How to change Client GRPC deadline?

[bloodcarter]

I randomly receive this error while streaming data:
Client GRPC deadline too short. Should be at least: 3 * audio-duration + 5 seconds. Current deadline is: 59.699356980447192 second(s). Required at least: 60 second(s).

What should I do with that?

[jmdobry]

@jmuk How would one increase the deadline?

[jmuk]

The autogen layer can customize the deadline through the CallOption parameter, however @google-cloud/speech hand-written layer doesn't allow passing timeout by themselves.

Which method are you using? I'm wild-guessing that asyncRecognize does not have such restriction.

[jmdobry]

Something like this would work:

const v1beta1 = require('@google-cloud/speech').v1beta1();
const speech = v1beta1.speechClient();

const request = {
  audio: {
    uri: 'gs://my-bucket/filename.raw'
  },
  config: {
    encoding: 'LINEAR16',
    sampleRate: 16000
  }
};

const options = {
  longrunning: {
    initialRetryDelayMillis: 100, , // Default
    retryDelayMultiplier: 1.3, // Default
    maxRetryDelayMillis: 60000, // Default
    initialRpcTimeoutMillis: null, // Default
    rpcTimeoutMultiplier: null, // Default
    maxRpcTimeoutMillis: null, // Default
    totalTimeoutMillis: 600000 // Default
  }
};

speech.syncRecognize(request, options)
  .then((results) => {
    const transcription = results[0];

    console.log(transcription);
  });

[bloodcarter]

@jmdobry Would this work with recognizeStream?

[jmdobry]

I'm not sure the bi-directional streaming uses exponential backoff. @jmuk?

[jmuk]

@bloodcarter: createRecognizeStream can customize the timeout as https://github.com/GoogleCloudPlatform/google-cloud-node/blob/master/packages/speech/src/index.js#L378

const speech = require('@google-cloud')();
const stream = speech.createRecognizeStream(..., {timeout: 300});
...

@jmdobry: bidi streaming does not use exponential backoff, but uses timeout field in the options.

const v1beta1 = require('@google-cloud/speech').v1beta1();
const speech = v1beta1.speechClient();

// set the timeout to 300 seconds.
const stream = speech.streamingRecognize({timeout: 300});
stream.write(...);
...

This should customize the timeout.