Cloud Run compatibility

[grayside]

I've built a simple Hello World express service for Cloud Run, and can sometimes see data show up in the Trace console. However, most of the time it's not getting through.

This could easily be because CPU is withdrawn from the Cloud Run service as soon as a response is sent and there is no mechanism for cleanup operations when a container instance is halted.

Here's the most recent code I've tried:

// Initialize tracing agent.
const tracer = require('@google-cloud/trace-agent').start({
  samplingRate: 0,
  bufferSize: 1,
  flushDelaySeconds: 1,
});

// Bootstrap express app.
const sleep = require('util').promisify(setTimeout);
const express = require('express');
const app = express();

app.get('/', async (req, res) => {
  const span = tracer.createChildSpan({name: 'greeting'});
  const greeting = 'Howdy Friends';
  sleep(2000)
  span.endSpan();

  // Existing endSpan() call in express wrapper happens after response sent.
  // Adding this here to try forcing the data to publish.
  const root = tracer.getCurrentRootSpan();
  root.endSpan();

  await sleep(process.env.SLEEP_TIME || 0)
  res.status(200).send(greeting);
});

This code works fine from localhost but has only managed to submit telemetry in 2-3 requests from Cloud Run across many curl attempts in the last hour.

My expectation is that a guarantee of sending telemetry from Cloud Run will require something like the following:

const tracer = require('@google-cloud/trace-agent').start();

app.get('/', (req, res) => {
  const json = { fancy: 'value' };
  await tracer.FlushAndPublishNow();
  res.send(200).send(json);
});

[kjin]

I ran your code exactly as written (adding only a statement to listen to a port) and could not seem to reproduce this -- I see 98/100 requests being traced (the remaining 2 might be due to your explanation above. This is after essentially following the Cloud Run tutorial (https://cloud.google.com/run/docs/quickstarts/build-and-deploy) and replacing the tutorial code with yours. Is your setup different?

For the record, I think you also need to await the sleep(2000) statement but that should be inconsequential here.

[grayside]

Alright, tried this again, and tonight it's working much more consistently. The question then: was the 90 minute period in which I tested this previously a fluke, or does this race condition vary between minor and severe impact?

I am essentially following the quickstart.

[kjin]

I'm not very familiar with the internal workings of Cloud Run, so unfortunately I can't answer that question.

Please re-open this bug if you observe traces not getting published again. Thanks!