QUIC

The 'node:quic' module provides an implementation of the QUIC protocol. To access it, start Node.js with the --experimental-quic option and:

import quic from 'node:quic';

The module is only available under the node: scheme.

The quic module provides APIs for creating QUIC clients and servers.

The QUIC and HTTP/3 protocols are defined by a collection of RFCs produced primarily by the IETF QUIC Working Group. A familiarity with these documents is strongly recommended for users of this module.

Core QUIC transport:

• RFC 8999 — Version-Independent Properties of QUIC
• RFC 9000 — QUIC: A UDP-Based Multiplexed and Secure Transport
• RFC 9001 — Using TLS to Secure QUIC
• RFC 9002 — QUIC Loss Detection and Congestion Control

Core HTTP/3:

• RFC 9114 — HTTP/3
• RFC 9204 — QPACK: Field Compression for HTTP/3

QUIC extensions:

• RFC 9221 — An Unreliable Datagram Extension to QUIC
• RFC 9287 — Greasing the QUIC Bit
• RFC 9368 — Compatible Version Negotiation for QUIC
• RFC 9369 — QUIC Version 2
• RFC 9443 — Multiplexing Scheme Updates for QUIC

HTTP/3 extensions:

• RFC 9218 — Extensible Prioritization Scheme for HTTP
• RFC 9220 — Bootstrapping WebSockets with HTTP/3
• RFC 9297 — HTTP Datagrams and the Capsule Protocol
• RFC 9412 — The ORIGIN Extension in HTTP/3

Operational and informational:

• RFC 9308 — Applicability of the QUIC Transport Protocol
• RFC 9312 — Manageability of the QUIC Transport Protocol

The quic module is built around three core abstractions:

• QuicEndpoint: represents the local UDP socket binding for QUIC. It is used to send and receive QUIC packets and can be shared across multiple sessions. A single endpoint can be used as both a client and a server simultaneously.

• QuicSession: represents a QUIC connection between the local endpoint and a remote peer. A session is created either by initiating a connection to a remote peer using quic.connect() or by accepting an incoming connection from a remote peer via quic.listen().

• QuicStream: represents a QUIC stream within a session. Streams are created by either local or remote peers and can be bidirectional or unidirectional.

Unlike traditional TCP-based protocols, QUIC "connections" are not inherently tied to a specific local port / remote port pair. A session is initiated via a QUIC endpoint but may be migrated to a different local or remote address over its lifetime, outlive the endpoint that created it, and may even be associated with multiple endpoints simultaneously. This flexibility allows for advanced use cases such as connection migration, multi-homing, and load balancing. Most often, however, a simple one-to-one relationship between endpoint and session is sufficient.

The QUIC protocol integrates TLS 1.3 directly into the protocol for connection establishment and security. The quic module's API reflects this integration by exposing TLS-related information and configuration options. It is currently not possible to use QUIC without TLS or to use a different version of TLS.

Every QUIC session starts with the client and server performing a TLS handshake to negotiate the application protocol (via ALPN), authenticate the server (and optionally the client), exchange transport parameters, and establish shared keys for encryption.

Every QuicSession is associated with a single application protocol, negotiated via ALPN during the TLS handshake. The quic module is designed to be application-agnostic in general but includes built-in support for HTTP/3 as a specific application protocol. When using HTTP/3, the quic module provides additional APIs for handling HTTP/3-specific features such as headers, trailers, and prioritization. For other application protocols, users can implement their own message framing and multiplexing on top of the core QUIC transport features.

When initiating a TLS handshake, the client will include a list of supported ALPN protocols in the ClientHello. The server selects one of these protocols (if any) and includes it in the ServerHello. The negotiated protocol determines how the QuicSession and QuicStream APIs behave. For example, when the h3 protocol is negotiated for HTTP/3, the QuicSession and QuicStream will support HTTP/3-specific features.

Currently, the quic module only supports HTTP/3 as a built-in application protocol. All other protocols must be implemented by the user on top of the provided JavaScript API.

The QUIC API is designed to be flexible and highly configurable to support a wide range of use cases. Users can configure various aspects of the QUIC transport, TLS handshake, and application behavior via options passed to the quic.connect() and quic.listen() functions, as well as dynamically on QuicEndpoint and QuicSession instances. The API also provides access to detailed statistics and events for monitoring and debugging.

QUIC transport parameters are exchanged during the TLS handshake to negotiate various transport-level settings such as maximum stream counts, idle timeouts, and datagram support. The quic module allows users to configure the transport parameters their endpoint advertises to peers, as well as access the transport parameters advertised by peers. These configure the capabilities and limits of the QUIC connection in coordination with the peer.

A rich set of local settings is also available for configuring the behavior of the local endpoint and sessions. These include settings for connection limits, congestion control, stream prioritization, and more.

The quic module uses a combination of callbacks and promises for asynchronous operations. For example, initiating a connection with quic.connect() returns a promise for the established session, while incoming sessions on the server side are handled via a callback passed to quic.listen(). Within a session, events such as incoming streams, datagrams, and session state changes are handled via callbacks on the QuicSession instance. Promises are used for operations that have a clear completion point, such as completion of the TLS handshake or graceful closure of a session.

All callbacks are invoked synchronously and may either return synchronously or return a promise. If a callback returns a promise that rejects, or throws an error, the object will be destroyed with the error as the reason if an onerror callback is not specified.

Streams are the primary data-carrying abstraction in QUIC. A stream can be initiated by either the local endpoint or the remote peer once a session is established.

