import { Memcache } from 'memcache';// Create a new clientconst client = new Memcache();// Set a valueawait client.set('mykey', 'Hello, Memcache!');// Get a valueconst value = await client.get('mykey');console.log(value); // ['Hello, Memcache!']// Delete a valueawait client.delete('mykey');// Close the connectionawait client.quit();
You can also just pass in the uri into the constructor
// Single node as stringconst client = new Memcache('localhost:11211');// Single node with protocolconst client = new Memcache('memcache://192.168.1.100:11211');// Multiple nodes with optionsconst client = new Memcache({ nodes: ['localhost:11211', 'server2:11211'], timeout: 10000});
You can specify multiple Memcache nodes by passing an array of connection strings:
import { Memcache } from 'memcache';// Create a client with multiple nodesconst client = new Memcache({ nodes: ['localhost:11211', '192.168.1.100:11211', 'memcache://192.168.1.101:11211']});// Set and get values (automatically distributed across nodes)await client.set('mykey', 'Hello, Memcache!');const value = await client.get('mykey');console.log(value); // ['Hello, Memcache!']// Close the connectionawait client.quit();
You can also pass an array of MemcacheNode instances for advanced configuration:
import { Memcache, createNode } from 'memcache';// Create nodes with custom settingsconst node1 = createNode('localhost', 11211, { weight: 2 });const node2 = createNode('192.168.1.100', 11211, { weight: 1 });const node3 = createNode('192.168.1.101', 11211, { weight: 1 });// Create a client with MemcacheNode instancesconst client = new Memcache({ nodes: [node1, node2, node3], timeout: 10000});// node1 will receive twice as much traffic due to higher weightawait client.set('mykey', 'Hello, Memcache!');const value = await client.get('mykey');console.log(value); // ['Hello, Memcache!']// Close the connectionawait client.quit();
API
Constructor
new Memcache(options?: string | MemcacheOptions)
Creates a new Memcache client instance. You can pass either:
A string representing a single node URI (uses default settings)
A MemcacheOptions object for custom configuration
Examples:
// Single node as stringconst client = new Memcache('localhost:11211');// Single node with protocolconst client = new Memcache('memcache://192.168.1.100:11211');// Multiple nodes with optionsconst client = new Memcache({ nodes: ['localhost:11211', 'server2:11211'], timeout: 10000});
Options
nodes?: (string | MemcacheNode)[] - Array of node URIs or MemcacheNode instances
timeout?: number - Milliseconds to wait for a connection to open, and for the server to respond while commands are pending (default: 5000). If pending commands get no response in time, the node emits timeout, the pending commands fail and the connection is closed; the next command reconnects. Idle connections stay open.
maxPendingCommands?: number - The most requests each node keeps waiting for a response or for its connection to open (default: 0, no limit). While a node has that many, a new request to it fails at once instead of waiting, so a slow or unreachable server can't build up an unbounded backlog. Refused requests are not retried. On an open connection a refused request fails like any other command: set() and the other writes resolve false, and get() resolves undefined. While the connection is being opened it fails the way a failed connection does: get(), set() and the other single-key commands reject, and gets() leaves out that node's keys. quit() still goes out, after the requests ahead of it. Called directly, node.command() and the node's binary* methods reject with a "Too many pending commands" error.
keepAliveDelay?: number - Keep alive delay in milliseconds (default: 1000)
hash?: HashProvider - Hash provider for consistent hashing (default: KetamaHash)
retries?: number - Number of retry attempts for failed commands (default: 0)
retryDelay?: number - Base delay in milliseconds between retries (default: 100)
retryBackoff?: RetryBackoffFunction - Function to calculate backoff delay (default: fixed delay)
retryOnlyIdempotent?: boolean - Only retry commands marked as idempotent (default: true)
sasl?: SASLCredentials - SASL PLAIN credentials for all nodes (see SASL Authentication)
lazyConnect?: boolean - When true, nodes will not connect until the first command is executed. When false, nodes connect eagerly during construction (default: true)
maxKeySize?: number - Maximum allowed key size in characters (default: 250, memcache protocol max)
maxValueSize?: number - Maximum allowed value size in bytes (default: 1048576, memcached default)
maxExpiration?: number - Maximum allowed expiration in seconds (default: 2592000, memcached's 30-day relative-time boundary). Values above this throw. 0 (no expiration) is always allowed. Raise this if you need to pass absolute Unix timestamps as expirations.
