Reload hook (#1445)

* Add reload directive

* gofmt

* Fix default jitter and error message

* remove unneeded call to NextArg, add a couple negative setup tests

* Review feedback

core/dnsserver/zdirectives.go

@@ -11,6 +11,7 @@ package dnsserver
 // care what plugin above them are doing.
 var Directives = []string{
 	"tls",
+	"reload",
 	"nsid",
 	"root",
 	"bind",

core/plugin/zplugin.go

@@ -26,6 +26,7 @@ import (
 	_ "github.com/coredns/coredns/plugin/nsid"
 	_ "github.com/coredns/coredns/plugin/pprof"
 	_ "github.com/coredns/coredns/plugin/proxy"
+	_ "github.com/coredns/coredns/plugin/reload"
 	_ "github.com/coredns/coredns/plugin/reverse"
 	_ "github.com/coredns/coredns/plugin/rewrite"
 	_ "github.com/coredns/coredns/plugin/root"

plugin.cfg

@@ -20,6 +20,7 @@
 # log:log
 
 tls:tls
+reload:reload
 nsid:nsid
 root:root
 bind:bind

plugin/reload/README.md

@@ -0,0 +1,58 @@
+# reload
+
+## Name
+
+*reload* - allows automatic reload of a changed Corefile
+
+## Description
+
+This plugin periodically checks if the Corefile has changed by reading
+it and calculating its MD5 checksum. If the file has changed, it reloads
+CoreDNS with the new Corefile. This eliminates the need to send a SIGHUP
+or SIGUSR1 after changing the Corefile.
+
+The reloads are graceful - you should not see any loss of service when the
+reload happens. Even if the new Corefile has an error, CoreDNS will continue
+to run the old config and an error message will be printed to the log.
+
+In some environments (for example, Kubernetes), there may be many CoreDNS 
+instances that started very near the same time and all share a common
+Corefile. To prevent these all from reloading at the same time, some
+jitter is added to the reload check interval. This is jitter from the
+perspective of multiple CoreDNS instances; each instance still checks on a
+regular interval, but all of these instances will have their reloads spread
+out across the jitter duration. This isn't strictly necessary given that the
+reloads are graceful, and can be disabled by setting the jitter to `0s`.
+
+Jitter is re-calculated whenever the Corefile is reloaded.
+
+## Syntax
+
+~~~ txt
+reload [INTERVAL] [JITTER]
+~~~
+
+* The plugin will check for changes every **INTERVAL**, subject to +/- the **JITTER** duration
+* **INTERVAL** and **JITTER** are Golang (durations)[https://golang.org/pkg/time/#ParseDuration]
+* Default **INTERVAL** is 30s, default **JITTER** is 15s
+* If **JITTER** is more than half of **INTERVAL**, it will be set to half of **INTERVAL**
+
+## Examples
+
+Check with the default intervals:
+
+~~~ corefile
+. {
+    reload
+    erratic
+}
+~~~
+
+Check every 10 seconds (jitter is automatically set to 10 / 2 = 5 in this case):
+
+~~~ corefile
+. {
+    reload 10s
+    erratic
+}
+~~~