Streams can be either bidirectional (data flows in both directions) or unidirectional (data flows in only one direction). The quic module provides separate APIs for creating each kind: session.createBidirectionalStream() and session.createUnidirectionalStream(). Streams initiated by a remote peer are delivered via the session.onstream callback.

There are two ways to write data to a stream:

• Body source — pass a body option when creating the stream (or call stream.setBody()). The body can be a string, ArrayBuffer, ArrayBufferView, Blob, FileHandle, AsyncIterable, sync Iterable, or Promise resolving to any of these. A null body closes the writable side immediately. This is the simplest approach when the data is available up front or can be expressed as an iterable.
• Writer — access stream.writer to push data incrementally. The writer exposes synchronous methods (writeSync(), writevSync(), endSync()) that return immediately, as well as async equivalents (write(), writev(), end()) that wait for drain when backpressured. writeSync() returns false when the write buffer is full; the caller should wait for drain before retrying.

These two approaches are mutually exclusive for a given stream.

Reading is done by iterating the stream as an async iterable. Each iteration yields a batch of Uint8Array chunks:

for await (const chunks of stream) {
  for (const chunk of chunks) {
    // Process each Uint8Array chunk
  }
}

Only one async iterator can be obtained per stream. The stream is also compatible with node:stream/iter utilities such as Stream.bytes(), Stream.text(), and Stream.pipeTo().

In addition to streams, QUIC supports unreliable datagrams (RFC 9221) for use cases that require low-latency, best-effort messaging.

Datagram support is enabled at two levels. At the QUIC transport level, both peers must advertise a non-zero maxDatagramFrameSize transport parameter during the handshake. For HTTP/3 sessions, both peers must additionally set application.enableDatagrams to true, which exchanges the SETTINGS_H3_DATAGRAM setting on the HTTP/3 control stream.

A datagram is sent with a single call to session.sendDatagram(). Each datagram must fit within a single QUIC packet — datagrams cannot be fragmented. The maximum payload size is determined by the peer's maxDatagramFrameSize and the path MTU. If a datagram is too large or the peer does not support datagrams, sendDatagram() returns 0n rather than throwing an error.

There is no guarantee of delivery. Datagrams may be lost, duplicated, or delivered out of order. The session.ondatagramstatus callback reports whether each sent datagram was 'acknowledged', 'lost', or 'abandoned' (never sent on the wire).

QUIC supports 0-RTT early data, allowing a client that has previously connected to a server to send application data with its very first packet, without waiting for the handshake to complete. This can eliminate a full round-trip of latency on reconnection.

Two pieces of state from a prior connection make this possible:

• A session ticket, received via the session.onsessionticket callback, enables TLS session resumption and 0-RTT encryption. Pass it as the sessionOptions.sessionTicket option on a subsequent connection to the same server.
• An address validation token, received via the session.onnewtoken callback, allows the client to skip the server's address validation step (avoiding a Retry round-trip). Pass it as the sessionOptions.token option.

If the server accepts the session ticket, any data sent before the handshake completes is 0-RTT early data. On the server side, stream.early is true for streams carrying early data. The server can reject the 0-RTT attempt (for example, if its configuration has changed since the ticket was issued). When this happens, all streams opened during the 0-RTT phase are destroyed and the client's session.onearlyrejected callback fires. The connection falls back to a normal 1-RTT handshake and the application can reopen streams.

Early data is less secure than data sent after the handshake completes — it can potentially be replayed by an attacker. Applications should treat 0-RTT data with appropriate caution and avoid performing non-idempotent operations during the early data phase.

A typical client session progresses through these stages:

1. Call quic.connect() with a server address and options. This returns a QuicSession.
2. The TLS handshake runs automatically. session.opened resolves when the handshake completes, providing the negotiated ALPN, cipher, and certificate validation results.
3. Open streams, send datagrams, and exchange data.
4. Call session.close() to initiate a graceful shutdown. Existing streams are allowed to finish, then the session is destroyed. The returned promise (also available as session.closed) resolves when teardown is complete.

On the server side, call quic.listen() with a callback. The callback fires for each incoming session after the TLS handshake begins. Incoming streams arrive via the session.onstream callback.

session.destroy() is available for immediate teardown — all open streams are destroyed and the session is closed without waiting for them to finish.

QuicEndpoint and QuicSession support Symbol.asyncDispose, so they can be used with await using for automatic cleanup.

Errors in the quic module are communicated through two complementary mechanisms: the onerror callback and the closed promise.

Both QuicSession and QuicStream expose an optional onerror callback. When a session or stream is destroyed with an error — including errors thrown by other user callbacks — the onerror callback is invoked with the error before the object is torn down. Setting onerror also marks the closed promise as handled, preventing unhandled rejection warnings. If onerror is not set, the error is delivered solely through the rejection of the closed promise.

The QuicError class carries an explicit numeric QUIC error code (error.errorCode) alongside the usual message and code properties. When a QuicError is passed to stream.destroy() or writer.fail(), its errorCode is used in the RESET_STREAM or STOP_SENDING frame sent to the peer. Any other error type falls back to the negotiated protocol's generic internal error code.

When using the Permission Model, the --allow-net flag must be passed to allow QUIC network operations. Without it, calling quic.connect() or quic.listen() will throw an ERR_ACCESS_DENIED error.