hashLargeKey?: boolean | Hashery - When true, keys longer than maxKeySize are deterministically hashed via hashery (djb2 sync by default) before being sent to the server, instead of throwing. Pass a configured Hashery instance (e.g. new Hashery({ defaultAlgorithmSync: 'fnv1' })) to choose a different sync algorithm or plug in custom providers. When false, oversized keys throw a validation error (default: false). Note: hashing is one-way and can collide; two distinct long keys could map to the same hashed key.
autoDiscover?: AutoDiscoverOptions - AWS ElastiCache Auto Discovery configuration (see Auto Discovery)
tls?: boolean | tls.ConnectionOptions - Enable TLS for all node connections. true uses Node's default trust store (enough for publicly-trusted server certs such as AWS ElastiCache Serverless); an object is passed through to tls.connect() for custom CAs, client certs, or SNI overrides (see TLS Support)
Properties
nodes: MemcacheNode[] (readonly)
Returns the list of all MemcacheNode instances in the cluster.
nodeIds: string[] (readonly)
Returns the list of node IDs (e.g., ["localhost:11211", "127.0.0.1:11212"]).
hash: HashProvider
Get or set the hash provider used for consistent hashing distribution.
timeout: number
Get or set the timeout in milliseconds (default: 5000). Setting it updates existing nodes and the Auto Discovery connection, and applies to commands that are already waiting.
maxPendingCommands: number
Get or set the most requests each node keeps waiting for a response or for its connection to open (default: 0, no limit). The value is rounded down, and anything below 1, or not a finite number, means no limit. Setting it updates existing nodes and applies to the next request; requests already waiting are not affected.
keepAlive: boolean
Get or set the keepAlive setting. Updates all existing nodes. Requires reconnect() to apply changes.
keepAliveDelay: number
Get or set the keep alive delay in milliseconds. Updates all existing nodes. Requires reconnect() to apply changes.
retries: number
Get or set the number of retry attempts for failed commands (default: 0).
retryDelay: number
Get or set the base delay in milliseconds between retry attempts (default: 100).
retryBackoff: RetryBackoffFunction
Get or set the backoff function for calculating retry delays.
retryOnlyIdempotent: boolean
Get or set whether retries are restricted to idempotent commands only (default: true).
maxKeySize: number
Get or set the maximum allowed key size in characters (default: 250). Memcache protocol max is 250.
maxValueSize: number
Get or set the maximum allowed value size in bytes (default: 1048576). Writes (set, add, replace, append, prepend, cas) throw when the encoded value exceeds this limit. Raise it if your memcached server is started with a larger -I item size.
maxExpiration: number
Get or set the maximum allowed expiration in seconds (default: 2592000). Writes that accept an expiration (set, add, replace, cas, touch) throw when exptime exceeds this limit. 0 (no expiration) is always allowed. Memcached treats any exptime greater than 2592000 as an absolute Unix timestamp, so the default guards against accidentally setting a TTL that memcached interprets as "already expired." Raise this if you need to pass Unix timestamps.
hashLargeKey: boolean
Get or set whether keys exceeding maxKeySize are hashed instead of throwing (default: false). When enabled, oversized keys are replaced with a short, deterministic hex digest (via the hashery library, djb2 by default) before being sent to memcache, so any string length is accepted. The same input always produces the same hashed key, but distinct long keys can collide. To change algorithm or providers, configure the hashery property.
Get or set the Hashery instance used to hash oversized keys when hashLargeKey is enabled. Always returns an instance, even when hashing is disabled, so you can pre-configure it (algorithm, custom providers, caching) before flipping hashLargeKey on. Hashery is re-exported from this package for convenience.
import Memcache, { Hashery } from 'memcache';// Simple — defaults to djb2 sync hashingconst simple = new Memcache({ hashLargeKey: true });// Advanced — supply a Hashery preconfigured for fnv1const advanced = new Memcache({ hashLargeKey: new Hashery({ defaultAlgorithmSync: 'fnv1' }),});// Or mutate the instance after constructionsimple.hashery.defaultAlgorithmSync = 'murmur';
lazyConnect: boolean (readonly)
Whether nodes defer connecting until the first command is executed (default: true).
