SMKV documentation

SMKV is a Rust database and persistent distributed cache for read-heavy applications. Run it on your own Linux hosts, keep the key index in memory, and store records on local disk. Use byte values or schemaless tables with typed columns.

Start with a working node

1. Install SMKV: install only the components each host needs.
2. Write your first records: try reads, writes, expiration, and restart recovery.
3. Configure a node: choose explicit storage and memory headroom.
4. Create a cluster: distribute partitions across primary and replica owners.
5. Connect your application: use the Go SDK and handle errors deliberately.

Choose the right release

Public APT baseline: 0.1.0-2. The signed repository provides server, tools, and dashboard packages for Ubuntu 24.04 amd64. It does not include all features described in these guides.

Development build: 0.1.1 source at a019cc5. Capacity protection, authority failover, replicated acknowledgments, resizing, optional transport security, scans, and manual backup belong to the newer development line. Obtain a matching build and SDK through early-access support before following those sections. A package with the same application version is not necessarily built from this revision.

Experimental build: S3 cross-cluster sync and scheduled sync checkpoints require an explicitly enabled build feature and use a separate compatibility contract. They are not enabled by ordinary installation.

Guides distinguish these tracks. Check installed package versions and each binary's --help; do not assume a source-tree feature is available in APT. Release availability was reviewed on October 11, 2026.

Understand the operating model

• A node owns persistent local files. Keep its complete data directory across restart.
• A partition has one primary and, in RF2 clusters, one replica on a different node.
• Default writes are asynchronous. A successful response does not guarantee survival of primary or machine failure.
• Local sync, replica confirmation, and durable replica confirmation provide different guarantees. Select them explicitly.
• Replica reads may be stale. Scans are best-effort and yield to foreground reads and writes.
• Partition count is chosen before ingestion; adding nodes moves ownership without changing that count.

SMKV is early access. Validate your record sizes, load, failure behavior, and recovery time before production use. There is no SQL interface, cross-record transaction, secondary index, or automatic capacity eviction.

Operate and recover

Use monitoring to inspect the cluster, backup and restore to create portable recovery archives, and operations for maintenance and resizing. The troubleshooting guide starts with symptoms and concrete checks.

For sizing, release access, or support, contact [email protected]. No public source repository is required to install the signed packages.