diff --git a/app/other/router/monitor.go b/app/other/router/monitor.go
index 9c088992..ed48ea7d 100644
--- a/app/other/router/monitor.go
+++ b/app/other/router/monitor.go
@@ -1,23 +1,63 @@
 package router
 
 import (
+	"context"
 	"net/http"
+	"time"
 
 	"github.com/gin-gonic/gin"
 	"github.com/go-admin-team/go-admin-core/v2/tools/transfer"
 	"github.com/prometheus/client_golang/prometheus/promhttp"
+
+	"go-admin/common/health"
 )
 
 func init() {
 	routerNoCheckRole = append(routerNoCheckRole, registerMonitorRouter)
 }
 
-// 需认证的路由代码
+// readyTimeout bounds the whole probe. It has to stay under whatever period
+// the orchestrator polls on, or a slow dependency turns a readiness check into
+// a queue of readiness checks.
+const readyTimeout = 2 * time.Second
+
+// 无需认证的路由代码
 func registerMonitorRouter(v1 *gin.RouterGroup) {
 	v1.GET("/metrics", transfer.Handler(promhttp.Handler()))
-	//健康检查
+
+	// 健康检查（存活）
+	//
+	// Stays a bare 200 on purpose. This is the answer to "should I restart
+	// you", and a process whose database is unreachable does not want
+	// restarting - that turns one outage into a crash loop and throws away the
+	// connection pool, the cache and every in-flight request along the way.
 	v1.GET("/health", func(c *gin.Context) {
 		c.Status(http.StatusOK)
 	})
 
-}
\ No newline at end of file
+	// 就绪检查
+	//
+	// The answer to "should I send you requests". It fails while a dependency
+	// is unreachable, and from the moment shutdown begins - which is before
+	// the server stops accepting, so a load balancer can take this instance
+	// out of the pool while it can still finish what it has.
+	v1.GET("/ready", func(c *gin.Context) {
+		if health.Draining() {
+			c.JSON(http.StatusServiceUnavailable, gin.H{
+				"status": "draining",
+				"checks": []health.Check{},
+			})
+			return
+		}
+		ctx, cancel := context.WithTimeout(c.Request.Context(), readyTimeout)
+		defer cancel()
+
+		checks := health.Ready(ctx)
+		status := http.StatusOK
+		if !health.Healthy(checks) {
+			status = http.StatusServiceUnavailable
+		}
+		c.JSON(status, gin.H{"status": http.StatusText(status), "checks": checks})
+	})
+
+}
diff --git a/cmd/api/server.go b/cmd/api/server.go
index 2237ac76..334a6c27 100644
--- a/cmd/api/server.go
+++ b/cmd/api/server.go
@@ -30,6 +30,7 @@ import (
 	"go-admin/app/jobs"
 	"go-admin/common/database"
 	"go-admin/common/global"
+	"go-admin/common/health"
 	common "go-admin/common/middleware"
 	"go-admin/common/middleware/handler"
 	"go-admin/common/storage"
@@ -217,6 +218,12 @@ func run() error {
 	// and the queue adapter, and re-registering consumers - on top of cleanup
 	// that has already run.
 	sdk.Runtime.BeginShutdown()
+	// Readiness fails from here, which is before the server stops accepting.
+	// The order is the whole point: a load balancer that is told "not ready"
+	// while this instance can still finish what it has in flight takes it out
+	// of the pool without dropping anything. Reversed, the connections are cut
+	// first and the health check reports it afterwards.
+	health.BeginDraining()
 