Connection Management
connect(nodeId?: string): Promise<void>
Connect to all Memcache servers or a specific node.
disconnect(): Promise<void>
Disconnect all connections.
reconnect(): Promise<void>
Reconnect all nodes by disconnecting and connecting them again.
quit(): Promise<void>
Quit all connections gracefully.
isConnected(): boolean
Check if any node is connected to a Memcache server.
Node Management
getNodes(): MemcacheNode[]
Get an array of all MemcacheNode instances.
getNode(id: string): MemcacheNode | undefined
Get a specific node by its ID (e.g., "localhost:11211").
Get the nodes for a given key using consistent hashing. Automatically connects to nodes if not already connected. With maxPendingCommands set, a call waiting for a node's connection counts toward that node's limit and, past it, rejects at once.
Get statistics from all Memcache servers. Returns a Map of node IDs to their stats.
version(): Promise<Map<string, string>>
Get the Memcache server version from all nodes. Returns a Map of node IDs to version strings.
Validation
validateKey(key: string): void
Validates a Memcache key according to protocol requirements. Throws error if:
Key is empty
Key exceeds 250 characters
Key contains spaces, newlines, or null characters
resolveKey(key: string): string
Returns the key that will actually be sent to memcache. When hashLargeKey is true and the key length exceeds maxKeySize, returns a short hex digest produced by the configured hashery instance (djb2 by default — 8 hex chars). Otherwise returns the original key unchanged. Called automatically before validateKey in every command, so calling it manually is only needed when you want to inspect the on-wire key.
The Memcache client extends Hookified to provide powerful hooks and events for monitoring and customizing behavior.
Events
The client emits various events during operations that you can listen to:
const client = new Memcache();// Connection eventsclient.on('connect', () => { console.log('Connected to Memcache server');});client.on('close', () => { console.log('Connection closed');});client.on('error', (error) => { console.error('Error:', error);});client.on('timeout', (nodeId) => { console.log(`No response from ${nodeId} in time`);});// Cache hit/miss eventsclient.on('hit', (key, value) => { console.log(`Cache hit for key: ${key}`);});client.on('miss', (key) => { console.log(`Cache miss for key: ${key}`);});
Available Events
connect - Emitted when connection to Memcache server is established
close - Emitted when connection is closed
error - Emitted when an error occurs
timeout - Emitted when a connection can't be opened within timeout, or pending commands get no response within timeout (the connection is then closed)
hit - Emitted when a key is found in cache (includes key and value)
miss - Emitted when a key is not found in cache
quit - Emitted when quit command is sent
warn - Emitted for warning messages
info - Emitted for informational messages
autoDiscover - Emitted on initial auto discovery with the cluster config
autoDiscoverUpdate - Emitted when auto discovery detects a topology change
autoDiscoverError - Emitted when auto discovery encounters an error
Hooks
Hooks allow you to intercept and modify behavior before and after operations. Every operation supports before and after hooks.
const client = new Memcache();// Add a before hook for get operationsclient.onHook('before:get', async ({ key }) => { console.log(`Getting key: ${key}`);});// Add an after hook for set operationsclient.onHook('after:set', async ({ key, value, success }) => { if (success) { console.log(`Successfully set ${key}`); }});// Hooks can be async and modify behaviorclient.onHook('before:set', async ({ key, value }) => { console.log(`About to set ${key} = ${value}`); // Perform validation, logging, etc.});
Available Hooks
All operations support before and after hooks with specific parameters:
get(key)
before:get - { key }
after:get - { key, value } (value is array or undefined)
const client = new Memcache();// Log all get operationsclient.onHook('before:get', async ({ key }) => { console.log(`[GET] Fetching key: ${key}`);});client.onHook('after:get', async ({ key, value }) => { console.log(`[GET] Key: ${key}, Found: ${value !== undefined}`);});// Log all set operations with timingclient.onHook('before:set', async (context) => { context.startTime = Date.now();});client.onHook('after:set', async (context) => { const duration = Date.now() - context.startTime; console.log(`[SET] Key: ${context.key}, Success: ${context.success}, Time: ${duration}ms`);});
Distribution Algorithms
Memcache supports pluggable distribution algorithms to determine how keys are distributed across nodes. You can configure the algorithm using the hash option.
