graceful

graceful

Lifecycle orchestration for long-running Go components: start everything together, wait for the context to be cancelled or the first failure, drain everything in reverse order — with optional zero-downtime binary upgrades on SIGHUP via tableflip.

Features

• Any component, not just HTTP. Add(name, start, stop) takes a pair of functions: start may block for the component's whole life (an accept loop) or return immediately after spawning its own work (a scheduler). A non-nil error from any start shuts the whole group down.
• Listeners built for upgrades. Listen(name, addr, srv) creates the listener inside Run — through tableflip when upgrades are enabled — so an upgraded process inherits the socket without dropping connections. *http.Server satisfies the Server interface directly.
• One shutdown story. A cancelled context, a component failure, or an upgrade handoff all funnel into the same drain: every started component is stopped in reverse registration order, under one shared deadline, and drain failures come back joined onto the returned error.
• Zero-downtime upgrades. WithUpgrade() makes SIGHUP re-exec the binary and hand off listeners; it silently falls back to plain listeners on Windows.
• Structured logging of every lifecycle event through log/slog.

Install

go get github.com/libtnb/graceful

Requires Go 1.27+.

Quick start

package main

import (
	"context"
	"net/http"
	"os/signal"
	"syscall"
	"time"

	"github.com/libtnb/graceful"
)

func main() {
	ctx, stop := signal.NotifyContext(context.Background(), syscall.SIGINT, syscall.SIGTERM)
	defer stop()

	srv := &http.Server{Handler: mux}

	g := graceful.New(
		graceful.WithUpgrade(),                      // SIGHUP = hot upgrade
		graceful.WithShutdownTimeout(30*time.Second),
	)
	g.Add("cron", cron.Start, cron.Stop)             // any start/stop pair
	g.Listen("http", ":8080", srv)                   // upgrade-aware listener

	if err := g.Run(ctx); err != nil {               // blocks until shutdown
		log.Fatal(err)
	}
}

Run starts entries in registration order and drains them in reverse: the HTTP listener above stops accepting before the scheduler is asked to finish, so in-flight requests can still schedule work.

Design notes

• start errors are fatal, stop errors are collected. A component that cannot run means the process is broken — everything comes down. A component that cannot stop cleanly must not block the rest from draining, so its error is logged and joined onto Run's return value instead.
• A scheduler-style start that returns nil is a successful launch, not a failure — only non-nil errors trigger shutdown. This makes Add fit both blocking accept loops and fire-and-forget starters without adapters.
• Registration order is the dependency order. Register infrastructure first, entry points last; reverse-order draining then closes the front door before the back office.
• The caller owns shutdown, the group owns SIGHUP. Shutdown arrives through the context (signal.NotifyContext in main), so the group composes with any cancellation source and tests need no real signals; SIGHUP stays internal because upgrades are the group's own feature.
• The group owns lifecycles, not resources. Database pools, log writers and the like belong to whatever built them (a DI container's cleanup); the group only coordinates starting and stopping.

License

MIT