caddyhttp: Honor grace period in background (#5043)

* caddyhttp: Honor grace period in background

This avoids blocking during config reloads.

* Don't quit process until servers shut down

* Make tests more likely to pass on fast CI (#5045)

* caddyhttp: Even faster shutdowns

Simultaneously shut down all HTTP servers, rather than one at a time.

In practice there usually won't be more than 1 that lingers. But this
code ensures that they all Shutdown() in their own goroutine
and then we wait for them at the end (if exiting).

We also wait for them to start up so we can be fairly confident the
shutdowns have begun; i.e. old servers no longer
accepting new connections.

* Fix comment typo

* Pull functions out of loop, for readability

caddy.go

@@ -31,6 +31,7 @@ import (
 	"strconv"
 	"strings"
 	"sync"
+	"sync/atomic"
 	"time"
 
 	"github.com/caddyserver/caddy/v2/notify"
@@ -676,6 +677,10 @@ func Validate(cfg *Config) error {
 // Errors are logged along the way, and an appropriate exit
 // code is emitted.
 func exitProcess(ctx context.Context, logger *zap.Logger) {
+	// let the rest of the program know we're quitting
+	atomic.StoreInt32(exiting, 1)
+
+	// give the OS or service/process manager our 2 weeks' notice: we quit
 	if err := notify.Stopping(); err != nil {
 		Log().Error("unable to notify service manager of stopping state", zap.Error(err))
 	}
@@ -739,6 +744,12 @@ func exitProcess(ctx context.Context, logger *zap.Logger) {
 	}()
 }
 
+var exiting = new(int32) // accessed atomically
+
+// Exiting returns true if the process is exiting.
+// EXPERIMENTAL API: subject to change or removal.
+func Exiting() bool { return atomic.LoadInt32(exiting) == 1 }
+
 // Duration can be an integer or a string. An integer is
 // interpreted as nanoseconds. If a string, it is a Go
 // time.Duration value such as `300ms`, `1.5h`, or `2h45m`;