PDE-2650 perf(core): improve logging performance

[eliangcs]

Improves z.console.log and HTTP logging performance. In this PR, we stream many logs in a single request to the /input endpoint on the log server. This reduces the overhead where the original /input endpoint only took one log in a request.

I used the following trigger code for benchmark:

const perform = async (z, bundle) => {
  const response = await z.request('https://httpbin.zapier-tooling.com/get');
  const requestId = response.data.headers['X-Request-Id'][0];

  for (let i = 0; i < 200; i++) {
    z.console.log(`hello ${i}`);
  }

  return [{ id: requestId }];
};

The above example code produces one HTTP log and 200 console logs. The old way (one log, one request) takes ~6.6 seconds to invoke the trigger. The new way (stream many logs in a request) takes ~3.5 seconds. I was expecting a bigger improvement but I think this is as good as it gets.

[eliangcs]

I was running zapier push and found the CLI tool sent HTTP logs to the log server. So I added !event.calledFromCli to make sure we don't patch http when it was from CLI.

[eliangcs]

This ensures HTTP logs and console logs share the same logger instance created by createLogger(). In the old implementation, there's no difference because every log is a separate request. But in the new implementation, every time you call createLogger(), you create a new request, ready to stream logs. Here we want to reuse the same request as much as possible.

[eliangcs]

@xavdid I found there's an issue with lost logs, which I'm still figuring out. So you don't have to review this PR just yet!

[eliangcs]

I moved applyEnvironment up because I found the logging endpoint could be different on the first invocation than it was on the subsequent invocations.

Here you can see the logging endpoint is either derived from the environment variable LOGGING_ENDPOINT or the constant DEFAULT_LOGGING_HTTP_ENDPOINT. But in the original (bad) code, we applyEnvironment AFTER createLogger. So on the first invocation, it would get the logging endpoint from the constant. But on the second or subsequent calls, since the environment variables have been loaded, it gets the logging endpoint from the environment variable.

We don't see any issues on production because the constant and the environment variables are the same (https://httplogger.zapier.com/input) on production.

To fix, we need to make sure we applyEnvironment BEFORE createLogger.

[eliangcs]

@xavdid this is ready for review!

[eliangcs]

Hopefully this fixes the test timeout issue.

[xavdid]

Very awesome! Looks great here.