goqite (pronounced Go-queue-ite) is a persistent message queue Go library built on SQLite and inspired by AWS SQS (but much simpler).
Made in 🇩🇰 by maragu, maker of online Go courses.
- Messages are persisted in a SQLite table.
- Messages are sent to and received from the queue, and are guaranteed to not be redelivered before a timeout occurs.
- Support for multiple queues in one table.
- Message timeouts can be extended, to support e.g. long-running tasks.
- A job runner abstraction is provided on top of the queue, for your background tasks.
- A simple HTTP handler is provided for your convenience.
- No non-test dependencies. Bring your own SQLite driver.
package main
import (
"context"
"database/sql"
"fmt"
"log"
"time"
_ "github.com/mattn/go-sqlite3"
"github.com/maragudk/goqite"
)
func main() {
// Bring your own database connection, since you probably already have it,
// as well as some sort of schema migration system.
// The schema is in the schema.sql file.
// Alternatively, use the goqite.Setup function to create the schema.
db, err := sql.Open("sqlite3", ":memory:?_journal=WAL&_timeout=5000&_fk=true")
if err != nil {
log.Fatalln(err)
}
db.SetMaxOpenConns(1)
db.SetMaxIdleConns(1)
if err := goqite.Setup(context.Background(), db); err != nil {
log.Fatalln(err)
}
// Create a new queue named "jobs".
// You can also customize the message redelivery timeout and maximum receive count,
// but here, we use the defaults.
q := goqite.New(goqite.NewOpts{
DB: db,
Name: "jobs",
})
// Send a message to the queue.
// Note that the body is an arbitrary byte slice, so you can decide
// what kind of payload you have. You can also set a message delay.
err = q.Send(context.Background(), goqite.Message{
Body: []byte("yo"),
})
if err != nil {
log.Fatalln(err)
}
// Receive a message from the queue, during which time it's not available to
// other consumers (until the message timeout has passed).
m, err := q.Receive(context.Background())
if err != nil {
log.Fatalln(err)
}
fmt.Println(string(m.Body))
// If you need more time for processing the message, you can extend
// the message timeout as many times as you want.
if err := q.Extend(context.Background(), m.ID, time.Second); err != nil {
log.Fatalln(err)
}
// Make sure to delete the message, so it doesn't get redelivered.
if err := q.Delete(context.Background(), m.ID); err != nil {
log.Fatalln(err)
}
}
package main
import (
"context"
"database/sql"
"fmt"
"log/slog"
"time"
_ "github.com/mattn/go-sqlite3"
"github.com/maragudk/goqite"
"github.com/maragudk/goqite/jobs"
)
func main() {
log := slog.Default()
// Setup the db and goqite schema.
db, err := sql.Open("sqlite3", ":memory:?_journal=WAL&_timeout=5000&_fk=true")
if err != nil {
log.Info("Error opening db", "error", err)
}
db.SetMaxOpenConns(1)
db.SetMaxIdleConns(1)
if err := goqite.Setup(context.Background(), db); err != nil {
log.Info("Error in setup", "error", err)
}
// Make a new queue for the jobs. You can have as many of these as you like, just name them differently.
q := goqite.New(goqite.NewOpts{
DB: db,
Name: "jobs",
})
// Make a job runner with a job limit of 1 and a short message poll interval.
r := jobs.NewRunner(jobs.NewRunnerOpts{
Limit: 1,
Log: slog.Default(),
PollInterval: 10 * time.Millisecond,
Queue: q,
})
// Register our "print" job.
r.Register("print", func(ctx context.Context, m []byte) error {
fmt.Println(string(m))
return nil
})
// Create a "print" job with a message.
if err := jobs.Create(context.Background(), q, "print", []byte("Yo")); err != nil {
log.Info("Error creating job", "error", err)
}
// Stop the job runner after a timeout.
ctx, cancel := context.WithTimeout(context.Background(), 30*time.Millisecond)
defer cancel()
// Start the job runner and see the job run.
r.Start(ctx)
}
Just for fun, some benchmarks. 🤓
On a MacBook Pro with M3 Ultra chip and SSD, sequentially sending, receiving, and deleting a message:
$ make benchmark
go test -cpu 1,2,4,8,16 -bench=.
goos: darwin
goarch: arm64
pkg: github.com/maragudk/goqite
BenchmarkQueue/send,_receive,_delete 21444 54262 ns/op
BenchmarkQueue/send,_receive,_delete-2 17278 68615 ns/op
BenchmarkQueue/send,_receive,_delete-4 16092 73888 ns/op
BenchmarkQueue/send,_receive,_delete-8 15346 78255 ns/op
BenchmarkQueue/send,_receive,_delete-16 15106 79517 ns/op
Note that the slowest result above is around 12,500 messages / second with 16 parallel producers/consumers. The fastest result is around 18,500 messages / second with just one producer/consumer. (SQLite only allows one writer at a time, so the parallelism just creates write contention.)