 	log.Info("Shutdown Server ... ")
 	if err := shutdownServer(srv, shutdownTimeout); err != nil {
diff --git a/common/health/health.go b/common/health/health.go
new file mode 100644
index 00000000..79c8ec01
--- /dev/null
+++ b/common/health/health.go
@@ -0,0 +1,133 @@
+// Package health answers whether this process should be sent traffic.
+//
+// The two questions an orchestrator asks are not the same one, and go-admin
+// answers them at two endpoints:
+//
+//   - /health is liveness: is the process there at all. It stays a bare 200,
+//     because the honest answer to "should I restart you" is almost always no.
+//     Restarting a process because its database is unreachable turns one
+//     outage into a crash loop that also loses the connection pool, the cache
+//     and every in-flight request.
+//   - /ready is readiness: should this instance receive requests now. It fails
+//     while the dependencies are unreachable, and - the part that only exists
+//     because of the life-cycle phases - it fails as soon as shutdown begins,
+//     before the server stops accepting, so a load balancer has a chance to
+//     take the instance out before connections are cut.
+package health
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"sync/atomic"
+	"time"
+
+	"github.com/go-admin-team/go-admin-core/v2/sdk"
+)
+
+// draining is set when the process starts shutting down.
+//
+// It is kept here rather than read back from core: BeginShutdown sets a flag on
+// the Application, but nothing exports it, and one host wanting to know is not
+// yet a reason to widen that interface.
+var draining atomic.Bool
+
+// BeginDraining records that shutdown has started, so readiness fails from now
+// on. It is called with BeginShutdown, before anything is taken apart.
+func BeginDraining() { draining.Store(true) }
+
+// Draining reports whether shutdown has begun.
+func Draining() bool { return draining.Load() }
+
+// Check is one dependency and what asking it produced.
+type Check struct {
+	Name string `json:"name"`
+	OK   bool   `json:"ok"`
+	Err  string `json:"error,omitempty"`
+}
+
+// Ready asks every dependency this process cannot serve a request without.
+//
+// The queue is deliberately absent. Nothing on AdapterQueue answers "are you
+// reachable" without publishing something, the memory backend cannot fail, and
+// a queue that is down degrades logging rather than stopping requests - which
+// is a reason to alert, not a reason to leave the load balancer pool.
+func Ready(ctx context.Context) []Check {
+	return []Check{
+		safely("database", func() error { return pingDB(ctx) }),
+		safely("cache", probeCache),
+	}
+}
+
+// safely turns a panic into a failed check.
+//
+// Not defensive habit: the accessors hand back wrappers, not the resources.
+// sdk.Runtime.GetCacheAdapter builds a runtime.Cache around whatever is
+// configured and returns it even when nothing is - so the value is not nil, the
+// cache inside it is, and the first call dereferences it. A nil check cannot
+// see that, and the same is true of GetQueueAdapter.
+//
+// Whatever the reason, a probe is the last thing that should be able to take
+// the process down: the caller is asking whether this instance is well, and
+// killing it to answer is the wrong reply.
+func safely(name string, fn func() error) (c Check) {
+	c = Check{Name: name}
+	defer func() {
+		if r := recover(); r != nil {
+			c.OK, c.Err = false, fmt.Sprintf("the check panicked: %v", r)
+		}
+	}()
+	if err := fn(); err != nil {
+		c.Err = err.Error()
+		return c
+	}
+	c.OK = true
+	return c
+}
+
+// Healthy reports whether every check passed.
+func Healthy(checks []Check) bool {
+	for _, c := range checks {
+		if !c.OK {
+			return false
+		}
+	}
+	return true
+}
+
+func pingDB(ctx context.Context) error {
+	db := sdk.Runtime.GetDb()
+	if db == nil {
+		return errors.New("no database configured")
+	}
+	sqlDB, err := db.DB()
+	if err != nil {
+		return err
+	}
+	return sqlDB.PingContext(ctx)
+}
+
+// cacheProbeKey is written and read back rather than only read: a cache that
+// answers "miss" for every key is indistinguishable from a healthy one on a
+// read alone, and that is precisely the failure - a client pointed at the wrong
+// server - this is meant to catch.
+const cacheProbeKey = "go-admin:health"
+
+func probeCache() error {
+	adapter := sdk.Runtime.GetCacheAdapter()
+	if adapter == nil {
+		return errors.New("no cache configured")
+	}
+	want := time.Now().Format(time.RFC3339Nano)
+	if err := adapter.Set(cacheProbeKey, want, 30); err != nil {
+		return err
+	}
+	got, err := adapter.Get(cacheProbeKey)
+	if err != nil {
+		return err
+	}
+	if got != want {
+		return errors.New("the cache returned a different value than was written")
+	}
+	return nil
+}
diff --git a/common/health/health_test.go b/common/health/health_test.go
new file mode 100644
index 00000000..c3d7a64c
--- /dev/null
+++ b/common/health/health_test.go
@@ -0,0 +1,141 @@
+package health
+
+import (
+	"context"
+	"errors"
