Skip to content

Error Handling

Contents | Previous: Connection Lifecycle | Next: Browser

View in another language — C++ · C#/.NET · Java · Kotlin · Node/TypeScript

After reading this chapter

You can read the error code of a failed call and judge what that code means for the connection, so you can choose between retrying, recovering, and stopping.

The connector's error codes form a closed set. An implementation neither adds nor removes a value, so once a handling is written per code, no new code appears to leave a branch missing. This chapter covers how those codes reach the caller and what each one means.

1. Where Errors Arrive

The delivery differs by surface while the meaning stays the same. A surface that awaits completion reports the failure there, a surface that takes a callback reports it through a result object, and an error that belongs to no request is delivered as an error event. In every case the receiver can read the code.

try {
  const reply = await connector
    .request(new LoginRequest('player-1', 'tok-abc123'))
    .submit<LoginReply>();
} catch (failure) {
  if (failure instanceof ZlinkStreamException
    && failure.error.code === ZlinkStreamErrorCode.RequestTimeout) {
    retryLogin();
  }
}

A standard exception of the language is never thrown as it is. The standard types for a bad argument or a bad state have no place to carry a code, so the caller cannot tell a configuration error from a validation failure. A language that delivers errors as exceptions uses a dedicated exception type that carries the code, and a build with exceptions off delivers the same code as a result value.

2. Errors That Belong to No Request

A frame that could not be decoded, or a server error unrelated to any request, has no call waiting for it, so it is delivered as an error event. This handler also returns a value that can be released.

const errors = connector.onErrorReceived(error => {
  logError(error.code, error.message);
});

3. Error Codes

Code Meaning
disconnected There is no connection, or it dropped
ConfigurationError The configuration is wrong — endpoint scheme against transport, a transport the runtime does not support
ValidationFailed A pre-send check, an option range check, or an observation condition of a wait surface did not hold
RequestTimeout The wait for an answer ran out of time
ConnectTimeout The wait for a connection ran out of time
FrameDecodeFailed A frame or header could not be decoded
FrameTooLarge A received payload exceeded the receive limit
SendFailed The transmission failed
CompressionFailed Compression failed
DecompressionFailed Decompression failed
TlsValidationFailed TLS validation failed
UserCallbackFailed A user callback failed
RemoteError The server answered with an error

The two codes for running out of time mean different things. A request that ran out of time did not receive its answer; a wait surface that ran out of time observed something other than what was expected. That is why the second is reported as a validation failure — the caller has to handle the two cases differently.

To return a domain error as a normal answer, the server uses a successful answer with its own payload rather than an error answer. The payload of an error answer is always JSON carrying a code and a message, whatever the codec setting is.

4. What an Error Means for the Connection

A failure that ends only the current call and a failure that ends the connection call for different handling.

Code Current call Connection Automatic reconnect
ConfigurationError · ValidationFailed fails kept no
RequestTimeout only that request fails kept no
ConnectTimeout · TlsValidationFailed connect fails disconnected applies the attempt policy
disconnected — transport dropped the call in progress fails disconnected; the close reason is a transport error applies it when enabled
disconnected caused by close the call in progress fails closed; the close reason is client close no
SendFailed — request sequence exhausted only that call fails kept no
SendFailed — transport write failed only that write's call fails with SendFailed; other calls in progress fail with disconnected ends; close reason is TransportError applies it when enabled
FrameDecodeFailed (frame or header) · FrameTooLarge the frame is not delivered and pending requests fail ends; the close reason is a protocol error applies it when enabled
CompressionFailed only that send fails kept no
DecompressionFailed only that received packet or pending request fails kept no
UserCallbackFailed · RemoteError delivered as an error event or to the related call kept no

A call in progress that fails because the connection ended fails with disconnected whatever ended it, and the cause remains in the close reason. When a transport write failure ends the connection, only the call of that write fails with SendFailed. Reading the close reason is covered by Connection Lifecycle.

5. Common Handling

No connection. A send that fails as no connection means reconnection is in progress or has already been given up. Register a connection-state handler and send again once the connection is back. Packets whose value expires with time are not sent again.

Answer timed out. Only that request failed and the connection stayed, so the same request can be sent again. The server may have processed it already and only the answer was late, so a request that must not be processed twice is made recognizable on the server side.

Receive limit exceeded. A received payload over the receive limit is not delivered and the connection ends with the close reason protocol error. If the server is configured to send larger packets, raise the receive limit in Connector Options.

Server error answer. An error answer from the server fails that request, and one that matches no request is delivered as an error event. The connection is kept, so other packets are unaffected.

6. A Code That Does Not Occur on Every Runtime

TLS validation failure does not occur on a browser runtime, because the browser's WebSocket API does not distinguish a TLS failure from an ordinary connection failure. The code stays in the set and is simply unused there, so a branch written per code does not change from runtime to runtime.