From 29ff791f08a56ec0542a486456c39944778a6eff Mon Sep 17 00:00:00 2001
From: Lantao Liu <lantaol@google.com>
Date: Fri, 3 Jun 2016 01:48:26 -0700
Subject: [PATCH] Hack for unsupported OS distros.

---
 pkg/kernelmonitor/kernel_log_watcher.go | 16 ++++++++++++++++
 pkg/problemdetector/problem_detector.go |  6 +-----
 2 files changed, 17 insertions(+), 5 deletions(-)

diff --git a/pkg/kernelmonitor/kernel_log_watcher.go b/pkg/kernelmonitor/kernel_log_watcher.go
index 5072b70c..c74f8c8e 100644
--- a/pkg/kernelmonitor/kernel_log_watcher.go
+++ b/pkg/kernelmonitor/kernel_log_watcher.go
@@ -71,6 +71,22 @@ func (k *kernelLogWatcher) Watch() (<-chan *types.KernelLog, error) {
 	if k.cfg.KernelLogPath != "" {
 		path = k.cfg.KernelLogPath
 	}
+	// NOTE(random-liu): This is a hack. KernelMonitor doesn't support some OS distros e.g. GCI. Ideally,
+	// KernelMonitor should only run on nodes with supported OS distro. However, NodeProblemDetector is
+	// running as DaemonSet, it has to be deployed on each node (There is no node affinity support for
+	// DaemonSet now #22205). If some nodes have unsupported OS distro e.g. the OS distro of master node
+	// in gke/gce is GCI, KernelMonitor will keep throwing out error, and NodeProblemDetector will be
+	// restarted again and again.
+	// To avoid this, we decide to add this temporarily hack. When KernelMonitor can't find the kernel
+	// log file, it will print a log and then return nil channel and no error. Since nil channel will
+	// always be blocked, the NodeProblemDetector will block forever.
+	// TODO(random-liu):
+	// 1. Add journald supports to support GCI.
+	// 2. Schedule KernelMonitor only on supported node (with node label and selector)
+	if _, err := os.Stat(path); os.IsNotExist(err) {
+		glog.Infof("kernel log %q is not found, kernel monitor doesn't support the os distro", path)
+		return nil, nil
+	}
 	start, err := k.getStartPoint(path)
 	if err != nil {
 		return nil, err
diff --git a/pkg/problemdetector/problem_detector.go b/pkg/problemdetector/problem_detector.go
index 04ff6e60..f1067d13 100644
--- a/pkg/problemdetector/problem_detector.go
+++ b/pkg/problemdetector/problem_detector.go
@@ -60,11 +60,7 @@ func (p *problemDetector) Run() error {
 	glog.Info("Problem detector started")
 	for {
 		select {
-		case status, ok := <-ch:
-			if !ok {
-				glog.Errorf("Monitor stopped unexpectedly")
-				break
-			}
+		case status := <-ch:
 			for _, event := range status.Events {
 				p.client.Eventf(util.ConvertToAPIEventType(event.Severity), status.Source, event.Reason, event.Message)
 			}