$ node --permission --allow-fs-read=* --experimental-quic index.mjs
Error: Access to this API has been restricted. Use --allow-net to manage permissions.
  code: 'ERR_ACCESS_DENIED',
  permission: 'Net',
}

Creating a QuicEndpoint instance without connecting or listening is permitted even without --allow-net, since no network I/O occurs until quic.connect() or quic.listen() is called.

quic.connect(address, options?): Promise

Initiate a new client-side session.

import { connect } from 'node:quic';
import { Buffer } from 'node:buffer';

const enc = new TextEncoder();
const alpn = 'foo';
const client = await connect('123.123.123.123:8888', { alpn });
await client.createUnidirectionalStream({
  body: enc.encode('hello world'),
});

By default, every call to connect(...) will create a new local QuicEndpoint instance bound to a new random local IP port. To specify the exact local address to use, or to multiplex multiple QUIC sessions over a single local port, pass the endpoint option with either a QuicEndpoint or EndpointOptions as the argument.

import { QuicEndpoint, connect } from 'node:quic';

const endpoint = new QuicEndpoint({
  address: '127.0.0.1:1234',
});

const client = await connect('123.123.123.123:8888', { endpoint });

quic.listen(onsession, options?): Promise

Configures the endpoint to listen as a server. When a new session is initiated by a remote peer, the given onsession callback will be invoked with the created session.

import { listen } from 'node:quic';

const endpoint = await listen((session) => {
  // ... handle the session
});

// Closing the endpoint allows any sessions open when close is called
// to complete naturally while preventing new sessions from being
// initiated. Once all existing sessions have finished, the endpoint
// will be destroyed. The call returns a promise that is resolved once
// the endpoint is destroyed.
await endpoint.close();

By default, every call to listen(...) will create a new local QuicEndpoint instance bound to a new random local IP port. To specify the exact local address to use, or to multiplex multiple QUIC sessions over a single local port, pass the endpoint option with either a QuicEndpoint or EndpointOptions as the argument.

At most, any single QuicEndpoint can only be configured to listen as a server once.

An object containing commonly used constants for QUIC configuration.

Congestion control algorithm identifiers, for use with the sessionOptions.cc option:

• quic.constants.cc.RENO — Reno congestion control.
• quic.constants.cc.CUBIC — CUBIC congestion control.
• quic.constants.cc.BBR — BBR congestion control.

The default TLS 1.3 cipher suite list used when sessionOptions.ciphers is not specified.

The default TLS 1.3 key-exchange group list used when sessionOptions.groups is not specified.

A QuicEndpoint encapsulates the local UDP-port binding for QUIC. It can be used as both a client and a server.

new QuicEndpoint(options?): void

The local UDP socket address to which the endpoint is bound, if any.

If the endpoint is not currently bound then the value will be undefined. Read only.

When endpoint.busy is set to true, the endpoint will temporarily reject new sessions from being created. Read/write.

// Mark the endpoint busy. New sessions will be prevented.
endpoint.busy = true;

// Mark the endpoint free. New session will be allowed.
endpoint.busy = false;

The busy property is useful when the endpoint is under heavy load and needs to temporarily reject new sessions while it catches up.

endpoint.close(): Promise

Gracefully close the endpoint. The endpoint will close and destroy itself when all currently open sessions close. Once called, new sessions will be rejected.

Returns a promise that is fulfilled when the endpoint is destroyed.

A promise that is fulfilled when the endpoint is destroyed. This will be the same promise that is returned by the endpoint.close() function. Read only.

True if endpoint.close() has been called and closing the endpoint has not yet completed. Read only.

endpoint.destroy(error?): void

Forcefully closes the endpoint by forcing all open sessions to be immediately closed.

True if endpoint.destroy() has been called. Read only.

True if the endpoint is actively listening for incoming connections. Read only.

The maximum number of concurrent connections allowed per remote IP address. 0 means unlimited (default). Can be set at construction time via the maxConnectionsPerHost option and changed dynamically at any time. The valid range is 0 to 65535.

The maximum total number of concurrent connections across all remote addresses. 0 means unlimited (default). Can be set at construction time via the maxConnectionsTotal option and changed dynamically at any time. The valid range is 0 to 65535.

endpoint.setSNIContexts(entries, options?): void

Replaces or updates the SNI TLS contexts for this endpoint. This allows changing the TLS identity (key/certificate) used for specific host names without restarting the endpoint. Existing sessions are unaffected — only new sessions will use the updated contexts.

endpoint.setSNIContexts({
  'api.example.com': { keys: [newApiKey], certs: [newApiCert] },
});

// Replace the entire SNI map
endpoint.setSNIContexts({
  'api.example.com': { keys: [newApiKey], certs: [newApiCert] },
}, { replace: true });

The statistics collected for an active endpoint. Read only.

endpoint[Symbol.asyncDispose](): void

Calls endpoint.close() and returns a promise that fulfills when the endpoint has closed.

A view of the collected statistics for an endpoint.

A QuicSession represents the local side of a QUIC connection.

session.close(options?): Promise

Initiate a graceful close of the session. Existing streams will be allowed to complete but no new streams will be opened. Once all streams have closed, the session will be destroyed. The returned promise will be fulfilled once the session has been destroyed. If a non-zero code is specified, the promise will reject with an ERR_QUIC_TRANSPORT_ERROR or ERR_QUIC_APPLICATION_ERROR depending on the type.

A promise that is fulfilled once the TLS handshake completes successfully. The resolved value contains information about the established session including the negotiated protocol, cipher suite, certificate validation status, and 0-RTT early data status.

