start a tally of working, totals, not working, etc

Signed-off-by: Jeff Carr <[email protected]>

2 files changed, 30 insertions, 9 deletions

diff --git a/http.go b/http.go
index 85dc520..5bf5300 100644
--- a/http.go
+++ b/http.go
@@ -59,17 +59,18 @@ func okHandler(w http.ResponseWriter, r *http.Request) {
 		return
 	}
 	if tmp == "/uptime" {
-		if clusterHealthy() {
-			log.Info("Handling URL:", tmp, "1 GOOD=true")
-			fmt.Fprintln(w, "GOOD=true")
+		b, s := clusterHealthy()
+		if b {
+			log.Info("Handling URL:", tmp, "cluster is ok", s)
+			fmt.Fprintln(w, s)
 		} else {
-			log.Info("Handling URL:", tmp, "1 GOOD=false")
-			fmt.Fprintln(w, "GOOD=false")
+			log.Info("Handling URL:", tmp, "cluster is not right yet", s)
+			fmt.Fprintln(w, s)
 		}
 		for _, h := range me.hypers {
 			url := "http://" + h.Hostname + ":2520/kill"
 			dur := time.Since(h.lastpoll) // Calculate the elapsed time
-			if dur > 90 * time.Second {
+			if dur > 90*time.Second {
 				log.Info("KILL DAEMON ON", h.Hostname, shell.FormatDuration(dur), "curl", url)
 				// s := shell.Wget(url)
 				// log.Info("curl got:", s)
diff --git a/poll.go b/poll.go
index 2979052..dc9d171 100644
--- a/poll.go
+++ b/poll.go
@@ -1,6 +1,7 @@
 package main
 
 import (
+	"fmt"
 	"strings"
 	"time"
 
@@ -68,8 +69,12 @@ func findDroplet(name string) *DropletT {
 	return nil
 }
 
-func clusterHealthy() bool {
-	var good = true
+// check the state of the cluster and return a string
+// that is intended to be sent to an uptime monitor like Kuma
+func clusterHealthy() (bool, string) {
+	var good bool = true
+	var working int
+	var failed int
 	for _, d := range me.droplets {
 		if d.State != "ON" {
 			continue
@@ -88,13 +93,28 @@ func clusterHealthy() bool {
 				log.Info("GOOD STATE MISSING", d.Hostname, d.hname, shell.FormatDuration(dur))
 				good = false
 				d.CurrentState = "MISSING"
+				failed += 1
+				continue
 			}
 			l := shell.FormatDuration(dur)
 			if l == "" {
 				log.Info("DUR IS EMPTY", dur)
+				continue
 			}
+			working += 1
 			// log.Info("GOOD STATE ON", d.Hostname, d.hname, "dur =", l)
 		}
 	}
-	return good
+	var summary string = "("
+	if working > 0 {
+		summary += fmt.Sprintf("working = %d", working)
+	}
+	if failed > 0 {
+		summary += fmt.Sprintf("failed = %d", failed)
+	}
+	summary += ")"
+	if good {
+		return good, "GOOD=true " + summary
+	}
+	return good, "GOOD=false " + summary
 }