feat: stamp Queue shard status with availability zones

Resolves each shard topic's replica broker IDs (internal/kafka.Admin.
ReplicaBrokerIDs) to the topology.kubernetes.io/zone labels of the nodes
hosting those brokers (ZoneLocator), and writes the result into
ShardStatus.AvailabilityZones each reconcile. Uses mgr.GetAPIReader()
rather than the cached client for the Pod/Node lookups, since the cached
client would otherwise require cluster-wide list/watch RBAC on Pods just
to serve occasional point Gets.
This commit is contained in:
riotpiaole
2026-06-22 17:30:26 -07:00
parent 1ff1ecd563
commit c7eeed2617
12 changed files with 227 additions and 0 deletions
+7
View File
@@ -121,6 +121,13 @@ type ShardStatus struct {
// CreatedAt timestamps when this shard was created, used to enforce
// ShardSplitCooldownSeconds.
CreatedAt metav1.Time `json:"createdAt,omitempty"`
// AvailabilityZones lists the topology.kubernetes.io/zone values of every
// node currently hosting a Kafka replica of this shard's topic, resolved
// from the broker pods' node placement each reconcile. Empty until the
// first successful resolution (e.g. node lookup failed transiently).
// +optional
AvailabilityZones []string `json:"availabilityZones,omitempty"`
}
// QueueStatus defines the observed state of a Queue.
+10
View File
@@ -67,6 +67,16 @@ func main() {
Admin: admin,
Redis: rdb,
Now: time.Now,
Zones: &operator.ZoneLocator{
// GetAPIReader(), not GetClient(): the latter is cache-backed and
// would make controller-runtime List+Watch all Pods/Nodes
// cluster-wide just to serve occasional point Gets here. A direct
// (uncached) reader needs only "get" RBAC, not "list"/"watch".
Client: mgr.GetAPIReader(),
Namespace: getEnv("KMSVC_NAMESPACE", "sqs"),
ClusterName: getEnv("KMSVC_KAFKA_CLUSTER_NAME", "kmsvc"),
PoolName: getEnv("KMSVC_KAFKA_POOL_NAME", "kmsvc-pool"),
},
}
err = ctrl.NewControllerManagedBy(mgr).
+9
View File
@@ -211,6 +211,15 @@ spec:
description: ShardStatus describes one shard backing a Queue (design.md
§2a/§2c).
properties:
availabilityZones:
description: |-
AvailabilityZones lists the topology.kubernetes.io/zone values of every
node currently hosting a Kafka replica of this shard's topic, resolved
from the broker pods' node placement each reconcile. Empty until the
first successful resolution (e.g. node lookup failed transiently).
items:
type: string
type: array
createdAt:
description: |-
CreatedAt timestamps when this shard was created, used to enforce
+27
View File
@@ -81,6 +81,33 @@ func (a *Admin) DescribeTopic(ctx context.Context, topic string) (partitions int
return int32(len(detail.Partitions)), true, nil
}
// ReplicaBrokerIDs returns the set of broker IDs holding any replica of any
// partition of topic, used by the queue-operator to resolve which
// availability zones (node labels) a shard's data actually lives on
// (design.md §2a AZ-awareness).
func (a *Admin) ReplicaBrokerIDs(ctx context.Context, topic string) ([]int32, error) {
td, err := a.client.ListTopics(ctx, topic)
if err != nil {
return nil, fmt.Errorf("list topics %q: %w", topic, err)
}
detail, found := td[topic]
if !found || detail.Err != nil {
return nil, nil
}
seen := make(map[int32]bool)
var ids []int32
for _, p := range detail.Partitions {
for _, id := range p.Replicas {
if !seen[id] {
seen[id] = true
ids = append(ids, id)
}
}
}
return ids, nil
}
// LogEndOffsetSum returns the sum of the high-watermark offsets across every
// partition of a topic — a monotonically increasing proxy for total records
// produced, used by the queue-operator's shard-split sampler (design.md §2c)
+1
View File
@@ -13,4 +13,5 @@ type TopicAdmin interface {
DeleteTopic(ctx context.Context, topic string) error
LogEndOffsetSum(ctx context.Context, topic string) (int64, error)
ConsumerLag(ctx context.Context, group, topic string) (int64, error)
ReplicaBrokerIDs(ctx context.Context, topic string) ([]int32, error)
}
+9
View File
@@ -51,6 +51,15 @@ func (f *fakeAdmin) ConsumerLag(_ context.Context, group, topic string) (int64,
return f.lag[group+"/"+topic], nil
}
func (f *fakeAdmin) ReplicaBrokerIDs(_ context.Context, topic string) ([]int32, error) {
f.mu.Lock()
defer f.mu.Unlock()
if _, ok := f.topics[topic]; !ok {
return nil, nil
}
return []int32{0}, nil
}
func (f *fakeAdmin) setOffsetSum(topic string, v int64) {
f.mu.Lock()
defer f.mu.Unlock()
+45
View File
@@ -15,6 +15,7 @@ import (
metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
"sigs.k8s.io/controller-runtime/pkg/client"
"sigs.k8s.io/controller-runtime/pkg/controller/controllerutil"
ctrllog "sigs.k8s.io/controller-runtime/pkg/log"
kmsvcv1 "github.com/rockliang/kafka-management-service/apis/kmsvc/v1"
"github.com/rockliang/kafka-management-service/internal/kafka"
@@ -39,6 +40,11 @@ type QueueReconciler struct {
Redis *goredis.Client
Now func() time.Time
// Zones resolves shard topics' broker placement to availability zones
// (design.md §2a AZ-awareness). Nil disables zone annotation entirely --
// tests and any deployment without zone-labeled nodes can leave it unset.
