[0.13] Limit PerClusterState to less than 1mb

internal/resourcestatus/resourcekey.go

@@ -1,6 +1,9 @@
 package resourcestatus
 
 import (
+	"cmp"
+	"maps"
+	"slices"
 	"sort"
 	"strings"
 
@@ -50,10 +53,10 @@ func clusterID(bd fleet.BundleDeployment) string {
 // fromResources inspects a list of BundleDeployments and returns a list of per-cluster states per resource keys.
 // It also returns a list of errors messages produced that may have occurred during processing
 func fromResources(items []fleet.BundleDeployment) resourceStatesByResourceKey {
+	// ensure cluster order is stable, so we do not trigger reconciles
 	sort.Slice(items, func(i, j int) bool {
 		return clusterID(items[i]) < clusterID(items[j])
 	})
-
 	resources := make(resourceStatesByResourceKey)
 	for _, bd := range items {
 		for key, entry := range bundleDeploymentResources(bd) {
@@ -155,8 +158,18 @@ func bundleDeploymentResources(bd fleet.BundleDeployment) map[fleet.ResourceKey]
 }
 
 func aggregateResourceStatesClustersMap(resourceKeyStates resourceStatesByResourceKey) []fleet.Resource {
+	// sort to make the result stable in case of truncation
+	keys := slices.SortedFunc(maps.Keys(resourceKeyStates), func(a, b fleet.ResourceKey) int {
+		return cmp.Compare(
+			(b.Kind + b.APIVersion + " " + b.Namespace + "/" + b.Name),
+			(a.Kind + a.APIVersion + " " + a.Namespace + "/" + a.Name),
+		)
+	})
+
+	size := 0
 	result := make([]fleet.Resource, 0, len(resourceKeyStates))
-	for resourceKey, entries := range resourceKeyStates {
+	for _, resourceKey := range keys {
+		entries := resourceKeyStates[resourceKey]
 		resource := &fleet.Resource{
 			Kind:       resourceKey.Kind,
 			APIVersion: resourceKey.APIVersion,
@@ -172,7 +185,10 @@ func aggregateResourceStatesClustersMap(resourceKeyStates resourceStatesByResour
 				resource.IncompleteState = true
 			}
 
-			appendToPerClusterState(&resource.PerClusterState, entry.state, entry.clusterID)
+			size += len(entry.clusterID)
+			if size < 1024*1024 { // 1mb
+				appendToPerClusterState(&resource.PerClusterState, entry.state, entry.clusterID)
+			}
 
 			// top-level state is set from first non "Ready" per-cluster state
 			if resource.State == "Ready" {
@@ -183,6 +199,7 @@ func aggregateResourceStatesClustersMap(resourceKeyStates resourceStatesByResour
 		result = append(result, *resource)
 	}
 
+	// sort to make result easier to read
 	sort.Slice(result, func(i, j int) bool {
 		return key(result[i]) < key(result[j])
 	})

internal/resourcestatus/resourcekey_test.go

@@ -1,6 +1,9 @@
 package resourcestatus
 
 import (
+	"encoding/json"
+	"fmt"
+	"slices"
 	"testing"
 
 	"github.com/stretchr/testify/assert"
@@ -245,6 +248,7 @@ func TestPerClusterState(t *testing.T) {
 			},
 		}
 	}
+
 	tests := []struct {
 		name              string
 		bundleDeployments []fleet.BundleDeployment
@@ -305,3 +309,88 @@ func TestPerClusterState(t *testing.T) {
 		})
 	}
 }
+
+func TestPerClusterStateTruncation(t *testing.T) {
+	percluster := func(b, c int) fleet.BundleDeployment {
+		workload := fmt.Sprintf("workload%02d", b)
+		bd := fleet.BundleDeployment{
+			ObjectMeta: metav1.ObjectMeta{
+				Name:      fmt.Sprintf("bundlename%d", b),
+				Namespace: fmt.Sprintf("ns-cluster%d", c),
+				Labels: map[string]string{
+					fleet.ClusterLabel:          fmt.Sprintf("d0-k3k-downstream%04d-downstream%04d", c, c),
+					fleet.ClusterNamespaceLabel: "fleet-default",
+				},
+			},
+			Spec: fleet.BundleDeploymentSpec{
+				DeploymentID: "fakeid",
+			},
+			Status: fleet.BundleDeploymentStatus{
+				AppliedDeploymentID: "fakeid",
+				Resources: []fleet.BundleDeploymentResource{
+					{Kind: "ConfigMap", APIVersion: "v1", Namespace: workload, Name: "cm-web"},
+					{Kind: "Deployment", APIVersion: "v1", Namespace: workload, Name: "web"},
+					{Kind: "Service", APIVersion: "v1", Namespace: workload, Name: "web-svc"},
+				},
+				ModifiedStatus: []fleet.ModifiedStatus{
+					{Kind: "Secret", APIVersion: "v1", Namespace: workload, Name: "cm-creds", Create: true},
+				},
+				NonReadyStatus: []fleet.NonReadyStatus{
+					{Kind: "Deployment", APIVersion: "v1", Namespace: workload, Name: "db", Summary: summary.Summary{State: "NotReady"}},
+				},
+			},
+		}
+		return bd
+	}
+	// we are not comparing the whole struct
+	sizeOf := func(res []fleet.Resource) int {
+		size := 0
+		for _, r := range res {
+			for _, s := range r.PerClusterState.Ready {
+				size = size + len(s)
+			}
+			for _, s := range r.PerClusterState.NotReady {
+				size = size + len(s)
+			}
+			for _, s := range r.PerClusterState.Missing {
+				size = size + len(s)
+			}
+		}
+		return size
+	}
+
+	n := 0
+	maxBundle := 50
+	maxCluster := 800
+	var items = make([]fleet.BundleDeployment, maxBundle*maxCluster)
+	for c := range maxCluster {
+		for b := range maxBundle {
+			items[n] = percluster(b, c)
+			n = n + 1
+		}
+	}
+
+	// different order should produce the same truncation
+	ritems := slices.Clone(items)
+	slices.Reverse(ritems)
+
+	var status fleet.GitRepoStatus
+	SetResources(items, &status.StatusBase)
+
+	assert.Less(t, sizeOf(status.Resources), 1024*1024, "resources should be truncated to be less than 1MB")
+
+	js, err := json.Marshal(status.Resources)
+	assert.NoError(t, err)
+
+	// and the truncation is stable
+	SetResources(items, &status.StatusBase)
+	js2, err := json.Marshal(status.Resources)
+	assert.NoError(t, err)
+	// avoid the long diff from assert.Equal
+	assert.True(t, string(js) == string(js2), "truncation should produce stable json for the same input")
+
+	SetResources(ritems, &status.StatusBase)
+	js2, err = json.Marshal(status.Resources)
+	assert.NoError(t, err)
+	assert.True(t, string(js) == string(js2), "truncation should produce stable json, when items are in a different order")
+}