KetamaHash (Default)
KetamaHash uses the Ketama consistent hashing algorithm, which minimizes key redistribution when nodes are added or removed. This is the default and recommended algorithm for production environments with dynamic scaling.
import { Memcache } from 'memcache';// KetamaHash is used by defaultconst client = new Memcache({ nodes: ['server1:11211', 'server2:11211', 'server3:11211']});
Characteristics:
Minimal key redistribution (~1/n keys move when adding/removing nodes)
Uses virtual nodes for better distribution
Supports weighted nodes
Best for production environments with dynamic scaling
ModulaHash
ModulaHash uses a simple modulo-based hashing algorithm (hash(key) % nodeCount). This is a simpler algorithm that may redistribute all keys when nodes change.
import { Memcache, ModulaHash } from 'memcache';// Use ModulaHash for distributionconst client = new Memcache({ nodes: ['server1:11211', 'server2:11211', 'server3:11211'], hash: new ModulaHash()});// With a custom hash algorithm (default is sha1)const client2 = new Memcache({ nodes: ['server1:11211', 'server2:11211'], hash: new ModulaHash('md5')});
Characteristics:
Simple and fast algorithm
All keys may be redistributed when nodes are added or removed
Supports weighted nodes (nodes with higher weight appear more in the distribution)
Best for fixed-size clusters or testing environments
Weighted Nodes with ModulaHash
ModulaHash supports weighted nodes, where nodes with higher weights receive proportionally more keys:
import { Memcache, ModulaHash, createNode } from 'memcache';// Create nodes with different weightsconst node1 = createNode('server1', 11211, { weight: 3 }); // 3x trafficconst node2 = createNode('server2', 11211, { weight: 1 }); // 1x trafficconst client = new Memcache({ nodes: [node1, node2], hash: new ModulaHash()});// server1 will receive approximately 75% of keys// server2 will receive approximately 25% of keys
BroadcastHash
BroadcastHash sends every operation to all nodes in the cluster. Instead of partitioning keys across nodes, every getNodesByKey() call returns all nodes, so reads and writes are broadcast to every server.
import { Memcache, BroadcastHash } from 'memcache';// Use BroadcastHash for full replicationconst client = new Memcache({ nodes: ['server1:11211', 'server2:11211', 'server3:11211'], hash: new BroadcastHash()});// Every set/get/delete hits all three nodesawait client.set('mykey', 'Hello!');
Characteristics:
Every operation targets all nodes
No key partitioning — all nodes hold the same data
Reads return the first successful result from any node
Writes succeed only if all nodes succeed
Best for replication, broadcast invalidation, or small clusters where all nodes should be in sync
Choosing an Algorithm
Feature
KetamaHash
ModulaHash
BroadcastHash
Key redistribution on node change
Minimal (~1/n keys)
All keys may move
N/A (all nodes always)
Complexity
Higher (virtual nodes)
Lower (simple modulo)
Simplest
Performance
Slightly slower
Faster
Depends on node count
Best for
Dynamic scaling
Fixed clusters
Replication
Weighted nodes
Yes
Yes
No
Use KetamaHash (default) when:
Your cluster size may change dynamically
You want to minimize cache invalidation during scaling
You're running in production
Use ModulaHash when:
Your cluster size is fixed
You prefer simplicity over minimal redistribution
You're in a testing or development environment
Use BroadcastHash when:
You want all nodes to hold the same data
You need broadcast cache invalidation across all nodes
You're running a small cluster where replication is more important than partitioning
Retry Configuration
The Memcache client supports automatic retry of failed commands with configurable backoff strategies.
Basic Retry Setup
Enable retries by setting the retries option:
import { Memcache } from 'memcache';const client = new Memcache({ nodes: ['localhost:11211'], retries: 3, // Retry up to 3 times retryDelay: 100 // 100ms between retries});
You can also modify retry settings at runtime:
client.retries = 5;client.retryDelay = 200;
Backoff Strategies
The client includes two built-in backoff functions:
Important: By default, retries are only performed for commands explicitly marked as idempotent. This prevents accidental double-execution of non-idempotent operations like incr, decr, append, and prepend.
