From 60dacecb39a44e5e9e914d1498843763a08e1980 Mon Sep 17 00:00:00 2001
From: Ryan Breen <rbreen@vistaprint.com>
Date: Wed, 17 Dec 2014 21:44:12 -0500
Subject: [PATCH] Add a randomized start before running CheckMonitors.

---
 command/agent/check.go | 24 +++++++++++++++++++++++-
 1 file changed, 23 insertions(+), 1 deletion(-)

diff --git a/command/agent/check.go b/command/agent/check.go
index 180f6db13a..82b8c84f61 100644
--- a/command/agent/check.go
+++ b/command/agent/check.go
@@ -5,6 +5,7 @@ import (
 	"github.com/armon/circbuf"
 	"github.com/hashicorp/consul/consul/structs"
 	"log"
+	"math/rand"
 	"os/exec"
 	"sync"
 	"syscall"
@@ -91,9 +92,30 @@ func (c *CheckMonitor) Stop() {
 	}
 }
 
+// getInitialPauseTime returns the random duration we should wait before starting this CheckMonitor,
+// preventing potentially large numbers of checks from firing concurrently by staggering their starts.
+func (c *CheckMonitor) getInitialPauseTime() time.Duration {
+	var initialPauseTime time.Duration
+	intervalSeconds := int(c.Interval.Seconds())
+	if intervalSeconds > 0 {
+		// If the check interval is greater than 500ms, as it will be in all real-world cases due to the
+		// application of MinInterval, start after some random number of seconds between 0 and c.Interval
+		initialPauseTime = time.Duration(rand.Intn(intervalSeconds)) * time.Second
+	} else {
+		// Test cases may use sub-second intervals.  In this case, return 0 as the pause duration.
+		initialPauseTime = time.Duration(0)
+	}
+	return initialPauseTime
+}
+
 // run is invoked by a goroutine to run until Stop() is called
 func (c *CheckMonitor) run() {
-	next := time.After(0)
+
+	// Get the randomized initial pause time
+	initialPauseTime := c.getInitialPauseTime()
+
+	c.Logger.Printf("[DEBUG] agent: pausing %ds before first invocation of %s", initialPauseTime, c.Script)
+	next := time.After(initialPauseTime)
 	for {
 		select {
 		case <-next: