Add ZMQ connection retry configuration

Signed-off-by: zhengkezhou1 <madzhou1@gmail.com>

Author: zhengkezhou1

pkg/common/config.go

@@ -126,6 +126,8 @@ type Configuration struct {
 
 	// ZMQEndpoint is the ZMQ address to publish events, the default value is tcp://localhost:5557
 	ZMQEndpoint string `yaml:"zmq-endpoint"`
+	// ZMQRetriesTimes defines the maximum number of retries when ZMQ connection fails
+	ZMQRetriesTimes uint `yaml:"zmq-retries-times"`
 	// EventBatchSize is the maximum number of kv-cache events to be sent together, defaults to 16
 	EventBatchSize int `yaml:"event-batch-size"`
 
@@ -354,6 +356,9 @@ func (c *Configuration) validate() error {
 	if c.EventBatchSize < 1 {
 		return errors.New("event batch size cannot less than 1")
 	}
+	if c.ZMQRetriesTimes > 10 {
+		return errors.New("zmq retries times cannot be more than 10")
+	}
 
 	if c.FakeMetrics != nil {
 		if c.FakeMetrics.RunningRequests < 0 || c.FakeMetrics.WaitingRequests < 0 {
@@ -415,6 +420,7 @@ func ParseCommandParamsAndLoadConfig() (*Configuration, error) {
 	f.StringVar(&config.TokenizersCacheDir, "tokenizers-cache-dir", config.TokenizersCacheDir, "Directory for caching tokenizers")
 	f.StringVar(&config.HashSeed, "hash-seed", config.HashSeed, "Seed for hash generation (if not set, is read from PYTHONHASHSEED environment variable)")
 	f.StringVar(&config.ZMQEndpoint, "zmq-endpoint", config.ZMQEndpoint, "ZMQ address to publish events")
+	f.UintVar(&config.ZMQRetriesTimes, "zmq-retries-times", config.ZMQRetriesTimes, "Number of times to retry ZMQ requests")
 	f.IntVar(&config.EventBatchSize, "event-batch-size", config.EventBatchSize, "Maximum number of kv-cache events to be sent together")
 
 	// These values were manually parsed above in getParamValueFromArgs, we leave this in order to get these flags in --help

pkg/common/publisher.go

@@ -23,6 +23,7 @@ import (
 	"errors"
 	"fmt"
 	"sync/atomic"
+	"time"
 
 	zmq "github.com/pebbe/zmq4"
 	"github.com/vmihailenco/msgpack/v5"
@@ -38,24 +39,33 @@ type Publisher struct {
 
 // NewPublisher creates a new ZMQ publisher.
 // endpoint is the ZMQ address to bind to (e.g., "tcp://*:5557").
-func NewPublisher(endpoint string) (*Publisher, error) {
+func NewPublisher(endpoint string, retries uint) (*Publisher, error) {
 	socket, err := zmq.NewSocket(zmq.PUB)
 	if err != nil {
 		return nil, fmt.Errorf("failed to create ZMQ PUB socket: %w", err)
 	}
 
-	if err := socket.Connect(endpoint); err != nil {
-		errClose := socket.Close()
-		return nil, errors.Join(
-			fmt.Errorf("failed to connect to %s: %w", endpoint, err),
-			errClose,
-		)
+	// Retry connection with specified retry times and intervals
+	for i := uint(0); i <= retries; i++ {
+		err = socket.Connect(endpoint)
+		if err == nil {
+			return &Publisher{
+				socket:   socket,
+				endpoint: endpoint,
+			}, nil
+		}
+
+		// If not the last attempt, wait before retrying
+		if i < retries {
+			time.Sleep(1 * time.Second)
+		}
 	}
 
-	return &Publisher{
-		socket:   socket,
-		endpoint: endpoint,
-	}, nil
+	errClose := socket.Close()
+	return nil, errors.Join(
+		fmt.Errorf("failed to connect to %s after %d retries: %w", endpoint, retries+1, err),
+		errClose,
+	)
 }
 
 // PublishEvent publishes a KV cache event batch to the ZMQ topic.

pkg/common/publisher_test.go

@@ -32,6 +32,7 @@ const (
 	topic    = "test-topic"
 	endpoint = "tcp://localhost:5557"
 	data     = "Hello"
+	retries  = 0
 )
 
 var _ = Describe("Publisher", func() {
@@ -49,7 +50,7 @@ var _ = Describe("Publisher", func() {
 
 		time.Sleep(100 * time.Millisecond)
 
-		pub, err := NewPublisher(endpoint)
+		pub, err := NewPublisher(endpoint, retries)
 		Expect(err).NotTo(HaveOccurred())
 
 		ctx, cancel := context.WithCancel(context.Background())

pkg/kv-cache/block_cache.go

@@ -48,7 +48,7 @@ func newBlockCache(config *common.Configuration, logger logr.Logger) (*blockCach
 	// TODO read size of channel from config
 	eChan := make(chan EventData, 10000)
 
-	publisher, err := common.NewPublisher(config.ZMQEndpoint)
+	publisher, err := common.NewPublisher(config.ZMQEndpoint, config.ZMQRetriesTimes)
 	if err != nil {
 		return nil, err
 	}

pkg/kv-cache/kv_cache_test.go

@@ -200,11 +200,12 @@ var _ = Describe("KV cache", Ordered, func() {
 				time.Sleep(300 * time.Millisecond)
 
 				config := &common.Configuration{
-					Port:           1234,
-					Model:          "model",
-					KVCacheSize:    test.cacheSize,
-					ZMQEndpoint:    endpoint,
-					EventBatchSize: 1,
+					Port:            1234,
+					Model:           "model",
+					KVCacheSize:     test.cacheSize,
+					ZMQEndpoint:     endpoint,
+					ZMQRetriesTimes: 3,
+					EventBatchSize:  1,
 				}
 
 				sub, topic := createSub(config)
@@ -303,10 +304,11 @@ var _ = Describe("KV cache", Ordered, func() {
 
 		It("should send events correctly", func() {
 			config := &common.Configuration{
-				Port:        1234,
-				Model:       "model",
-				KVCacheSize: 4,
-				ZMQEndpoint: endpoint,
+				Port:            1234,
+				Model:           "model",
+				KVCacheSize:     4,
+				ZMQEndpoint:     endpoint,
+				ZMQRetriesTimes: 3,
 			}
 
 			sub, topic := createSub(config)
@@ -412,10 +414,11 @@ var _ = Describe("KV cache", Ordered, func() {
 		for _, testCase := range testCases {
 			It(testCase.name, func() {
 				config := common.Configuration{
-					Port:        1234,
-					Model:       "model",
-					KVCacheSize: testCase.cacheSize,
-					ZMQEndpoint: endpoint,
+					Port:            1234,
+					Model:           "model",
+					KVCacheSize:     testCase.cacheSize,
+					ZMQEndpoint:     endpoint,
+					ZMQRetriesTimes: 3,
 				}
 				blockCache, err := newBlockCache(&config, GinkgoLogr)
 				Expect(err).NotTo(HaveOccurred())