If the handshake fails or the session is destroyed before the handshake completes, the promise will be rejected.

A promise that is fulfilled once the session is destroyed.

True if session.close() has been called and the session has not yet been destroyed. Read only.

session.destroy(error?, options?): void

Immediately destroy the session. All streams will be destroyed and the session will be closed. If error is provided and session.onerror is set, the onerror callback is invoked before destruction. The session.closed promise will reject with the error. If options is provided, the CONNECTION_CLOSE frame sent to the peer will include the specified error code, type, and reason.

True if session.destroy() has been called. Read only.

The endpoint that created this session. Returns null if the session has been destroyed. Read only.

An optional callback invoked when the session is destroyed with an error. This includes errors caused by user callbacks that throw or reject (see Callback error handling). The callback receives a single argument: the error that triggered the destruction. If the onerror callback itself throws or returns a promise that rejects, the error is surfaced as an uncaught exception. Read/write.

Can also be set via the onerror option in quic.connect() or quic.listen().

The callback to invoke when a new stream is initiated by a remote peer. Read/write.

The callback to invoke when a new datagram is received from a remote peer. Read/write.

The callback to invoke when the status of a datagram is updated. Read/write.

The callback to invoke when the server rejects 0-RTT early data. When this fires, all streams that were opened during the 0-RTT phase have been destroyed. The application should re-open streams if needed. Read/write.

This callback only fires on the client side when the server rejects the client's 0-RTT attempt. The connection falls back to 1-RTT and continues normally.

The callback to invoke when the path validation is updated. Read/write.

The callback to invoke when a new session ticket is received. Read/write.

The callback to invoke when a version negotiation is initiated. Read/write.

The callback to invoke when the TLS handshake is completed. Read/write.

The callback to invoke when a NEW_TOKEN token is received from the server. The token can be passed as the token option on a future connection to the same server to skip address validation. Read/write.

The callback to invoke when an ORIGIN frame (RFC 9412) is received from the server, indicating which origins the server is authoritative for. Read/write.

The callback to invoke when the peer sends an HTTP/3 GOAWAY frame, indicating it is initiating a graceful shutdown. The callback receives (lastStreamId) where lastStreamId is a {bigint}:

• When lastStreamId is -1n, the peer sent a shutdown notice (intent to close) without specifying a stream boundary. All existing streams may still be processed.
• When lastStreamId is >= 0n, it is the highest stream ID the peer may have processed. Streams with IDs above this value were NOT processed and can be safely retried on a new connection.

After GOAWAY is received, session.createBidirectionalStream() will throw ERR_INVALID_STATE. Existing streams continue until they complete or the session closes.

This callback is only relevant for HTTP/3 sessions. Read/write.

The callback to invoke when TLS key material is available. Requires sessionOptions.keylog to be true. Each invocation receives a single line of NSS Key Log Format text (including a trailing newline). This is useful for decrypting packet captures with tools like Wireshark. Read/write.

Can also be set via the onkeylog option in quic.connect() or quic.listen().

The callback to invoke when qlog data is available. Requires sessionOptions.qlog to be true. The callback receives a string chunk of JSON-SEQ formatted qlog data and a boolean fin flag. When fin is true, the chunk is the final qlog output for this session and the concatenated chunks form a complete qlog trace. Read/write.

Qlog data arrives during the connection lifecycle. The first chunk contains the qlog header with format metadata. Subsequent chunks contain trace events. The final chunk (with fin set to true) is emitted during session destruction and completes the JSON-SEQ output.

Can also be set via the onqlog option in quic.connect() or quic.listen().

session.createBidirectionalStream(options?): Promise

Open a new bidirectional stream. If the body option is not specified, the outgoing stream will be half-closed. The priority and incremental options are only used when the session supports priority (e.g. HTTP/3). The headers, onheaders, ontrailers, oninfo, and onwanttrailers options are only used when the session supports headers (e.g. HTTP/3).

session.createUnidirectionalStream(options?): Promise

Open a new unidirectional stream. If the body option is not specified, the outgoing stream will be closed. The priority and incremental options are only used when the session supports priority (e.g. HTTP/3).

The local and remote socket addresses associated with the session. Read only.

session.sendDatagram(datagram, encoding?): Promise

Sends an unreliable datagram to the remote peer, returning a promise for the datagram ID.

If datagram is a string, it will be encoded using the specified encoding.

If datagram is an ArrayBufferView, the bytes are copied into an internal buffer; the caller's source buffer is unchanged and may be reused or mutated immediately after the call returns. Callers that want to ensure their source cannot be mutated after the call (for example, when handing the buffer off to another async consumer) can call ArrayBuffer.prototype.transfer() themselves before passing the buffer.

If datagram is a Promise, it will be awaited before sending. If the session closes while awaiting, 0n is returned silently (datagrams are inherently unreliable).

If the datagram payload is zero-length (empty string after encoding, detached buffer, or zero-length view), 0n is returned and no datagram is sent.

For HTTP/3 sessions, the peer must advertise SETTINGS_H3_DATAGRAM=1 (via application: { enableDatagrams: true }) for datagrams to be sent. If the peer's setting is 0, sendDatagram() returns 0n (per RFC 9297 §3, an endpoint MUST NOT send HTTP Datagrams unless the peer indicated support).

