node/pprof,metrics: make them stop at the end of app life

Like 85b31b9 but for shutdown order. In
practice, there was an SN app that could not stop, but it was impossible to
debug anything because no pprof or metrics were available. Closes #2976.

Signed-off-by: Pavel Karpy <[email protected]>

CHANGELOG.md

@@ -67,6 +67,7 @@ notes) if you've not done it previously.
 - `ObjectService`'s `Search` and `Replicate` RPC handlers cache up to 1000 lists of container nodes (#2892)
 - Default max_traceable_blocks Morph setting lowered to 17280 from 2102400 (#2897)
 - `ObjectService`'s `Get`/`Head`/`GetRange` RPC handlers cache up to 10K lists of per-object sorted container nodes (#2896)
+- Pprof and metrics services stop at the end of SN's application lifecycle (#2976)
 
 ### Updated
 - neofs-contract dependency to 0.20.0 (#2872)

cmd/neofs-node/config.go

@@ -285,6 +285,9 @@ type internals struct {
 	wg      *sync.WaitGroup
 	workers []worker
 	closers []func()
+	// services that are useful for debug (e.g. when a regular closer does not
+	// close), must be close at the very end of application life cycle
+	veryLastClosers []func()
 
 	apiVersion   version.Version
 	healthStatus atomic.Int32

cmd/neofs-node/main.go

@@ -97,7 +97,7 @@ func preRunAndLog(c *cfg, name string, srv *httputil.Server) {
 		})
 	}()
 
-	c.closers = append(c.closers, func() {
+	c.closers = append(c.veryLastClosers, func() {
 		c.log.Debug(fmt.Sprintf("shutting down %s service", name))
 
 		err := srv.Shutdown()
@@ -185,6 +185,9 @@ func shutdown(c *cfg) {
 	for _, closer := range c.closers {
 		closer()
 	}
+	for _, lastCloser := range c.veryLastClosers {
+		lastCloser()
+	}
 
 	c.log.Debug("waiting for all processes to stop")