From 4b1953d949e57ce33b12176d5f9b7c64baa83a0e Mon Sep 17 00:00:00 2001
From: Pierre Fersing <pierre.fersing@bleemeo.com>
Date: Mon, 16 Oct 2017 11:55:19 +0200
Subject: [PATCH] Fix CPU system plugin that get stuck after suspend

Signed-off-by: Pierre Fersing <pierre.fersing@bleemeo.com>
---
 CHANGELOG.md                      |  1 +
 plugins/inputs/system/cpu.go      |  5 ++-
 plugins/inputs/system/cpu_test.go | 69 +++++++++++++++++++++++++++++++
 3 files changed, 73 insertions(+), 2 deletions(-)

diff --git a/CHANGELOG.md b/CHANGELOG.md
index efaeb4476a7dc..444e23a72abb5 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -49,6 +49,7 @@
 - [#3136](https://github.com/influxdata/telegraf/issues/3136): Fix webhooks input address in use during reload.
 - [#3258](https://github.com/influxdata/telegraf/issues/3258): Unlock Statsd when stopping to prevent deadlock.
 - [#3319](https://github.com/influxdata/telegraf/issues/3319): Fix cloudwatch output requires unneeded permissions.
+- [#3342](https://github.com/influxdata/telegraf/pull/3342): Fix CPU input plugin stuck after suspend on Linux.
 
 ## v1.4.3 [unreleased]
 
diff --git a/plugins/inputs/system/cpu.go b/plugins/inputs/system/cpu.go
index 55378c93e14a3..99fa451b37e3a 100644
--- a/plugins/inputs/system/cpu.go
+++ b/plugins/inputs/system/cpu.go
@@ -96,7 +96,8 @@ func (s *CPUStats) Gather(acc telegraf.Accumulator) error {
 		totalDelta := total - lastTotal
 
 		if totalDelta < 0 {
-			return fmt.Errorf("Error: current total CPU time is less than previous total CPU time")
+			err = fmt.Errorf("Error: current total CPU time is less than previous total CPU time")
+			break
 		}
 
 		if totalDelta == 0 {
@@ -126,7 +127,7 @@ func (s *CPUStats) Gather(acc telegraf.Accumulator) error {
 		s.lastStats[cts.CPU] = cts
 	}
 
-	return nil
+	return err
 }
 
 func totalCpuTime(t cpu.TimesStat) float64 {
diff --git a/plugins/inputs/system/cpu_test.go b/plugins/inputs/system/cpu_test.go
index fabff8a7d47cb..b6c0c9ef1e61b 100644
--- a/plugins/inputs/system/cpu_test.go
+++ b/plugins/inputs/system/cpu_test.go
@@ -184,3 +184,72 @@ func TestCPUCountIncrease(t *testing.T) {
 	err = cs.Gather(&acc)
 	require.NoError(t, err)
 }
+
+// TestCPUTimesDecrease tests that telegraf continue to works after
+// CPU times decrease, which seems to occur when Linux system is suspended.
+func TestCPUTimesDecrease(t *testing.T) {
+	var mps MockPS
+	defer mps.AssertExpectations(t)
+	var acc testutil.Accumulator
+
+	cts := cpu.TimesStat{
+		CPU:       "cpu0",
+		User:      18,
+		Idle:      80,
+		Iowait:    2,
+	}
+
+	cts2 := cpu.TimesStat{
+		CPU:       "cpu0",
+		User:      38,     // increased by 20
+		Idle:      40,     // decreased by 40
+		Iowait:    1,      // decreased by 1
+	}
+
+	cts3 := cpu.TimesStat{
+		CPU:       "cpu0",
+		User:      56,       // increased by 18
+		Idle:      120,      // increased by 80
+		Iowait:    3,        // increased by 2
+	}
+
+	mps.On("CPUTimes").Return([]cpu.TimesStat{cts}, nil)
+
+	cs := NewCPUStats(&mps)
+
+	cputags := map[string]string{
+		"cpu": "cpu0",
+	}
+
+	err := cs.Gather(&acc)
+	require.NoError(t, err)
+
+	// Computed values are checked with delta > 0 becasue of floating point arithmatic
+	// imprecision
+	assertContainsTaggedFloat(t, &acc, "cpu", "time_user", 18, 0, cputags)
+	assertContainsTaggedFloat(t, &acc, "cpu", "time_idle", 80, 0, cputags)
+	assertContainsTaggedFloat(t, &acc, "cpu", "time_iowait", 2, 0, cputags)
+
+	mps2 := MockPS{}
+	mps2.On("CPUTimes").Return([]cpu.TimesStat{cts2}, nil)
+	cs.ps = &mps2
+
+	// CPU times decreased. An error should be raised
+	err = cs.Gather(&acc)
+	require.Error(t, err)
+
+	mps3 := MockPS{}
+	mps3.On("CPUTimes").Return([]cpu.TimesStat{cts3}, nil)
+	cs.ps = &mps3
+
+	err = cs.Gather(&acc)
+	require.NoError(t, err)
+
+	assertContainsTaggedFloat(t, &acc, "cpu", "time_user", 56, 0, cputags)
+	assertContainsTaggedFloat(t, &acc, "cpu", "time_idle", 120, 0, cputags)
+	assertContainsTaggedFloat(t, &acc, "cpu", "time_iowait", 3, 0, cputags)
+
+	assertContainsTaggedFloat(t, &acc, "cpu", "usage_user", 18, 0.0005, cputags)
+	assertContainsTaggedFloat(t, &acc, "cpu", "usage_idle", 80, 0.0005, cputags)
+	assertContainsTaggedFloat(t, &acc, "cpu", "usage_iowait", 2, 0.0005, cputags)
+}