Datagrams cannot be fragmented — each must fit within a single QUIC packet. The maximum datagram size is determined by the peer's maxDatagramFrameSize transport parameter (which the peer advertises during the handshake). If the peer sets this to 0, datagrams are not supported and 0n will be returned. If the datagram exceeds the peer's limit, it will be silently dropped and 0n returned. The local maxDatagramFrameSize transport parameter (default: 1200 bytes) controls what this endpoint advertises to the peer as its own maximum.

The local certificate as an object with properties such as subject, issuer, valid_from, valid_to, fingerprint, etc. Returns undefined if the session is destroyed or no certificate is available.

The peer's certificate as an object with properties such as subject, issuer, valid_from, valid_to, fingerprint, etc. Returns undefined if the session is destroyed or the peer did not present a certificate.

The ephemeral key information for the session, with properties such as type, name, and size. Only available on client sessions. Returns undefined for server sessions or if the session is destroyed.

The maximum datagram payload size in bytes that the peer will accept. This is derived from the peer's maxDatagramFrameSize transport parameter minus the DATAGRAM frame overhead (type byte and variable-length integer encoding). Returns 0 if the peer does not support datagrams or if the handshake has not yet completed. Datagrams larger than this value will not be sent.

The maximum number of datagrams that can be queued for sending. Datagrams are queued when sendDatagram() is called and sent opportunistically alongside stream data by the packet serialization loop. When the queue is full, the sessionOptions.datagramDropPolicy determines whether the oldest or newest datagram is dropped. Dropped datagrams are reported as lost via the ondatagramstatus callback.

This property can be changed dynamically to adjust queue capacity based on application activity or memory pressure. The valid range is 0 to 65535.

Return the current statistics for the session. Read only.

session.updateKey(): void

Initiate a key update for the session.

session[Symbol.asyncDispose](): void

Calls session.close() and returns a promise that fulfills when the session has closed.

A QuicError is an Error subclass that carries an explicit numeric QUIC error code. Use it to abort a QUIC stream or session with a specific application-protocol-defined error code rather than letting the implementation pick a generic fallback.

The class is exported from node:quic:

import { QuicError } from 'node:quic';

When a QuicError is supplied to APIs that emit a wire frame (writer.fail(), stream.destroy()), the QUIC stack uses error.errorCode as the wire code for the resulting frame. When any other value is supplied (for example a plain Error), the implementation falls back to the negotiated application protocol's "internal error" code (H3_INTERNAL_ERROR (0x102) for HTTP/3, or the QUIC transport-layer INTERNAL_ERROR (0x1) for raw QUIC).

The Node.js error code (error.code) defaults to 'ERR_QUIC_STREAM_ABORTED'. Callers who need a more specific code string can override it via options.code — the numeric QUIC code is unaffected.

The Node.js error code is fixed at 'ERR_QUIC_STREAM_ABORTED' so that catch blocks can distinguish a QuicError from other Node.js errors without checking the prototype chain. The numeric QUIC code lives on the separate error.errorCode property to avoid colliding with the Node.js convention that error.code is a string.

new QuicError(message, options): void

import { QuicError } from 'node:quic';

const err = new QuicError('rejecting stream', { errorCode: 0x10cn });
console.log(err.code);       // 'ERR_QUIC_STREAM_ABORTED'
console.log(err.errorCode);  // 268n
console.log(err.type);       // 'application'

const custom = new QuicError('custom failure', {
  errorCode: 0x10cn,
  code: 'ERR_MY_QUIC_FAILURE',
});
console.log(custom.code);    // 'ERR_MY_QUIC_FAILURE'

The numeric QUIC error code carried by this error.

Either 'application' or 'transport'. Indicates the namespace of error.errorCode.

A promise that is fulfilled when the stream is fully closed. It resolves when the stream closes cleanly (including idle timeout). It rejects with an ERR_QUIC_APPLICATION_ERROR or ERR_QUIC_TRANSPORT_ERROR when the stream is closed due to a QUIC error (e.g., stream reset by the peer, CONNECTION_CLOSE with a non-zero error code).

stream.destroy(error?, options?): void

Immediately and abruptly destroys the stream. If error is provided and stream.onerror is set, the onerror callback is invoked before destruction. The stream.closed promise rejects with the error.

When the stream is destroyed with an error (or with an explicit options.code), the QUIC stack signals the abort to the peer:

• If the writable side is still open, a RESET_STREAM frame is sent.
• If the readable side is still open (a bidirectional stream, or a remote-initiated unidirectional stream), a STOP_SENDING frame is sent.

Both frames carry the same wire code, resolved with the following precedence:

1. options.code, when explicitly provided.
2. error.errorCode, when error is a QuicError.
3. The negotiated application protocol's "internal error" code (H3_INTERNAL_ERROR (0x102) for HTTP/3, or the QUIC transport-layer INTERNAL_ERROR (0x1) for raw QUIC).

A clean destroy — no error and no options.code — does not emit RESET_STREAM or STOP_SENDING; the stream's existing close machinery handles teardown.

See Aborting a stream for an overview of the available stream-abort APIs.

True if stream.destroy() has been called.

A QuicStream can be aborted in three ways, each producing different wire-frame side effects:

• writer.fail(reason) — Aborts only the writable side. Sends RESET_STREAM to the peer. The readable side is unaffected; any data already buffered for read remains available.
• stream.destroy() with an error argument — Tears the stream down completely. Sends RESET_STREAM on any still-open writable side and STOP_SENDING on any still-open readable side. The wire code is derived from error (see stream.destroy() for the precedence rules).
• stream.destroy() with an explicit options.code — Same as the previous form but with a caller-supplied wire code, which takes precedence over any code carried by error.

