From d45a4bb0a3c39b20a4c7feabd9bd0d1b8cb6a9a7 Mon Sep 17 00:00:00 2001
From: David Vossel <davidvossel@gmail.com>
Date: Tue, 10 Apr 2018 13:27:53 -0400
Subject: [PATCH] Ensure monitor loop exits early if pod shutdown is invoked
 and no pid is found

This helps some of the inconsistencies we're seeing in func tests. The qemu
pid would get created and destroyed so fast that the monitor polling loop
would never even see the pid at all. In this case, SIGTERM signal was being
ignored.

Signed-off-by: David Vossel <davidvossel@gmail.com>
---
 pkg/virt-launcher/monitor.go      | 27 ++++++++++++++++++---------
 pkg/virt-launcher/monitor_test.go | 24 ++++++++++++++++++++++++
 2 files changed, 42 insertions(+), 9 deletions(-)

diff --git a/pkg/virt-launcher/monitor.go b/pkg/virt-launcher/monitor.go
index 395fcc6736b5..840450a4a11e 100644
--- a/pkg/virt-launcher/monitor.go
+++ b/pkg/virt-launcher/monitor.go
@@ -179,6 +179,16 @@ func NewProcessMonitor(commandPrefix string,
 	}
 }
 
+func (mon *monitor) isGracePeriodExpired() bool {
+	if mon.gracePeriodStartTime != 0 {
+		now := time.Now().UTC().Unix()
+		if (now - mon.gracePeriodStartTime) > int64(mon.gracePeriod) {
+			return true
+		}
+	}
+	return false
+}
+
 func (mon *monitor) refresh() {
 	if mon.isDone {
 		log.Log.Error("Called refresh after done!")
@@ -187,6 +197,8 @@ func (mon *monitor) refresh() {
 
 	log.Log.Debugf("Refreshing. CommandPrefix %s pid %d", mon.commandPrefix, mon.pid)
 
+	expired := mon.isGracePeriodExpired()
+
 	// is the process there?
 	if mon.pid == 0 {
 		var err error
@@ -200,6 +212,9 @@ func (mon *monitor) refresh() {
 			if mon.timeout > 0 && elapsed >= mon.timeout {
 				log.Log.Infof("%s not found after timeout", mon.commandPrefix)
 				mon.isDone = true
+			} else if expired {
+				log.Log.Infof("%s not found after grace period expired", mon.commandPrefix)
+				mon.isDone = true
 			}
 			return
 		}
@@ -219,12 +234,9 @@ func (mon *monitor) refresh() {
 		return
 	}
 
-	if mon.gracePeriodStartTime != 0 {
-		now := time.Now().UTC().Unix()
-		if (now - mon.gracePeriodStartTime) > int64(mon.gracePeriod) {
-			log.Log.Infof("Grace Period expired, shutting down.")
-			mon.shutdownCallback(mon.pid)
-		}
+	if expired {
+		log.Log.Infof("Grace Period expired, shutting down.")
+		mon.shutdownCallback(mon.pid)
 	}
 
 	return
@@ -252,9 +264,6 @@ func (mon *monitor) monitorLoop(startTimeout time.Duration, signalChan chan os.S
 			mon.refresh()
 		case s := <-signalChan:
 			log.Log.Infof("Received signal %d.", s)
-			if mon.pid == 0 {
-				continue
-			}
 
 			if mon.gracePeriodStartTime != 0 {
 				continue
diff --git a/pkg/virt-launcher/monitor_test.go b/pkg/virt-launcher/monitor_test.go
index 78fb0980b91b..bf76abd7a5a8 100644
--- a/pkg/virt-launcher/monitor_test.go
+++ b/pkg/virt-launcher/monitor_test.go
@@ -145,6 +145,30 @@ var _ = Describe("VirtLauncher", func() {
 				Expect(exited).To(Equal(true))
 			})
 
+			It("verify monitor loop exits when signal arrives and no pid is present", func() {
+				signalChannel := make(chan os.Signal, 1)
+				done := make(chan string)
+
+				go func() {
+					mon.monitorLoop(1*time.Second, signalChannel)
+					done <- "exit"
+				}()
+
+				time.Sleep(time.Second)
+
+				signalChannel <- syscall.SIGQUIT
+				noExitCheck := time.After(5 * time.Second)
+				exited := false
+
+				select {
+				case <-noExitCheck:
+				case <-done:
+					exited = true
+				}
+
+				Expect(exited).To(Equal(true))
+			})
+
 			It("verify graceful shutdown trigger works", func() {
 				signalChannel := make(chan os.Signal, 1)
 				done := make(chan string)