LEGEND

• Clients connect over the native Java client, JDBC, JSON/HTTP & REST, C++, and Python — plus sqlcmd, the csvloader/kafkaloader/jdbcloader bulk loaders, and the VMC web console for operations.
• Stored procedures are the unit of work: each call is a single ACID transaction, precompiled and deterministic, written in Java + SQL. Single-partition, multi-partition, directed, and ad-hoc SQL all run with no locks and no logs at serializable isolation.
• Distributed hosts & K-safe partitions — the cluster shards data across sites on every host. Each host owns partition masters and holds K-safety replicas of other hosts' partitions, so a node can fail without data loss or downtime.
• Partitioned tables (e.g. ORDERS, USERS, EVENTS) are sharded by a partition column; a procedure touching one partition key runs lock-free, single-threaded on that site. Replicated tables (e.g. PRODUCTS, COUNTRIES) keep a full copy on every host so procedures can join against them while staying single-partition.
• Export & replication stream committed data out to Kafka, Iceberg, Snowflake, Databricks, S3, Elasticsearch, HTTP, and JDBC, with Active(N) XDCR for cross-datacenter replication.
• Durability & operations pair in-memory speed with on-disk safety — command logs, snapshots, XDCR, live online operations, and cloud-native Kubernetes deployment.