Zones *ZoneLocator
// sampleState tracks the last (offsetSum, time) seen per shard topic, used
// to compute a throughput estimate between reconciles. Keyed by topic name.
sampleState map[string]sample
@@ -90,6 +96,8 @@ func (r *QueueReconciler) Reconcile(ctx context.Context, namespace, name string)
return r.setFailed(ctx, &queue, "EnsureTopics", err)
}
r.annotateAvailabilityZones(ctx, &queue)
if err := r.reconcileSplits(ctx, &queue); err != nil {
return r.setFailed(ctx, &queue, "ShardSplit", err)
}
@@ -139,6 +147,43 @@ func (r *QueueReconciler) ensureShardTopics(ctx context.Context, queue *kmsvcv1.
return nil
}
// annotateAvailabilityZones resolves and stamps each non-closed shard's
// AvailabilityZones (design.md §2a). Best-effort: a resolution failure for
// one shard is logged via setFailed-style swallowing -- it must never block
// the rest of reconciliation, since AZ info is status metadata, not
// load-bearing for the queue's actual operation.
func (r *QueueReconciler) annotateAvailabilityZones(ctx context.Context, queue *kmsvcv1.Queue) {
if r.Zones == nil {
return
}
logger := ctrllog.FromContext(ctx)
for i := range queue.Status.Shards {
s := &queue.Status.Shards[i]
if s.Phase == kmsvcv1.ShardPhaseClosed {
continue
}
brokerIDs, err := r.Admin.ReplicaBrokerIDs(ctx, s.Topic)
if err != nil {
logger.Error(err, "resolve replica broker IDs", "topic", s.Topic)
continue
}
if len(brokerIDs) == 0 {
logger.Info("no replica broker IDs returned", "topic", s.Topic)
continue
}
zones, err := r.Zones.ZonesForBrokers(ctx, brokerIDs)
if err != nil {
logger.Error(err, "resolve zones for brokers", "topic", s.Topic, "brokerIDs", brokerIDs)
continue
}
if len(zones) == 0 {
logger.Info("no zones resolved", "topic", s.Topic, "brokerIDs", brokerIDs)
continue
}
s.AvailabilityZones = zones
}
}
func (r *QueueReconciler) setFailed(ctx context.Context, queue *kmsvcv1.Queue, reason string, cause error) error {
queue.Status.Phase = kmsvcv1.QueuePhaseFailed
if err := r.Client.Status().Update(ctx, queue); err != nil {
+93
View File
@@ -0,0 +1,93 @@
package operator
import (
"context"
"fmt"
"sort"
corev1 "k8s.io/api/core/v1"
apierrors "k8s.io/apimachinery/pkg/api/errors"
"sigs.k8s.io/controller-runtime/pkg/client"
)
// zoneLabel is the standard k8s topology label this cluster's Talos nodes
// carry (confirmed via `kubectl get nodes --show-labels`).
const zoneLabel = "topology.kubernetes.io/zone"
// ZoneLocator resolves which availability zones (node labels) a set of
// Kafka broker IDs currently run in, for AZ-aware Queue status (design.md
// §2a). Broker pods follow Strimzi's StrimziPodSet naming convention:
// "<clusterName>-<poolName>-<brokerID>" (confirmed live: broker id 0 is pod
// kmsvc-kmsvc-pool-0, etc. -- broker ID equals the pod's ordinal suffix).
type ZoneLocator struct {
// Client only needs Get, so callers can pass an uncached client.Reader
// (e.g. mgr.GetAPIReader()) and avoid requiring cluster-wide list/watch
// RBAC on Pods/Nodes just to serve occasional point lookups here.
