Fix cache overshoot (#200)

* count by hostname

* changelog

tools/wasme/changelog/v0.0.28/cache.yaml

@@ -0,0 +1,4 @@
+changelog:
+  - type: FIX
+    description: Fix a bug where the operator miscounted cache events.
+    issueLink: https://github.com/solo-io/wasm/issues/199

tools/wasme/cli/operator/install/wasme/templates/deployment.yaml

@@ -122,8 +122,12 @@ spec:
         - /etc/wasme-cache/images.txt
         - --cache-ns
         - '{{ .Release.Namespace }}'
-{{- if $wasmeCache.env }}
         env:
+        - name: NODE_HOSTNAME
+          valueFrom:
+            fieldRef:
+              fieldPath: spec.nodeName
+{{- if $wasmeCache.env }}
 {{ toYaml $wasmeCache.env | indent 10 }}
 {{- end }}
         volumeMounts:

tools/wasme/cli/pkg/cache/notifying_cache.go

@@ -3,6 +3,7 @@ package cache
 import (
 	"crypto/md5"
 	"fmt"
+	"os"
 
 	"github.com/hashicorp/go-multierror"
 	v1 "k8s.io/api/core/v1"
@@ -56,6 +57,7 @@ func (n *Notifier) Notify(err error, image string) error {
 		Message: message,
 		Source: v1.EventSource{
 			Component: "wasme-cache",
+			Host:      os.Getenv("NODE_HOSTNAME"),
 		},
 	})
 

tools/wasme/cli/pkg/deploy/istio/istio_provider.go

@@ -290,17 +290,17 @@ func (p *Provider) waitForCacheEvents(image string) error {
 				return errors.Wrapf(err, "getting events for image %v", image)
 			}
 			// expect an event for each cache instance
-			var successEvents int32
+			successEvents := map[string]bool{}
 
 			for _, evt := range events {
 				if evt.Reason == cache.Reason_ImageError {
 					logrus.Warnf("event %v was in Error state: %+v", evt.Name, evt)
 					continue
 				}
-				successEvents++
+				successEvents[evt.Source.Host] = true
 			}
 
-			if successEvents != cacheDaemonset.Status.NumberReady {
+			if len(successEvents) != int(cacheDaemonset.Status.NumberReady) {
 				eventsErr = errors.Errorf("expected %v image-ready events for image %v, only found %v", cacheDaemonset.Status.NumberReady, image, successEvents)
 				logrus.Warnf("event err: %v", eventsErr)
 				continue