Why This Matters
If a network timeout occurs after the server applies a mutation but before the client receives the response, retrying would apply the mutation twice:
Counter incremented twice instead of once
Data appended twice instead of once
Safe Usage Patterns
For read operations (always safe to retry):
// Mark read operations as idempotentawait client.execute('get mykey', nodes, { idempotent: true });
For idempotent writes (safe to retry):
// SET with the same value is idempotentawait client.execute('set mykey 0 0 5\r\nhello', nodes, { idempotent: true });
Disable safety for all commands (use with caution):
const client = new Memcache({ retries: 3, retryOnlyIdempotent: false // Allow retries for ALL commands});
Behavior Summary
retryOnlyIdempotent
idempotent flag
Retries enabled?
true (default)
false (default)
No
true (default)
true
Yes
false
(any)
Yes
Methods Without Retry Support
The following methods do not use the retry mechanism and have their own error handling:
get() - Returns undefined on failure
gets() - Returns partial results on node failure
flush() - Operates directly on nodes
stats() - Operates directly on nodes
version() - Operates directly on nodes
To use retries with read operations, use the execute() method directly:
The Memcache client supports SASL (Simple Authentication and Security Layer) authentication using the PLAIN mechanism. This allows you to connect to memcached servers that require authentication.
Enabling SASL Authentication
import { Memcache } from 'memcache';const client = new Memcache({ nodes: ['localhost:11211'], sasl: { username: 'myuser', password: 'mypassword', },});await client.connect();// Client is now authenticated and ready to use
SASL Options
The sasl option accepts an object with the following properties:
username: string - The username for authentication (required)
password: string - The password for authentication (required)
mechanism?: 'PLAIN' - The SASL mechanism to use (default: 'PLAIN')
Currently, only the PLAIN mechanism is supported.
Binary Protocol Methods
Important: Memcached servers with SASL enabled (-S flag) require the binary protocol for all operations after authentication. The standard text-based methods (client.get(), client.set(), etc.) will not work on SASL-enabled servers.
Use the binary* methods on nodes for SASL-enabled servers:
import { Memcache } from 'memcache';const client = new Memcache({ nodes: ['localhost:11211'], sasl: { username: 'user', password: 'pass' },});await client.connect();// Access the node directly for binary operationsconst node = client.nodes[0];// Binary protocol operationsawait node.binarySet('mykey', 'myvalue', 3600); // Set with 1 hour expiryconst value = await node.binaryGet('mykey'); // Get valueawait node.binaryDelete('mykey'); // Delete key// Other binary operationsawait node.binaryAdd('newkey', 'value'); // Add (only if not exists)await node.binaryReplace('existingkey', 'newvalue'); // Replace (only if exists)await node.binaryIncr('counter', 1); // Incrementawait node.binaryDecr('counter', 1); // Decrementawait node.binaryAppend('mykey', '-suffix'); // Append to valueawait node.binaryPrepend('mykey', 'prefix-'); // Prepend to valueawait node.binaryTouch('mykey', 7200); // Update expirationawait node.binaryFlush(); // Flush allconst version = await node.binaryVersion(); // Get server versionconst stats = await node.binaryStats(); // Get server stats
The binary* methods can run concurrently on the same node. Requests are pipelined on the node's connection, and each caller gets the response to its own request:
If a response doesn't match the request it should answer, every pending binary request is rejected and the connection is closed instead of returning another request's data.
Per-Node SASL Configuration
You can also configure SASL credentials when creating individual nodes:
import { createNode } from 'memcache';// Create a node with SASL credentialsconst node = createNode('localhost', 11211, { sasl: { username: 'user', password: 'pass' },});// Connect and use binary methodsawait node.connect();await node.binarySet('mykey', 'hello');const value = await node.binaryGet('mykey');
Authentication Events
You can listen for authentication events on both nodes and the client:
The Memcache client supports AWS ElastiCache Auto Discovery, which automatically detects cluster topology changes and adds or removes nodes as needed. When enabled, the client connects to a configuration endpoint, retrieves the current list of cache nodes, and periodically polls for changes.
