Your first records

Applies to: the public baseline and newer development builds. Install smkv and smkv-tools first. Commands below use the loopback data port.

Start one node

For a package installation, use the service configured in installation:

sudo systemctl start smkv

Alternatively, run a foreground evaluation process with a new directory:

smkv-server --data-dir ./smkv-demo --bind 127.0.0.1:7379

Use one approach at a time; two processes cannot share the same port or data directory. Keep the foreground terminal open and use another for the following commands.

Write and read

smkv --addr 127.0.0.1:7379 --sync put greeting hello
smkv --addr 127.0.0.1:7379 get greeting

The read prints hello without a trailing newline. Keys and values passed through this CLI are UTF-8 arguments; use the SDK for arbitrary bytes. --sync requests local durable storage before acknowledgment. It does not wait for a replica.

Expire a record

smkv --addr 127.0.0.1:7379 put session active --ttl-ms 1000
sleep 2
smkv --addr 127.0.0.1:7379 get session
echo $?

The expired read prints no value and returns exit code 2. Omit TTL for no expiration. Expiration removes logical visibility; it does not immediately reclaim every byte on disk.

Verify restart recovery

For the package service:

sudo systemctl restart smkv
smkv --addr 127.0.0.1:7379 get greeting

For the foreground process, press Ctrl-C and rerun the identical server command with the same directory. Wait for startup before reading. greeting should still be hello. Ctrl-C and SIGTERM drain accepted work and synchronize partitions. This orderly restart check does not prove resilience to every machine failure.

Delete a record

smkv --addr 127.0.0.1:7379 --sync delete greeting
smkv --addr 127.0.0.1:7379 get greeting
echo $?

The read returns exit code 2. Operational failures return 1. Never interpret an empty value alone as a miss; empty values are valid.

Choose write guarantees

Default writes can be lost after failure. --sync makes the affected local partition durable. Newer quorum-managed development clusters additionally support --replica-ack and --durable-replica-ack; the latter waits for both copies to sync. These flags do not create transactions or protect against losing both disks.

A timed-out mutation may already have completed. Resolve that uncertainty in the application instead of blindly replaying writes. Next, configure storage budgets or use the Go SDK.