A software project that starts at the foundations.

datumujo

An embedded database for Go

• v0.1.0
• EUPL-1.2
• Go standard library only

No server, nothing to install. You get datumujo like any other Go library, and your data lives in one file next to your program. The name is Esperanto for roughly "data container".

What v0.1.0 does:

• Transactions that survive a crash or a power loss. When a commit returns, it is on disk.
• One writer at a time, and readers never wait for it.
• Tables with a primary key, indexes (unique ones too), scans in both directions, and counters.
• A checksum on every page of the file.
• Backup while writes go on.

What it looks like

Create a table, write a row, read everything back. This program runs as shown against v0.1.0:

package main

import (
	"fmt"

	"github.com/dyadik-eu/datumujo"
)

func main() {
	db, err := datumujo.Open("notizen.db", datumujo.Options{})
	if err != nil {
		panic(err)
	}
	defer db.Close()

	err = db.Update(func(tx *datumujo.Tx) error {
		err := tx.CreateTable(datumujo.Def{
			Name: "notiz",
			Columns: []datumujo.Column{
				{Name: "id", Type: datumujo.Int64},
				{Name: "text", Type: datumujo.String},
			},
			Key: []string{"id"},
		})
		if err != nil {
			return err
		}
		return tx.Insert("notiz", datumujo.Row{int64(1), "Hallo, Leibniz"})
	})
	if err != nil {
		panic(err)
	}

	err = db.Read(func(v *datumujo.View) error {
		rows, err := v.Scan("notiz", datumujo.ScanOptions{})
		if err != nil {
			return err
		}
		for rows.Next() {
			fmt.Println(rows.Row())
		}
		return rows.Err()
	})
	if err != nil {
		panic(err)
	}
}

Output:

[1 Hallo, Leibniz]

And if a byte flips?

Then you get an error with the page number. You do not get wrong data. Tried on a small database with one table: a copy, with a single byte on page 2 overwritten.

$ datumujo check notizen.db
file: 12288 bytes, log: 28 bytes
pages: 3 of 4096 bytes, 0 free, 1 in the catalog
schema version: 1
table notiz: 1 rows, 1 pages
intact
$ cp notizen.db kaputt.db
$ printf '\xff' | dd of=kaputt.db bs=1 seek=8292 conv=notrunc status=none
$ datumujo check kaputt.db
fault: page 2: checksum does not match
file: 12288 bytes, log: 0 bytes
pages: 3 of 4096 bytes, 0 free, 0 in the catalog
schema version: 0
1 faults

datumujo check then exits with code 1. For an intact file the code is 0, and 2 if it could not check at all.

What is still missing

SQL. I am working on it now, as the second stage. Until v1, the API and the file format can still change. The changelog says the same.