mirror of
https://github.com/Kong/insomnia
synced 2024-11-08 23:00:30 +00:00
262 lines
6.9 KiB
JavaScript
262 lines
6.9 KiB
JavaScript
// @flow
|
|
|
|
import * as grpc from '@grpc/grpc-js';
|
|
|
|
import * as models from '../../models';
|
|
import * as protoLoader from './proto-loader';
|
|
import callCache from './call-cache';
|
|
import type { ServiceError } from './service-error';
|
|
import { GrpcStatusEnum } from './service-error';
|
|
import type { Call } from './call-cache';
|
|
import parseGrpcUrl from './parse-grpc-url';
|
|
|
|
const _createClient = (req: GrpcRequest, respond: ResponseCallbacks): Object | undefined => {
|
|
const { url, enableTls } = parseGrpcUrl(req.url);
|
|
|
|
if (!url) {
|
|
respond.sendError(req._id, new Error('URL not specified'));
|
|
return undefined;
|
|
}
|
|
|
|
const credentials = enableTls ? grpc.credentials.createSsl() : grpc.credentials.createInsecure();
|
|
|
|
console.log(`[gRPC] connecting to url=${url} ${enableTls ? 'with' : 'without'} TLS`);
|
|
const Client = grpc.makeGenericClientConstructor({});
|
|
return new Client(url, credentials);
|
|
};
|
|
|
|
export const sendUnary = async (requestId: string, respond: ResponseCallbacks): Promise<void> => {
|
|
// Load method
|
|
const req = await models.grpcRequest.getById(requestId);
|
|
const selectedMethod = await protoLoader.getSelectedMethod(req);
|
|
|
|
if (!selectedMethod) {
|
|
respond.sendError(requestId, new Error(`Method definition could not be found`));
|
|
return;
|
|
}
|
|
|
|
// Load initial message
|
|
const messageBody = _parseMessage(req, respond);
|
|
if (!messageBody) {
|
|
return;
|
|
}
|
|
|
|
// Create client
|
|
const client = _createClient(req, respond);
|
|
if (!client) {
|
|
return;
|
|
}
|
|
|
|
// Create callback
|
|
const callback = _createUnaryCallback(requestId, respond);
|
|
|
|
// Make call
|
|
const call = client.makeUnaryRequest(
|
|
selectedMethod.path,
|
|
selectedMethod.requestSerialize,
|
|
selectedMethod.responseDeserialize,
|
|
messageBody,
|
|
callback,
|
|
);
|
|
|
|
_setupStatusListener(call, requestId, respond);
|
|
respond.sendStart(requestId);
|
|
|
|
// Save call
|
|
callCache.set(requestId, call);
|
|
};
|
|
|
|
export const startClientStreaming = async (
|
|
requestId: string,
|
|
respond: ResponseCallbacks,
|
|
): Promise<void> => {
|
|
// Load method
|
|
const req = await models.grpcRequest.getById(requestId);
|
|
const selectedMethod = await protoLoader.getSelectedMethod(req);
|
|
|
|
if (!selectedMethod) {
|
|
respond.sendError(requestId, new Error(`Method definition could not be found`));
|
|
return;
|
|
}
|
|
|
|
// Create client
|
|
const client = _createClient(req, respond);
|
|
|
|
if (!client) {
|
|
return;
|
|
}
|
|
|
|
// Create callback
|
|
const callback = _createUnaryCallback(requestId, respond);
|
|
|
|
// Make call
|
|
const call = client.makeClientStreamRequest(
|
|
selectedMethod.path,
|
|
selectedMethod.requestSerialize,
|
|
selectedMethod.responseDeserialize,
|
|
callback,
|
|
);
|
|
|
|
_setupStatusListener(call, requestId, respond);
|
|
respond.sendStart(requestId);
|
|
|
|
// Save call
|
|
callCache.set(requestId, call);
|
|
};
|
|
|
|
export const startServerStreaming = async (
|
|
requestId: string,
|
|
respond: ResponseCallbacks,
|
|
): Promise<void> => {
|
|
const req = await models.grpcRequest.getById(requestId);
|
|
const selectedMethod = await protoLoader.getSelectedMethod(req);
|
|
|
|
if (!selectedMethod) {
|
|
respond.sendError(
|
|
requestId,
|
|
new Error(`The gRPC method ${req.protoMethodName} could not be found`),
|
|
);
|
|
return;
|
|
}
|
|
|
|
// Load initial message
|
|
const messageBody = _parseMessage(req, respond);
|
|
if (!messageBody) {
|
|
return;
|
|
}
|
|
|
|
// Create client
|
|
const client = _createClient(req, respond);
|
|
if (!client) {
|
|
return;
|
|
}
|
|
|
|
// Make call
|
|