Client client.Reader
Namespace string
ClusterName string
PoolName string
// nodeZoneCache avoids a Node lookup per broker per reconcile -- node
// zone labels don't change while a node is up, and this is best-effort
// (cleared implicitly by process restart) rather than watched for
// correctness.
nodeZoneCache map[string]string
}
// ZonesForBrokers returns the sorted, de-duplicated set of zones the given
// broker IDs currently run in. A broker whose pod or node can't be resolved
// is skipped rather than failing the whole call -- AZ info is best-effort
// status, not load-bearing for reconciliation.
func (z *ZoneLocator) ZonesForBrokers(ctx context.Context, brokerIDs []int32) ([]string, error) {
if z.nodeZoneCache == nil {
z.nodeZoneCache = make(map[string]string)
}
seen := make(map[string]bool)
var zones []string
for _, id := range brokerIDs {
nodeName, err := z.nodeNameForBroker(ctx, id)
if err != nil || nodeName == "" {
continue
}
zone, err := z.zoneForNode(ctx, nodeName)
if err != nil || zone == "" {
continue
}
if !seen[zone] {
seen[zone] = true
zones = append(zones, zone)
}
}
sort.Strings(zones)
return zones, nil
}
func (z *ZoneLocator) nodeNameForBroker(ctx context.Context, brokerID int32) (string, error) {
podName := fmt.Sprintf("%s-%s-%d", z.ClusterName, z.PoolName, brokerID)
var pod corev1.Pod
if err := z.Client.Get(ctx, client.ObjectKey{Namespace: z.Namespace, Name: podName}, &pod); err != nil {
if apierrors.IsNotFound(err) {
return "", nil
}
return "", fmt.Errorf("get broker pod %s: %w", podName, err)
}
return pod.Spec.NodeName, nil
}
func (z *ZoneLocator) zoneForNode(ctx context.Context, nodeName string) (string, error) {
if zone, ok := z.nodeZoneCache[nodeName]; ok {
return zone, nil
}
var node corev1.Node
if err := z.Client.Get(ctx, client.ObjectKey{Name: nodeName}, &node); err != nil {
if apierrors.IsNotFound(err) {
return "", nil
}
return "", fmt.Errorf("get node %s: %w", nodeName, err)
}
zone := node.Labels[zoneLabel]
z.nodeZoneCache[nodeName] = zone
return zone, nil
}
+9
View File
@@ -211,6 +211,15 @@ spec:
description: ShardStatus describes one shard backing a Queue (design.md
§2a/§2c).
properties:
availabilityZones:
description: |-
AvailabilityZones lists the topology.kubernetes.io/zone values of every
node currently hosting a Kafka replica of this shard's topic, resolved
from the broker pods' node placement each reconcile. Empty until the
first successful resolution (e.g. node lookup failed transiently).
items:
type: string
type: array
createdAt:
description: |-
CreatedAt timestamps when this shard was created, used to enforce
@@ -25,5 +25,13 @@ spec:
value: {{ .Values.redisAddr | quote }}
- name: GOMEMLIMIT
value: {{ .Values.goMemLimit | quote }}
- name: KMSVC_NAMESPACE
valueFrom:
fieldRef:
fieldPath: metadata.namespace
- name: KMSVC_KAFKA_CLUSTER_NAME
value: {{ .Values.kafkaClusterName | quote }}
- name: KMSVC_KAFKA_POOL_NAME
value: {{ .Values.kafkaPoolName | quote }}
resources:
{{- toYaml .Values.resources | nindent 12 }}
+3
View File
@@ -24,6 +24,9 @@ rules:
- apiGroups: [""]
resources: ["events"]
verbs: ["create", "patch"]
- apiGroups: [""]
resources: ["pods", "nodes"]
verbs: ["get"]
---
apiVersion: rbac.authorization.k8s.io/v1
kind: ClusterRoleBinding
+6
View File
@@ -8,6 +8,12 @@ image:
kafkaBrokers: "kmsvc-kafka-bootstrap.sqs.svc.cluster.local:9092"
redisAddr: "kmsvc-redis-master.sqs.svc.cluster.local:6379"
# Must match kafka-cluster chart's clusterName/derived pool name -- used to
# resolve "<kafkaClusterName>-<kafkaPoolName>-<brokerID>" broker pod names
# for AZ-aware Queue status (design.md §2a).
kafkaClusterName: kmsvc
kafkaPoolName: kmsvc-pool
resources:
requests:
cpu: 100m