chore: Adding cherry picked commits to release-v0.33.x (#979)

Co-authored-by: Todd Neal <tnealt@amazon.com>

hack/toolchain.sh

@@ -16,7 +16,7 @@ tools() {
     go install github.com/mikefarah/yq/v4@latest
     go install github.com/norwoodj/helm-docs/cmd/helm-docs@latest
     go install sigs.k8s.io/controller-runtime/tools/setup-envtest@latest
-    go install sigs.k8s.io/controller-tools/cmd/controller-gen@latest
+    go install sigs.k8s.io/controller-tools/cmd/controller-gen@v0.13.0
     go install github.com/sigstore/cosign/cmd/cosign@latest
     go install -tags extended github.com/gohugoio/hugo@v0.110.0
     go install golang.org/x/vuln/cmd/govulncheck@latest

pkg/controllers/provisioning/scheduling/topology.go

@@ -22,6 +22,7 @@ import (
 	"sigs.k8s.io/karpenter/pkg/controllers/state"
 	"sigs.k8s.io/karpenter/pkg/scheduling"
 	"sigs.k8s.io/karpenter/pkg/utils/functional"
+	"sigs.k8s.io/karpenter/pkg/utils/pretty"
 
 	"k8s.io/apimachinery/pkg/api/errors"
 
@@ -162,7 +163,7 @@ func (t *Topology) AddRequirements(podRequirements, nodeRequirements scheduling.
 		}
 		domains := topology.Get(p, podDomains, nodeDomains)
 		if domains.Len() == 0 {
-			return nil, fmt.Errorf("unsatisfiable topology constraint for %s, key=%s (counts = %v, podDomains = %v, nodeDomains = %v)", topology.Type, topology.Key, topology.domains, podDomains, nodeDomains)
+			return nil, fmt.Errorf("unsatisfiable topology constraint for %s, key=%s (counts = %s, podDomains = %v, nodeDomains = %v", topology.Type, topology.Key, pretty.Map(topology.domains, 5), podDomains, nodeDomains)
 		}
 		requirements.Add(domains)
 	}

pkg/controllers/provisioning/scheduling/topologygroup.go

@@ -61,8 +61,9 @@ type TopologyGroup struct {
 	selector   *metav1.LabelSelector
 	nodeFilter TopologyNodeFilter
 	// Index
-	owners  map[types.UID]struct{} // Pods that have this topology as a scheduling rule
-	domains map[string]int32       // TODO(ellistarn) explore replacing with a minheap
+	owners       map[types.UID]struct{} // Pods that have this topology as a scheduling rule
+	domains      map[string]int32       // TODO(ellistarn) explore replacing with a minheap
+	emptyDomains sets.Set[string]       // domains for which we know that no pod exists
 }
 
 func NewTopologyGroup(topologyType TopologyType, topologyKey string, pod *v1.Pod, namespaces sets.Set[string], labelSelector *metav1.LabelSelector, maxSkew int32, minDomains *int32, domains sets.Set[string]) *TopologyGroup {
@@ -76,15 +77,16 @@ func NewTopologyGroup(topologyType TopologyType, topologyKey string, pod *v1.Pod
 		nodeSelector = MakeTopologyNodeFilter(pod)
 	}
 	return &TopologyGroup{
-		Type:       topologyType,
-		Key:        topologyKey,
-		namespaces: namespaces,
-		selector:   labelSelector,
-		nodeFilter: nodeSelector,
-		maxSkew:    maxSkew,
-		domains:    domainCounts,
-		owners:     map[types.UID]struct{}{},
-		minDomains: minDomains,
+		Type:         topologyType,
+		Key:          topologyKey,
+		namespaces:   namespaces,
+		selector:     labelSelector,
+		nodeFilter:   nodeSelector,
+		maxSkew:      maxSkew,
+		domains:      domainCounts,
+		emptyDomains: domains.Clone(),
+		owners:       map[types.UID]struct{}{},
+		minDomains:   minDomains,
 	}
 }
 
@@ -104,6 +106,7 @@ func (t *TopologyGroup) Get(pod *v1.Pod, podDomains, nodeDomains *scheduling.Req
 func (t *TopologyGroup) Record(domains ...string) {
 	for _, domain := range domains {
 		t.domains[domain]++
+		t.emptyDomains.Delete(domain)
 	}
 }
 
@@ -118,6 +121,7 @@ func (t *TopologyGroup) Register(domains ...string) {
 	for _, domain := range domains {
 		if _, ok := t.domains[domain]; !ok {
 			t.domains[domain] = 0
+			t.emptyDomains.Insert(domain)
 		}
 	}
 }
@@ -245,6 +249,14 @@ func (t *TopologyGroup) nextDomainAffinity(pod *v1.Pod, podDomains *scheduling.R
 
 func (t *TopologyGroup) nextDomainAntiAffinity(domains *scheduling.Requirement) *scheduling.Requirement {
 	options := scheduling.NewRequirement(domains.Key, v1.NodeSelectorOpDoesNotExist)
+	// pods with anti-affinity must schedule to a domain where there are currently none of those pods (an empty
+	// domain). If there are none of those domains, then the pod can't schedule and we don't need to walk this
+	// list of domains.  The use case where this optimization is really great is when we are launching nodes for
+	// a deployment of pods with self anti-affinity.  The domains map here continues to grow, and we continue to
+	// fully scan it each iteration.
+	if len(t.emptyDomains) == 0 {
+		return options
+	}
 	for domain := range t.domains {
 		if domains.Has(domain) && t.domains[domain] == 0 {
 			options.Insert(domain)

pkg/utils/pretty/pretty.go

@@ -15,6 +15,7 @@ limitations under the License.
 package pretty
 
 import (
+	"bytes"
 	"encoding/json"
 	"fmt"
 	"strings"
@@ -43,3 +44,19 @@ func Slice[T any](s []T, maxItems int) string {
 	}
 	return sb.String()
 }
+
+// Map truncates a map after a certain number of max items to ensure that the
+// description in a log doesn't get too long
+func Map[K comparable, V any](values map[K]V, maxItems int) string {
+	var buf bytes.Buffer
+	for k, v := range values {
+		fmt.Fprintf(&buf, "%v: %v ", k, v)
+		if buf.Len() > maxItems {
+			break
+		}
+	}
+	if maxItems < buf.Len() {
+		fmt.Fprintf(&buf, "and %d other(s)", buf.Len()-maxItems)
+	}
+	return buf.String()
+}