Distributed & Cloud

DistributedCloud

AetheriusDB runs as a single database across a wide range — from a local replica on a device, out to a cluster spanning multiple cloud providers. This section covers the features that make that work: keeping edge replicas in sync, controlling cloud cost, treating object storage as a native tier, and backing up and restoring a whole cluster.

These are advanced and cloud-oriented capabilities. Each page carries an availability badge — most are Beta or Roadmap rather than on by default — so you can tell what you can rely on today versus what is still evolving.

Clustering, multi-tenancy and high availability are premium features. They exist only when the license key grants them (cluster, multi_tenancy, replication); no flag or environment variable turns them on. To license them, contact the AetheriusDB team at aetheriusdb.com or aetheriuslabs.com. See Licensing → Premium features.

Why it matters

The same database stretches from a phone-sized edge replica to a multi-cloud cluster, so you don’t re-platform as you scale out. Reads can be served locally for zero latency, cost can follow policy across providers, and a whole cluster can be snapshotted and rebuilt without a bespoke backup pipeline.

Local-first, offline-capable — an edge replica answers reads with no network hop and merges its writes back automatically when reconnected.
Object storage as a native tier — S3/GCS/Azure are storage the engine reads directly, not an import step.
Whole-cluster backup/restore — snapshot and clone the cluster from object storage instead of scripting per-table dumps.

How fast it is

0 ms reads at the edge. A local replica serves queries without leaving the device — no round trip to the cluster.
Boot query-ready from a backup in seconds. A restored cluster comes up answering queries from object storage, rather than replaying a long restore.

In this section

Table Clustering & the God Cluster — assign each table to the cluster that owns and writes it, keep tables that join each other in one cluster so the join runs on one worker, and plan for the God Cluster: shared tables replicated to every worker (designed; the topology key and write path are not wired yet).
Edge Sync — run a local AetheriusDB replica for zero-latency reads and offline writes that merge back to the cluster automatically.
Query Pushdown — send the question to the machine holding the data and get back just the answer, so a summary’s reply stays small however large the table grows.
Cloud Arbitrage — an autonomous cost layer that moves eligible workload to the cheapest viable cloud location within your policy.
Cloud Runtime — object storage, streaming sources, and cross-cloud failover as native storage tiers and SQL primitives.
Cluster Backup & Restore — fast snapshots, incremental uploads to object storage, and clusters that boot query-ready from a backup in seconds.
Cluster Execution Engine — an orchestrator compiles SQL to 64-bit bytecode; workers run it in a register VM over their own column stripes and return only answers. With reference pages for the Tiny VM ISA, the Bytecode Compiler, the Worker Node, Pushdown Joins & Aggregates and the Wire Protocol Versions.
Shadow HA Streamer — asynchronous replication of a primary’s containers to its standby, driven by heartbeats, never blocking ingest.
Multi-Tenancy (Fluid Clusters) — one cluster, many tenants: every connection executes for one tenant, each tenant’s data sits in its own directory, noisy neighbours are throttled before they reach a worker, and a tenant snapshots and restores in constant time.
Cluster Transactions — BEGIN and COMMIT across the cluster, readers that only see committed data, instant rollback, write conflicts that fail fast, and a watchdog for stalled transactions.
Deletes, Updates & the Vacuum — DELETE and UPDATE mark and append; a background vacuum compacts files without blocking a reader or a writer.
Secondary Indexes on Cluster Tables — point lookups, ranges and LIKE 'prefix%' without a scan, on BIGINT and VARCHAR columns, kept through the vacuum and every snapshot.
VARCHAR in Cluster Tables — text columns you can insert, select, compare, prefix-search and index.
Engine Hardening Log — the phase-by-phase record of how the engine landed, what it measured, and what broke on the way.