When error is a QuicError, its error.errorCode is used as the wire code for both writer.fail() and stream.destroy(). Otherwise the implementation falls back to the negotiated application protocol's "internal error" code (see QuicError).

True if any data on this stream was received as 0-RTT (early data) before the TLS handshake completed. Early data is less secure and could potentially be replayed by an attacker. Applications should treat early data with appropriate caution.

This property is only meaningful on the server side. On the client side, it is always false.

The directionality of the stream, or null if the stream has been destroyed or is still pending. Read only.

The maximum number of bytes that the writer will buffer before writeSync() returns false. When the buffered data exceeds this limit, the caller should wait for drain before writing more.

The value can be changed dynamically at any time. This is particularly useful for streams received via the onstream callback, where the default (65536) may need to be adjusted based on application needs. The valid range is 0 to 4294967295.

The stream ID, or null if the stream has been destroyed or is still pending. Read only.

An optional callback invoked when the stream is destroyed with an error. This includes errors caused by user callbacks that throw or reject (see Callback error handling). The callback receives a single argument: the error that triggered the destruction. If the onerror callback itself throws or returns a promise that rejects, the error is surfaced as an uncaught exception. Read/write.

The callback to invoke when the stream is blocked. Read/write.

The callback to invoke when the peer aborts a direction of the stream by sending a RESET_STREAM frame (the peer abandons their writable side, so no further data will arrive on our readable side) or a STOP_SENDING frame (the peer asks us to stop writing on our writable side).

The callback receives a Node.js error whose errorCode (bigint) property carries the application error code from the wire frame.

The stream is not automatically destroyed when this callback fires — the application chooses how to react. Common patterns are: ignore (and continue using the still-active direction on a bidirectional stream), abort the other direction with writer.fail(), or tear down the whole stream with stream.destroy(). Read/write.

The buffered initial headers received on this stream, or undefined if the application does not support headers or no headers have been received yet. For server-side streams, this contains the request headers (e.g., :method, :path, :scheme). For client-side streams, this contains the response headers (e.g., :status).

Header names are lowercase strings. Multi-value headers are represented as arrays. The object has __proto__: null.

The callback to invoke when initial headers are received on the stream. The callback receives (headers) where headers is an object (same format as stream.headers). For HTTP/3, this delivers request pseudo-headers on the server side and response headers on the client side. Throws ERR_INVALID_STATE if set on a session that does not support headers. Read/write.

The callback to invoke when trailing headers are received from the peer. The callback receives (trailers) where trailers is an object in the same format as stream.headers. Throws ERR_INVALID_STATE if set on a session that does not support headers. Read/write.

The callback to invoke when informational (1xx) headers are received from the server. The callback receives (headers) where headers is an object in the same format as stream.headers. Informational headers are sent before the final response (e.g., 103 Early Hints). Throws ERR_INVALID_STATE if set on a session that does not support headers. Read/write.

The callback to invoke when the application is ready for trailing headers to be sent. This is called synchronously — the user must call stream.sendTrailers() within this callback. Throws ERR_INVALID_STATE if set on a session that does not support headers. Read/write.

Set trailing headers to be sent automatically when the application requests them. This is an alternative to the stream.onwanttrailers callback for cases where the trailers are known before the body completes. Throws ERR_INVALID_STATE if set on a session that does not support headers. Read/write.

stream.sendHeaders(headers, options?): boolean

Sends initial or response headers on the stream. For client-side streams, this sends request headers. For server-side streams, this sends response headers. Throws ERR_INVALID_STATE if the session does not support headers.

stream.sendInformationalHeaders(headers): boolean

Sends informational (1xx) response headers. Server only. Throws ERR_INVALID_STATE if the session does not support headers.

stream.sendTrailers(headers): boolean

Sends trailing headers on the stream. Must be called synchronously during the stream.onwanttrailers callback, or set ahead of time via stream.pendingTrailers. Throws ERR_INVALID_STATE if the session does not support headers.

The current priority of the stream. Returns null if the session does not support priority (e.g. non-HTTP/3) or if the stream has been destroyed. Read only. Use stream.setPriority() to change the priority.

On client-side HTTP/3 sessions, the value reflects what was set via stream.setPriority(). On server-side HTTP/3 sessions, the value reflects the peer's requested priority (e.g., from PRIORITY_UPDATE frames).

stream.setPriority(options?): void

Sets the priority of the stream. Throws ERR_INVALID_STATE if the session does not support priority (e.g. non-HTTP/3). Has no effect if the stream has been destroyed.

stream[Symbol.asyncIterator](): AsyncIterableIterator

The stream implements Symbol.asyncIterator, making it directly usable in for await...of loops. Each iteration yields a batch of Uint8Array chunks.

Only one async iterator can be obtained per stream. A second call throws ERR_INVALID_STATE. Non-readable streams (outbound-only unidirectional or closed) return an immediately-finished iterator.

for await (const chunks of stream) {
  for (const chunk of chunks) {
    // Process each Uint8Array chunk
  }
}

Compatible with stream/iter utilities:

import Stream from 'node:stream/iter';
const body = await Stream.bytes(stream);
const text = await Stream.text(stream);
await Stream.pipeTo(stream, someWriter);

Returns a Writer object for pushing data to the stream incrementally. The Writer implements the stream/iter Writer interface with the try-sync-fallback-to-async pattern.