Enabling Auto Discovery
import { Memcache } from 'memcache';const client = new Memcache({ nodes: [], autoDiscover: { enabled: true, configEndpoint: 'my-cluster.cfg.use1.cache.amazonaws.com:11211', },});await client.connect();// The client automatically discovers and connects to all cluster nodes
If you omit configEndpoint, the first node in the nodes array is used as the configuration endpoint:
The autoDiscover option accepts an object with the following properties:
enabled: boolean - Enable auto discovery of cluster nodes (required)
pollingInterval?: number - How often to poll for topology changes, in milliseconds (default: 60000)
configEndpoint?: string - The configuration endpoint to use for discovery. This is typically the .cfg endpoint from ElastiCache. If not specified, the first node in the nodes array will be used
useLegacyCommand?: boolean - Use the legacy get AmazonElastiCache:cluster command instead of config get cluster (default: false)
Auto Discovery Events
The client emits events during the auto discovery lifecycle:
const client = new Memcache({ nodes: [], autoDiscover: { enabled: true, configEndpoint: 'my-cluster.cfg.use1.cache.amazonaws.com:11211', },});// Emitted on initial discovery with the full cluster configclient.on('autoDiscover', (config) => { console.log('Discovered nodes:', config.nodes); console.log('Config version:', config.version);});// Emitted when polling detects a topology changeclient.on('autoDiscoverUpdate', (config) => { console.log('Cluster topology changed:', config.nodes);});// Emitted when discovery encounters an error (non-fatal, retries on next poll)client.on('autoDiscoverError', (error) => { console.error('Discovery error:', error.message);});await client.connect();
Legacy Command Support
For ElastiCache engine versions older than 1.4.14, use the legacy discovery command:
const client = new Memcache({ nodes: [], autoDiscover: { enabled: true, configEndpoint: 'my-cluster.cfg.use1.cache.amazonaws.com:11211', useLegacyCommand: true, // Uses 'get AmazonElastiCache:cluster' instead of 'config get cluster' },});
IPv6 Support
The Memcache client fully supports IPv6 addresses using standard bracket notation in URIs.
Connecting to IPv6 Nodes
import { Memcache } from 'memcache';// IPv6 loopbackconst client = new Memcache('[::1]:11211');// Multiple IPv6 nodesconst client = new Memcache({ nodes: [ '[::1]:11211', '[2001:db8::1]:11211', 'memcache://[2001:db8::2]:11212', ],});await client.connect();
IPv6 in Auto Discovery
When auto discovery returns IPv6 node addresses, the client automatically brackets them for correct URI handling:
const client = new Memcache({ nodes: [], autoDiscover: { enabled: true, configEndpoint: '[2001:db8::1]:11211', },});await client.connect();// Discovered IPv6 nodes are added as [host]:port automatically
IPv6 Node IDs
Node IDs for IPv6 addresses use bracket notation to avoid ambiguity:
const client = new Memcache({ nodes: ['[::1]:11211', '[2001:db8::1]:11212'],});console.log(client.nodeIds);// ['[::1]:11211', '[2001:db8::1]:11212']
TLS Support
The client can connect to TLS-enabled memcached servers (memcached ships TLS
since 1.5.13 via -Z; the official Docker image supports it since 1.5.21).
Connection readiness waits for the completed TLS handshake (secureConnect),
so pipelined commands are never written into an unfinished handshake.
Connecting with TLS
Enable TLS for every node with the client-level tls option:
import { Memcache } from 'memcache';// TLS with Node's default trust store (publicly-trusted server certs)const client = new Memcache({ nodes: ['my-cache.example.com:11211'], tls: true,});await client.set('mykey', 'Hello over TLS!');
Or per node with the memcaches:// URI scheme (enables TLS for that node
even when the client-level option is unset):
const client = new Memcache('memcaches://my-cache.example.com:11211');
TLS Options
The tls option accepts:
true / {} — connect using TLS with Node's default trust store
a tls.ConnectionOptions object — passed through to tls.connect() (CA,
client certificates, servername, minVersion, …)
false / undefined (default) — plain TCP
Certificate verification is always on (standard Node behavior). To connect to
a server with a self-signed cert, pass its CA — do not disable verification.
The node's host, port (or Unix socket path), and keep-alive settings always
win over any host / port / path fields in a tls options object.
Per-Node TLS Configuration
You can also enable TLS when creating individual nodes:
node.tlsEnabled — true when TLS is configured for the node
node.tls — the TLS option the node was constructed with
node.uri — memcaches://host:port when TLS is enabled, so passing it
back into addNode() or the constructor keeps TLS on even without a
client-level tls option
The client's connect event fires after the TLS handshake completes.