const call = client.makeServerStreamRequest(
|
|
selectedMethod.path,
|
|
selectedMethod.requestSerialize,
|
|
selectedMethod.responseDeserialize,
|
|
messageBody,
|
|
);
|
|
|
|
_setupStatusListener(call, requestId, respond);
|
|
_setupServerStreamListeners(call, requestId, respond);
|
|
|
|
respond.sendStart(requestId);
|
|
|
|
// Save call
|
|
callCache.set(requestId, call);
|
|
};
|
|
|
|
export const startBidiStreaming = async (
|
|
requestId: string,
|
|
respond: ResponseCallbacks,
|
|
): Promise<void> => {
|
|
const req = await models.grpcRequest.getById(requestId);
|
|
const selectedMethod = await protoLoader.getSelectedMethod(req);
|
|
|
|
if (!selectedMethod) {
|
|
respond.sendError(
|
|
requestId,
|
|
new Error(`The gRPC method ${req.protoMethodName} could not be found`),
|
|
);
|
|
return;
|
|
}
|
|
|
|
// Create client
|
|
const client = _createClient(req, respond);
|
|
|
|
if (!client) {
|
|
return;
|
|
}
|
|
|
|
// Make call
|
|
const call = client.makeBidiStreamRequest(
|
|
selectedMethod.path,
|
|
selectedMethod.requestSerialize,
|
|
selectedMethod.responseDeserialize,
|
|
);
|
|
|
|
_setupStatusListener(call, requestId, respond);
|
|
_setupServerStreamListeners(call, requestId, respond);
|
|
|
|
respond.sendStart(requestId);
|
|
|
|
// Save call
|
|
callCache.set(requestId, call);
|
|
};
|
|
|
|
export const sendMessage = async (requestId: string, respond: ResponseCallbacks) => {
|
|
const req = await models.grpcRequest.getById(requestId);
|
|
const messageBody = _parseMessage(req, respond);
|
|
|
|
if (!messageBody) {
|
|
return;
|
|
}
|
|
|
|
callCache.get(requestId)?.write(messageBody, _streamWriteCallback);
|
|
};
|
|
|
|
export const commit = (requestId: string) => callCache.get(requestId)?.end();
|
|
|
|
export const cancel = (requestId: string) => callCache.get(requestId)?.cancel();
|
|
|
|
export const cancelMultiple = (requestIds: Array<string>) => requestIds.forEach(cancel);
|
|
|
|
const _setupStatusListener = (call: Call, requestId: string, respond: ResponseCallbacks) => {
|
|
call.on('status', s => respond.sendStatus(requestId, s));
|
|
};
|
|
|
|
const _setupServerStreamListeners = (call: Call, requestId: string, respond: ResponseCallbacks) => {
|
|
call.on('data', data => respond.sendData(requestId, data));
|
|
|
|
call.on('error', (error: ServiceError) => {
|
|
if (error?.code !== GrpcStatusEnum.CANCELLED) {
|
|
respond.sendError(requestId, error);
|
|
|
|
// Taken through inspiration from other implementation, needs validation
|
|
if (error.code === GrpcStatusEnum.UNKNOWN || error.code === GrpcStatusEnum.UNAVAILABLE) {
|
|
respond.sendEnd(requestId);
|
|
}
|
|
}
|
|
});
|
|
call.on('end', () => respond.sendEnd(requestId));
|
|
};
|
|
|
|
// This function returns a function
|
|
const _createUnaryCallback = (requestId: string, respond: ResponseCallbacks) => (
|
|
err: ServiceError,
|
|
value: Object,
|
|
) => {
|
|
if (err) {
|
|
// Don't do anything if cancelled
|
|
// TODO: test with other errors
|
|
if (err.code !== GrpcStatusEnum.CANCELLED) {
|
|
respond.sendError(requestId, err);
|
|
}
|
|
} else {
|
|
respond.sendData(requestId, value);
|
|
}
|
|
respond.sendEnd(requestId);
|
|
callCache.clear(requestId);
|
|
};
|
|
|
|
type WriteCallback = (error: Error | null | undefined) => void;
|
|
|
|
const _streamWriteCallback: WriteCallback = err => {
|
|
if (err) {
|
|
console.error('[gRPC] Error when writing to stream', err);
|
|
}
|
|
};
|
|
|
|
const _parseMessage = (request: Request, respond: ResponseCallbacks): Object | undefined => {
|
|
try {
|
|
return JSON.parse(request.body.text || '');
|
|
} catch (e) {
|
|
// TODO: How do we want to handle this case, where the message cannot be parsed?
|
|
// Currently an error will be shown, but the stream will not be cancelled.
|
|
respond.sendError(request._id, e);
|
|
return undefined;
|
|
}
|
|
};
|