Only available when no body source was provided at creation time or via stream.setBody(). Non-writable streams return an already-closed Writer. Throws ERR_INVALID_STATE if the outbound is already configured.

The Writer has the following methods:

• writeSync(chunk) — Synchronous write. Returns true if accepted, false if flow-controlled. Data is NOT accepted on false.
• write(chunk[, options]) — Async write with drain wait. options.signal is checked at entry but not observed during the write.
• writevSync(chunks) — Synchronous vectored write. All-or-nothing.
• writev(chunks[, options]) — Async vectored write.
• endSync() — Synchronous close. Returns total bytes or -1.
• end([options]) — Async close.
• fail(reason) — Errors the stream (sends RESET_STREAM to peer). When reason is a QuicError, its error.errorCode is used as the wire code on the resulting RESET_STREAM frame; otherwise the wire code falls back to the negotiated application protocol's "internal error" code (H3_INTERNAL_ERROR (0x102) for HTTP/3, or the QUIC transport-layer INTERNAL_ERROR (0x1) for raw QUIC). See stream.destroy() for a full-stream abort that also resets the readable side via STOP_SENDING.
• desiredSize — Available capacity in bytes, or null if closed/errored.

The bytes from each writeSync() / writevSync() / write() / writev() input chunk are copied into an internal buffer, so the caller's source buffer is unchanged and may be reused or mutated immediately after the call returns. Callers that want to ensure a source buffer cannot be mutated after handing it off can call ArrayBuffer.prototype.transfer() themselves before passing the buffer.

stream.setBody(body): void

Sets the outbound body source for the stream. Can only be called once. Mutually exclusive with stream.writer.

The following body source types are supported:

• null — The writable side is closed immediately (FIN sent with no data).
• string — UTF-8 encoded and sent as a single chunk.
• ArrayBuffer, SharedArrayBuffer, ArrayBufferView — Sent as a single chunk. The bytes are copied into an internal buffer, so the caller's source buffer is unchanged and may be reused or mutated immediately after the call returns. Callers wanting to ensure their source cannot be mutated after handing it off can call ArrayBuffer.prototype.transfer() themselves before passing the buffer.
• Blob — Sent from the Blob's underlying data queue.
• <FileHandle> — The file contents are read asynchronously via an fd-backed data source. The FileHandle must be opened for reading (e.g. via fs.promises.open(path, 'r')). Once passed as a body, the FileHandle is locked and cannot be used as a body for another stream. The FileHandle is automatically closed when the stream finishes.
• AsyncIterable, Iterable — Each yielded chunk (string or Uint8Array) is written incrementally in streaming mode.
• Promise — Awaited; the resolved value is used as the body (subject to the same type rules).

Throws ERR_INVALID_STATE if the outbound is already configured or if the writer has been accessed.

The session that created this stream, or null if the stream has been destroyed. Read only.

The current statistics for the stream. Read only.

The endpoint configuration options passed when constructing a new QuicEndpoint instance.

If not specified the endpoint will bind to IPv4 localhost on a random port.

The endpoint maintains an internal cache of validated socket addresses as a performance optimization. This option sets the maximum number of addresses that are cached. This is an advanced option that users typically won't have need to specify.

When true, the endpoint will not send stateless reset packets in response to packets from unknown connections. Stateless resets allow a peer to detect that a connection has been lost even when the server has no state for it. Disabling them may be useful in testing or when stateless resets are handled at a different layer.

The number of seconds an endpoint will remain alive after all sessions have closed and it is no longer listening. A value of 0 (default) means the endpoint is only destroyed when explicitly closed via endpoint.close() or endpoint.destroy(). A positive value starts an idle timer when the endpoint becomes idle; if no new sessions are created before the timer fires, the endpoint is automatically destroyed. This is useful for connection pooling where endpoints should linger briefly for reuse by future connect() calls.

When true, indicates that the endpoint should bind only to IPv6 addresses.

Specifies the maximum number of concurrent sessions allowed per remote IP address (ignoring port). When the limit is reached, new connections from the same IP are refused with CONNECTION_REFUSED. A value of 0 disables the limit. The maximum value is 65535.

This limit can also be changed dynamically after construction via endpoint.maxConnectionsPerHost.

Specifies the maximum total number of concurrent sessions across all remote addresses. When the limit is reached, new connections are refused with CONNECTION_REFUSED. A value of 0 disables the limit. The maximum value is 65535.

This limit can also be changed dynamically after construction via endpoint.maxConnectionsTotal.

Specifies the maximum number of QUIC retry attempts allowed per remote peer address.

Specifies the maximum number of stateless resets that are allowed per remote peer address.

Specifies the length of time a QUIC retry token is considered valid.

Specifies the 16-byte secret used to generate QUIC retry tokens.

Specifies the length of time a QUIC token is considered valid.

Specifies the 16-byte secret used to generate QUIC tokens.

When true, requires that the endpoint validate peer addresses using retry packets while establishing a new connection.

The ALPN (Application-Layer Protocol Negotiation) identifier(s).

For client sessions, this is a single string specifying the protocol the client wants to use (e.g. 'h3').

For server sessions, this is an array of protocol names in preference order that the server supports (e.g. ['h3', 'h3-29']). During the TLS handshake, the server selects the first protocol from its list that the client also supports.

The negotiated ALPN determines which Application implementation is used for the session. 'h3' and 'h3-*' variants select the HTTP/3 application; all other values select the default application.