AWS ElastiCache Serverless
ElastiCache Serverless (Memcached) requires TLS and only speaks the text
protocol — both are this client's defaults, so the only configuration needed
is tls: true (the server presents a publicly-trusted ACM certificate):
const client = new Memcache({ nodes: ['my-cache-xxxxxx.serverless.use1.cache.amazonaws.com:11211'], tls: true,});
Custom certificate authorities
Pass any tls.connect() options — private CAs, client certificates, servername
overrides — as an object:
import { readFileSync } from 'node:fs';const client = new Memcache({ nodes: ['memcached-internal:11211'], tls: { ca: readFileSync('/etc/ssl/private-ca.pem'), // cert, key, servername, ... are passed through to tls.connect() },});
Auto Discovery with TLS
Auto Discovery uses the same client-level tls (and sasl) options as data
nodes, including the configuration-endpoint connection. Set tls: true (or a
CA options object) when the cluster requires in-transit encryption:
A memcaches:// configuration endpoint also enables TLS for discovery. When
discovery returns a DNS hostname plus an IP, the client connects to the IP
(stable node IDs) and sets TLS SNI/servername to the hostname so certificate
verification matches ElastiCache node certs.
TLS and SASL together
TLS and SASL compose: the TCP handshake completes, then SASL PLAIN runs on
the encrypted socket. Use both for ElastiCache clusters that require
in-transit encryption and AUTH. SASL-enabled servers still require the
binary protocol after authentication (see SASL Authentication).
Tables below are auto-generated by pnpm benchmark:readme. Do not edit between the <!-- BENCHMARK:* --> markers.
Set, Get and Delete compares this client with other Node.js memcached clients on single-key and multi-key tasks run one at a time in random order, and the next section compares them under load and on multi-gets. The other tables show how this client behaves under load; in each one, every row does the same total work in a different shape.
All the tables come from one run of pnpm benchmark:readme, with Node.js 22 on Linux, against the benchmark containers' IPs (see Running the benchmarks). The absolute numbers depend on the machine; the comparisons within each table are what carry over.
Set, Get and Delete (one task at a time, in random order)
Each library runs 1,000 tasks that set, get and delete one key, and 1,000 that set, get and delete 10 keys. The tasks of all 3 libraries go into one queue in random order and run one at a time. The queue runs 5 times, shuffled each time, after an untimed warm-up of 100 tasks of each kind per library. Every result is checked.
Total time per library, the median of the 5 runs (lower is better). 🥇 marks the fastest library in each row, and the percentages compare the others with it. The text protocol has no multi-set or multi-delete, so for 10 keys every library sends its 10 sets, and then its 10 deletes, at once. memjs has no multi-get either, so it does the same with its gets. memcached pools up to 10 connections; the other clients use one.
Tasks
memcache (v1.11.0)
memjs (v1.3.2)
memcached (v2.2.2)
1,000 × set, get, delete 1 key
🥇 290 ms
322 ms (+11%)
333 ms (+15%)
1,000 × set, get, delete 10 keys
🥇 424 ms
566 ms (+33%)
789 ms (+86%)
All 2,000 tasks
🥇 713 ms
893 ms (+25%)
1.12 s (+57%)
Compared with memjs and memcached
Each cell is the median of 3 rounds, with the clients taking turns. 🥇 marks the fastest client in each row, and the percentages compare the other clients with it.
Requests per second, 500 requests per run (higher is better). memcached pools up to 10 connections; the other clients use one.
Workload
memcache (v1.11.0)
memjs (v1.3.2)
memcached (v2.2.2)
gets, 1 in flight
🥇 12.3K
11.8K (-3.5%)
10.5K (-14%)
gets, 10 in flight
🥇 86.8K
65.0K (-25%)
44.6K (-49%)
gets, 100 in flight
🥇 285.6K
104.0K (-64%)
59.7K (-79%)
gets, 500 in flight
🥇 298.1K
109.9K (-63%)
60.0K (-80%)
sets, 1 in flight
🥇 13.2K
11.8K (-11%)
10.7K (-19%)
sets, 10 in flight
🥇 106.8K
64.9K (-39%)
35.6K (-67%)
sets, 100 in flight
🥇 328.9K
92.1K (-72%)
42.9K (-87%)
sets, 500 in flight
🥇 272.1K
93.7K (-66%)
46.3K (-83%)
Time to fetch 10,000 keys in batches (lower is better). memjs has no multi-get, so it sends one get per key, all at once.
