.
Two nodesSecond Sentinel on 26380 on the replica; min-replicas-to-write 1.
AccountsOne ACL account per service; FLUSHALL, CONFIG, ACL, SHUTDOWN, REPLICAOF, DEBUG removed. /redis/acl, …/{service}/rotate.
Memorymaxmemory 1gb, noeviction at install.
CheckGET /redis/verify, /redis/topology.

Tenant databases

Tenant console (Databases & tasks) or operator. Name ozt<uid>_<name>, name part ^[a-z][a-z0-9_]{0,31}$.

POST /tenants/{id}/databases {"name":"shop","engine":"mysql"}        # mysql | postgres; owner user, password once
POST /tenants/{id}/db-users  {"name":"report","engine":"mysql"}
POST /tenants/{id}/databases/{name}/grants {"user":"ozt100001_report","privilege":"readonly"}

`privilege`	MariaDB	PostgreSQL
`readonly`	SELECT	CONNECT, USAGE, SELECT (incl. future)
`readwrite`	SELECT, INSERT, UPDATE, DELETE	Same + sequences
`all` (default)	Full owner list on that DB	All on tables, sequences

Passwords20 chars, shown once. Reset: …/databases/{name}/password, …/db-users/{user}/password.
DeleteDatabase + bundled user. PostgreSQL WITH (FORCE). User owning a DB: 409.
SuspendLogins locked, sessions closed, nothing dropped.
LimitsPlan databases, db_users. Size measured, not capped.
IsolationData private. PostgreSQL catalogue lists names; GET /postgresql/tenant-visibility.
Importmariadb to 127.0.0.1:3306 or psql to :6432. Upload endpoint: not available.
BackupOne gzip dump per database in each snapshot, checksummed. See Backups.

Client	MariaDB	PostgreSQL	Redis
Site, cron	127.0.0.1:3306	127.0.0.1:6432	127.0.0.1:6379, ACL
Container	`ozpanel-db:3306`	—	—
Internet	Never

Tenant · Databases & tasks. — ozpanel screenshot — **Tenant · Databases & tasks.**

Engine settings

Operator. GET/PUT /{module}/settings; validated, applied, rolled back on failure. Cluster-wide fields applied on every node, including later joiners. Config triad: /{module}/config/live, /stored, /diff.

Module	Cluster-wide (default)	Per node
`mariadb`	max_connections 151 · max_allowed_packet 16M · wait/interactive_timeout 28800 · thread_cache_size 151 · innodb_flush_log_at_trx_commit 2 · slow_query_log OFF · long_query_time 10 · general_log OFF	innodb_buffer_pool_size · innodb_io_capacity · table_open_cache · tmp_table_size · max_heap_table_size
`postgresql`	max_connections 200 · superuser_reserved 3 · max_worker_processes 8 (rolling restart) · idle_in_transaction 0 · max/min_wal_size 1GB/80MB · wal_keep_size 1GB · checkpoint 0.9 · wal_compression off · statistics 100 · log_min_duration -1 · log_lock_waits off · autovacuum on	shared_buffers · effective_cache_size · work_mem · maintenance_work_mem · random_page_cost · effective_io_concurrency · parallel workers
`pgbouncer`	pool_mode transaction · default_pool_size 25 · max_client_conn 500 · server_idle 600 · server_lifetime 3600 · query_wait 120 · client_login 60 · others 0
`repmgrd`	failover manual · connection_check_type · reconnect/monitor timing · monitoring_history off	priority
`redis`	noeviction · samples 5 · save 3600 1 300 100 60 10000 · appendonly yes · everysec · maxclients 10000 · tcp-keepalive 300 · loglevel notice · slowlog 10000/128	maxmemory · tcp-backlog
`redis-sentinel`	down-after 10000 · failover-timeout 60000 · parallel-syncs 1
`garbd`	galera_nodes · galera_group · galera_options · log_file

POST /postgresql/param sets other parameters (replication, listener, paths refused). POST /redis/param, /pgbouncer/param, /postfix/param likewise.

CLI

Command	Run by
`ozpanel galera-ready-gate`, `galera-position`, `galera-recover`	systemd
`ozpanel boot-gate [--pgdata] [--wait-for-peer]`	PostgreSQL pre-start
`ozpanel pgbouncer-follow-primary`	repmgr event hook
`ozpanel migrate`, `ozpanel recover`	Operator