+	"testing"
+	"time"
+
+	"github.com/go-admin-team/go-admin-core/v2/sdk"
+	"github.com/go-admin-team/go-admin-core/v2/sdk/runtime"
+	corestorage "github.com/go-admin-team/go-admin-core/v2/storage"
+)
+
+func freshRuntime(t *testing.T) {
+	t.Helper()
+	previous := sdk.Runtime
+	t.Cleanup(func() { sdk.Runtime = previous })
+	sdk.Runtime = runtime.NewConfig()
+}
+
+// fakeCache answers whatever the test needs it to.
+type fakeCache struct {
+	setErr  error
+	getErr  error
+	getBack string // returned instead of what was written, when non-empty
+	stored  string
+}
+
+func (c *fakeCache) String() string { return "fake" }
+func (c *fakeCache) Set(_ string, val interface{}, _ int) error {
+	if c.setErr != nil {
+		return c.setErr
+	}
+	c.stored, _ = val.(string)
+	return nil
+}
+func (c *fakeCache) Get(string) (string, error) {
+	if c.getErr != nil {
+		return "", c.getErr
+	}
+	if c.getBack != "" {
+		return c.getBack, nil
+	}
+	return c.stored, nil
+}
+func (c *fakeCache) Del(string) error                    { return nil }
+func (c *fakeCache) HashGet(_, _ string) (string, error) { return "", nil }
+func (c *fakeCache) HashDel(_, _ string) error           { return nil }
+func (c *fakeCache) Increase(string) error               { return nil }
+func (c *fakeCache) Decrease(string) error               { return nil }
+func (c *fakeCache) Expire(string, time.Duration) error  { return nil }
+
+var _ corestorage.AdapterCache = (*fakeCache)(nil)
+
+func named(checks []Check, name string) Check {
+	for _, c := range checks {
+		if c.Name == name {
+			return c
+		}
+	}
+	return Check{Name: name, Err: "check not reported at all"}
+}
+
+// A cache that accepts writes and answers every read with a different value is
+// the failure this probe exists for - a client pointed at the wrong server, or
+// one that silently drops everything. A read alone cannot tell that apart from
+// a healthy cache with a cold key, which is why the probe writes first.
+func TestCacheProbeFailsWhenTheValueDoesNotComeBack(t *testing.T) {
+	freshRuntime(t)
+	sdk.Runtime.SetCacheAdapter(&fakeCache{getBack: "something else"})
+
+	got := named(Ready(context.Background()), "cache")
+	if got.OK {
+		t.Error("the cache check passed although the value written was not the value read back")
+	}
+	if got.Err == "" {
+		t.Error("the failing check reported no reason")
+	}
+}
+
+func TestCacheProbePassesWhenTheValueComesBack(t *testing.T) {
+	freshRuntime(t)
+	sdk.Runtime.SetCacheAdapter(&fakeCache{})
+
+	if got := named(Ready(context.Background()), "cache"); !got.OK {
+		t.Errorf("the cache check failed for a cache that works: %s", got.Err)
+	}
+}
+
+func TestCacheProbeReportsAWriteFailure(t *testing.T) {
+	freshRuntime(t)
+	sdk.Runtime.SetCacheAdapter(&fakeCache{setErr: errors.New("connection refused")})
+
+	got := named(Ready(context.Background()), "cache")
+	if got.OK {
+		t.Error("the cache check passed although the write failed")
+	}
+}
+
+// Nothing configured is the case that used to take the process down rather
+// than answer. GetCacheAdapter builds a wrapper around whatever is configured
+// and returns it even when nothing is, so the value is not nil, the cache
+// inside it is, and Set dereferences it - a probe that panics is the worst
+// possible answer to "are you well".
+//
+// Every check has to be reported, passing or not. A probe that omits what it
+// could not reach reads as a shorter list of healthy things.
+func TestEveryDependencyIsReportedEvenWithNothingConfigured(t *testing.T) {
+	freshRuntime(t)
+
+	checks := Ready(context.Background())
+	for _, name := range []string{"database", "cache"} {
+		c := named(checks, name)
+		if c.Err == "check not reported at all" {
+			t.Errorf("%s was not reported", name)
+		}
+		if c.OK {
+			t.Errorf("%s passed with nothing configured", name)
+		}
+	}
+	if Healthy(checks) {
+		t.Error("Healthy said yes for a process with no database and no cache")
+	}
+}
+
+// Draining is what makes the shutdown graceful from the outside: it has to be
+// observable before the server stops accepting, or the load balancer learns
+// about the shutdown by having its connections cut.
+func TestDrainingIsObservableOnceItBegins(t *testing.T) {
+	previous := draining.Load()
+	t.Cleanup(func() { draining.Store(previous) })
+
+	draining.Store(false)
+	if Draining() {
+		t.Fatal("Draining reported true before shutdown began")
+	}
+	BeginDraining()
+	if !Draining() {
+		t.Error("Draining still reported false after BeginDraining")
+	}
+}
diff --git a/common/middleware/settings.go b/common/middleware/settings.go
index 248a640f..e1eac74e 100644
--- a/common/middleware/settings.go
+++ b/common/middleware/settings.go
@@ -34,6 +34,7 @@ var CasbinExclude = []UrlInfo{
 	{Url: "/api/v1/user/pwd", Method: "PUT"},
 	{Url: "/api/v1/metrics", Method: "GET"},
 	{Url: "/api/v1/health", Method: "GET"},
+	{Url: "/api/v1/ready", Method: "GET"},
 	{Url: "/", Method: "GET"},
 	{Url: "/api/v1/server-monitor", Method: "GET"},
 	{Url: "/api/v1/public/uploadFile", Method: "POST"},