plugin/reload/reload.go

@@ -0,0 +1,65 @@
+package reload
+
+import (
+	"crypto/md5"
+	"log"
+	"time"
+
+	"github.com/mholt/caddy"
+)
+
+// reload periodically checks if the Corefile has changed, and reloads if so
+
+type reload struct {
+	instance *caddy.Instance
+	interval time.Duration
+	sum      [md5.Size]byte
+	stopped  bool
+	quit     chan bool
+}
+
+func hook(event caddy.EventName, info interface{}) error {
+	if event != caddy.InstanceStartupEvent {
+		return nil
+	}
+
+	// if reload is removed from the Corefile, then the hook
+	// is still registered but setup is never called again
+	// so we need a flag to tell us not to reload
+	if r.stopped {
+		return nil
+	}
+
+	// this should be an instance. ok to panic if not
+	r.instance = info.(*caddy.Instance)
+	r.sum = md5.Sum(r.instance.Caddyfile().Body())
+
+	go func() {
+		tick := time.NewTicker(r.interval)
+
+		for {
+			select {
+			case <-tick.C:
+				corefile, err := caddy.LoadCaddyfile(r.instance.Caddyfile().ServerType())
+				if err != nil {
+					continue
+				}
+				s := md5.Sum(corefile.Body())
+				if s != r.sum {
+					_, err := r.instance.Restart(corefile)
+					if err != nil {
+						log.Printf("[ERROR] Corefile changed but reload failed: %s\n", err)
+						continue
+					}
+					// we are done, this hook gets called again with new instance
+					r.stopped = true
+					return
+				}
+			case <-r.quit:
+				return
+			}
+		}
+	}()
+
+	return nil
+}

plugin/reload/setup.go

@@ -0,0 +1,72 @@
+package reload
+
+import (
+	"math/rand"
+	"sync"
+	"time"
+
+	"github.com/coredns/coredns/plugin"
+
+	"github.com/mholt/caddy"
+)
+
+func init() {
+	caddy.RegisterPlugin("reload", caddy.Plugin{
+		ServerType: "dns",
+		Action:     setup,
+	})
+}
+
+var r *reload
+var once sync.Once
+
+func setup(c *caddy.Controller) error {
+	c.Next() // 'reload'
+	args := c.RemainingArgs()
+
+	if len(args) > 2 {
+		return plugin.Error("reload", c.ArgErr())
+	}
+
+	i := defaultInterval
+	if len(args) > 0 {
+		d, err := time.ParseDuration(args[0])
+		if err != nil {
+			return err
+		}
+		i = d
+	}
+
+	j := defaultJitter
+	if len(args) > 1 {
+		d, err := time.ParseDuration(args[1])
+		if err != nil {
+			return err
+		}
+		j = d
+	}
+
+	if j > i/2 {
+		j = i / 2
+	}
+
+	jitter := time.Duration(rand.Int63n(j.Nanoseconds()) - (j.Nanoseconds() / 2))
+	i = i + jitter
+
+	r = &reload{interval: i, quit: make(chan bool)}
+	once.Do(func() {
+		caddy.RegisterEventHook("reload", hook)
+	})
+
+	c.OnFinalShutdown(func() error {
+		r.quit <- true
+		return nil
+	})
+
+	return nil
+}
+
+const (
+	defaultInterval = 30 * time.Second
+	defaultJitter   = 15 * time.Second
+)

plugin/reload/setup_test.go

@@ -0,0 +1,39 @@
+package reload
+
+import (
+	"testing"
+
+	"github.com/mholt/caddy"
+)
+
+func TestSetupReload(t *testing.T) {
+	c := caddy.NewTestController("dns", `reload`)
+	if err := setup(c); err != nil {
+		t.Fatalf("Expected no errors, but got: %v", err)
+	}
+
+	c = caddy.NewTestController("dns", `reload 10s`)
+	if err := setup(c); err != nil {
+		t.Fatalf("Expected no errors, but got: %v", err)
+	}
+
+	c = caddy.NewTestController("dns", `reload 10s 2s`)
+	if err := setup(c); err != nil {
+		t.Fatalf("Expected no errors, but got: %v", err)
+	}
+
+	c = caddy.NewTestController("dns", `reload foo`)
+	if err := setup(c); err == nil {
+		t.Fatalf("Expected errors, but got: %v", err)
+	}
+
+	c = caddy.NewTestController("dns", `reload 10s foo`)
+	if err := setup(c); err == nil {
+		t.Fatalf("Expected errors, but got: %v", err)
+	}
+
+	c = caddy.NewTestController("dns", `reload 10s 5s foo`)
+	if err := setup(c); err == nil {
+		t.Fatalf("Expected errors, but got: %v", err)
+	}
+}