Default: 'h3'

HTTP/3 application-specific options. These only apply when the negotiated ALPN selects the HTTP/3 application ('h3').

const { listen } = await import('node:quic');

await listen((session) => { /* ... */ }, {
  application: {
    maxHeaderPairs: 64,
    qpackMaxDTableCapacity: 8192,
    enableDatagrams: true,
  },
  // ... other session options
});

The CA certificates to use for client sessions. For server sessions, CA certificates are specified per-identity in the sessionOptions.sni map.

Specifies the congestion control algorithm that will be used. Must be set to one of either 'reno', 'cubic', or 'bbr'.

This is an advanced option that users typically won't have need to specify.

The TLS certificates to use for client sessions. For server sessions, certificates are specified per-identity in the sessionOptions.sni map.

The list of supported TLS 1.3 cipher algorithms.

The CRL to use for client sessions. For server sessions, CRLs are specified per-identity in the sessionOptions.sni map.

When true, enables TLS 0-RTT early data for this session. Early data allows the client to send application data before the TLS handshake completes, reducing latency on reconnection when a valid session ticket is available. Set to false to disable early data support.

The list of supported TLS 1.3 cipher groups.

When true, enables TLS key logging for the session. Key material is delivered to the session.onkeylog callback in NSS Key Log Format. Each callback invocation receives a single line of key material. The output can be used with tools such as Wireshark to decrypt captured QUIC traffic.

The TLS crypto keys to use for client sessions. For server sessions, keys are specified per-identity in the sessionOptions.sni map.

Specifies the maximum UDP packet payload size.

Specifies the maximum stream flow-control window size.

Specifies the maximum session flow-control window size.

The minimum QUIC version number to allow. This is an advanced option that users typically won't have need to specify.

When the remote peer advertises a preferred address, this option specifies whether to use it or ignore it.

When true, enables qlog diagnostic output for the session. Qlog data is delivered to the session.onqlog callback as chunks of JSON-SEQ formatted text. The output can be analyzed with qlog visualization tools such as qvis.

Controls which datagram to drop when the pending datagram queue (sized by session.maxPendingDatagrams) is full. Must be one of 'drop-oldest' (discard the oldest queued datagram to make room) or 'drop-newest' (reject the incoming datagram). Dropped datagrams are reported as lost via the ondatagramstatus callback.

This option is immutable after session creation.

The maximum number of SendPendingData cycles a datagram can survive without being sent before it is abandoned. When a datagram cannot be sent due to congestion control or packet size constraints, it remains in the queue and the attempt counter increments. Once the limit is reached, the datagram is dropped and reported as 'abandoned' via the ondatagramstatus callback. Valid range: 1 to 255.

A multiplier applied to the Probe Timeout (PTO) to compute the draining period duration after receiving a CONNECTION_CLOSE frame from the peer. RFC 9000 Section 10.2 requires the draining period to persist for at least three times the current PTO. The valid range is 3 to 255. Values below 3 are clamped to 3.

Specifies the maximum number of milliseconds a TLS handshake is permitted to take to complete before timing out.

Specifies the keep-alive timeout in milliseconds. When set to a non-zero value, PING frames will be sent automatically to keep the connection alive before the idle timeout fires. The value should be less than the effective idle timeout (maxIdleTimeout transport parameter) to be useful.

The peer server name to target (SNI). Defaults to 'localhost'.

An object mapping host names to TLS identity options for Server Name Indication (SNI) support. This is required for server sessions and must contain at least one entry. The special key '*' specifies the optional default/fallback identity used when no other host name matches. If no wildcard entry is provided, connections with unrecognized server names will be rejected with a TLS unrecognized_name alert. Each entry may contain:

const endpoint = await listen(callback, {
  sni: {
    '*': { keys: [defaultKey], certs: [defaultCert] },
    'api.example.com': { keys: [apiKey], certs: [apiCert], port: 8443 },
    'www.example.com': { keys: [wwwKey], certs: [wwwCert], ca: [customCA] },
    'internal.example.com': { keys: [intKey], certs: [intCert], authoritative: false },
  },
});

Shared TLS options (such as ciphers, groups, keylog, and verifyClient) are specified at the top level of the session options and apply to all identities. Each SNI entry overrides only the per-identity certificate fields.

The SNI map can be replaced at runtime using endpoint.setSNIContexts(), which atomically swaps the map for new sessions while existing sessions continue to use their original identity.

True to enable TLS tracing output.

An opaque address validation token previously received from the server via the session.onnewtoken callback. Providing a valid token on reconnection allows the client to skip the server's address validation, reducing handshake latency.

The QUIC transport parameters to use for the session.

Specifies the maximum number of unacknowledged packets a session should allow.

If true, the peer certificate is verified against the list of supplied CAs. An error is emitted if verification fails; the error can be inspected via the validationErrorReason and validationErrorCode fields in the handshake callback. If false, peer certificate verification errors are ignored.

When true (the default), connect() will attempt to reuse an existing endpoint rather than creating a new one for each session. This provides connection pooling behavior — multiple sessions can share a single UDP socket. The reuse logic will not return an endpoint that is listening on the same address as the connect target (to prevent CID routing conflicts).

Set to false to force creation of a new endpoint for the session. This is useful when endpoint isolation is required (e.g., testing stateless reset behavior where source port identity matters).

True to require verification of TLS client certificate.

True to require private key verification for client sessions. For server sessions, this option is specified per-identity in the sessionOptions.sni map.

The QUIC version number to use. This is an advanced option that users typically won't have need to specify.