Vertex AI: Getting a GRPC Exception when sending a prediction request in Java

[Roborg]

Hi,

I have deployed a custom model (from a Docker image) in a Vertex AI endpoint.

When I try to get a prediction in Java with the following code:

private PredictResponse predict(String endpointId, String query, String project, String location) throws IOException {
    try (PredictionServiceClient serviceClient = getPredictionServiceClient()) {
        EndpointName endpointName = EndpointName.of(project, location, endpointId);
        ListValue.Builder listValue = ListValue.newBuilder();
        JsonFormat.parser().merge(query, listValue);
        List<Value> instanceList = listValue.getValuesList();
        PredictRequest request = PredictRequest.newBuilder()
                .setEndpoint(endpointName.toString())
                .addAllInstances(instanceList)
                .build();
        return serviceClient.predict(request);
    }
}

I got a io.grpc.StatusRuntimeException: INTERNAL: RST_STREAM closed stream. HTTP/2 error code: INTERNAL_ERROR.

I see no error when I check the logs in the Google Cloud Console. It looks like the request has been processed without error.

The strange part is that everything is working if:

• I call another Vertex model endpoint (with a simpler model inside a Docker image)
• I use a HTTP REST request (with curl) to get the prediction from the initial model (instead of Java code)

The request in CLI works so it seems that the model+endpoint is working.

The request in Java to another endpoint is working, so the code seems correct.

Any idea how to debug/fix this?

[Roborg]

Update: if I use the CLI to send the request, there is also an error.

gcloud ai endpoints predict $ENDPOINT_ID --json-request=input.json --region=europe-west4

Returns

Using endpoint [https://europe-west4-prediction-aiplatform.googleapis.com/]
ERROR: gcloud crashed (ContentDecodingError): ('Received response with content-encoding: gzip, gzip, but failed to decode it.', error('Error -3 while decompressing data: incorrect header check'))

[Roborg · Accepted solution]

I have found an open issue on this topic: https://issuetracker.google.com/issues/234474507

I have posted a workaround that also solves the bug in the Java SDK.