From bdce4288036fc862f4c32cd03e67e95baa08c625 Mon Sep 17 00:00:00 2001
From: Nathan Coad <nathan@thecoads.com>
Date: Mon, 18 Mar 2024 10:05:44 +1100
Subject: [PATCH] use HA unreachable events when trying to find a failure

---
 main.go | 42 +++++++++++++++++++++++++++++++-----------
 1 file changed, 31 insertions(+), 11 deletions(-)

diff --git a/main.go b/main.go
index 13a02be..716ab72 100644
--- a/main.go
+++ b/main.go
@@ -42,14 +42,15 @@ type HostFailureResults struct {
 }
 
 var (
-	c           *govmomi.Client
-	ctx         context.Context
-	cancel      context.CancelFunc
-	location    *time.Location
-	sha1ver     string // sha1 revision used to build the program
-	buildTime   string // when the executable was built
-	results     []OutageResults
-	hostResults []HostFailureResults
+	c                   *govmomi.Client
+	ctx                 context.Context
+	cancel              context.CancelFunc
+	location            *time.Location
+	sha1ver             string // sha1 revision used to build the program
+	buildTime           string // when the executable was built
+	results             []OutageResults
+	hostResults         []HostFailureResults
+	haUnreachableEvents []types.Event
 )
 
 // This function optionally filters events by a single MoRef, any additonal MoRefs are ignored
@@ -269,6 +270,18 @@ func main() {
 		log.Printf("Found at least one host failure, proceeding with VM restart search\n")
 		vmFailures := getEvents([]string{"com.vmware.vc.ha.VmRestartedByHAEvent"}, []types.ManagedObjectReference{}, *begin, *end)
 
+		log.Printf("Searching for ha status change events\n")
+		haStatusChanges := getEvents([]string{"com.vmware.vc.HA.HostStateChangedEvent"}, []types.ManagedObjectReference{}, *begin, *end)
+
+		// filter ha status changed messages for unreachable ones
+		for _, h := range haStatusChanges {
+			unreachableMessage := strings.Contains(strings.ToLower(h.FullFormattedMessage), "changed to unreachable")
+			if unreachableMessage {
+				haUnreachableEvents = append(haUnreachableEvents, h)
+				log.Printf("Host %s unreachable HA status event at %s : '%s'\n", h.Host.Name, h.CreatedTime.In(location).Format(time.ANSIC), h.FullFormattedMessage)
+			}
+		}
+
 		// Sort the host failure events by time
 		sort.Slice(hostFailures[:], func(i, j int) bool {
 			return hostFailures[i].CreatedTime.Before(hostFailures[j].CreatedTime)
@@ -338,13 +351,20 @@ func main() {
 			} else { // Didn't find any VM disconnected events
 				log.Printf("could not determine previous host for this VM. Filtering all host failures for events prior to fuzzy VM restart time '%s'\n", fuzzyTime)
 
-				// Search for host failures
-				for _, hostEvent := range hostFailures {
+				// TODO Use HA unreachable events to find the host
+				for _, hostEvent := range haUnreachableEvents {
 					if hostEvent.CreatedTime.In(location).Before(fuzzyTime) || hostEvent.CreatedTime.In(location).Equal(fuzzyTime) {
 						possibleHosts = append(possibleHosts, hostEvent)
 					}
 				}
-
+				/*
+					// Search for host failures
+					for _, hostEvent := range hostFailures {
+						if hostEvent.CreatedTime.In(location).Before(fuzzyTime) || hostEvent.CreatedTime.In(location).Equal(fuzzyTime) {
+							possibleHosts = append(possibleHosts, hostEvent)
+						}
+					}
+				*/
 				log.Printf("Based on event times there were %d possible hosts this VM was running on\n", len(possibleHosts))
 
 				if len(possibleHosts) == 0 {