Batches
memcache (v1.11.0)
memjs (v1.3.2)
memcached (v2.2.2)
100 × 100 keys
🥇 28 ms
93 ms (+230%)
68 ms (+141%)
10 × 1,000 keys
🥇 18 ms
97 ms (+443%)
132 ms (+642%)
1 × 10,000 keys
🥇 21 ms
146 ms (+606%)
2.63 s (+12649%)
Concurrent Requests (500 per operation)
name
summary
ops/sec
time/op
margin
samples
500 sets, 100 in flight
🥇
725
1ms
±1.27%
697
500 gets, 500 in flight
-16%
606
2ms
±1.71%
571
500 sets, 500 in flight
-19%
588
2ms
±1.74%
558
500 gets, 100 in flight
-24%
549
2ms
±1.50%
526
500 sets, 10 in flight
-70%
216
5ms
±2.38%
209
500 gets, 10 in flight
-75%
182
6ms
±1.97%
179
500 sets, 1 in flight
-96%
27
37ms
±1.68%
64
500 gets, 1 in flight
-96%
26
39ms
±1.91%
64
Multi-Get (10,000 keys per operation)
name
summary
ops/sec
time/op
margin
samples
10 × gets() of 1,000 keys
🥇
56
18ms
±1.77%
166
1 × gets() of 10,000 keys
-14%
48
21ms
±2.02%
143
100 × gets() of 100 keys
-30%
39
26ms
±2.39%
116
Large Values (4 MB per operation)
name
summary
ops/sec
time/op
margin
samples
TCP: 4 × 1 MB sets
🥇
157
7ms
±2.15%
154
TCP: 1 × 4 MB sets
-1.6%
155
7ms
±1.68%
153
TCP: 16 × 256 KB sets
-4.2%
151
7ms
±1.84%
149
TLS: 4 × 1 MB sets
-23%
121
8ms
±2.74%
118
TLS: 1 × 4 MB sets
-24%
120
8ms
±1.50%
120
TCP: 1 × 4 MB gets
-29%
112
9ms
±2.23%
110
TCP: 16 × 256 KB gets
-30%
110
10ms
±4.03%
104
TLS: 16 × 256 KB sets
-34%
104
10ms
±3.18%
102
TCP: 4 × 1 MB gets
-35%
102
10ms
±2.67%
100
TLS: 4 × 1 MB gets
-40%
94
11ms
±4.72%
90
TLS: 1 × 4 MB gets
-40%
94
11ms
±5.67%
87
TLS: 16 × 256 KB gets
-54%
73
14ms
±3.92%
72
Bursts (60,000 gets per operation)
name
summary
ops/sec
time/op
margin
samples
6 × 10,000 concurrent gets
🥇
4
268ms
±6.48%
19
2 × 30,000 concurrent gets
-8.3%
4
292ms
±7.15%
18
1 × 60,000 concurrent gets
-8.4%
4
293ms
±7.67%
18
Cold Start
concurrent first requests
sockets opened
50
1
Running the benchmarks
Each table has its own script: pnpm benchmark:set-get, benchmark:compare, benchmark:concurrency, benchmark:multi-get, benchmark:large-values, benchmark:bursts and benchmark:cold-start. pnpm benchmark runs them all, and pnpm benchmark:readme runs them and updates the tables above; pass section ids, such as pnpm benchmark:readme compare, to update only those tables. They run against dedicated memcached containers that the test services don't start:
To target other servers, set MEMCACHE_BENCH_HOST and MEMCACHE_BENCH_PORT, plus MEMCACHE_BENCH_TLS_HOST and MEMCACHE_BENCH_TLS_PORT for the TLS measurements. TLS trusts the test CA by default. For a server with its own certificate, set MEMCACHE_BENCH_TLS_CA to its CA bundle, and MEMCACHE_BENCH_TLS_SERVERNAME if the certificate's name differs from the host. Set MEMCACHE_BENCH_SKIP_TLS=1 to leave TLS out.
On Linux, published Docker ports go through docker-proxy, which adds per-packet overhead. For numbers closer to a real network, use the containers' IPs with port 11211: