Slow require times (between 2 to 3 seconds) of logging-bunyan when used with Firebase functions

Environment details

• OS: Firebase functions
• Node.js version: 6.11.5
• npm version: yarn 1.9.4
• @google-cloud/logging-bunyan version: 0.8.2

Steps to reproduce

I am seeing particularly long cold start times consistently with Firebase functions and so did some basic instrumentation about which requires were the most time consuming. I’m seeing the time needed to require logging-bunyan to be consistently between 2 and 3 seconds when deployed to Firebase, but when running in local emulator to be around 200ms.

I’ve also seen some intermittent errors causing the function to crash due to Error: 14 UNAVAILABLE: TCP Read failed https://github.com/GoogleCloudPlatform/google-cloud-node/issues/2438#issuecomment-412200264

Not sure if they’re related but I’m wondering if there’s some underlying connectivity issue causing these long require times on firebase.

[ofrobots]

I did a require-time profile of this library (v0.9.4) using require-so-slow. It seems that the library takes 344ms to require. 340ms out of that spend requiring @google-cloud/logging. Profile can be viewed here.

I am going to move this issue over to the logging library.

[JustinBeckwith]

Good news! We’ve cut this down ever further. We’re now down to about 250ms of time to load. As we move towards a native http2 implementation in grpc in the next few months, it will get even better:

One thing you want to be really careful about - how do you instantiate your logging objects? Where you instantiate the client (inside of the function or outside) makes a VERY bug difference on how long it will take to first load.

Given the fact that we’re down pretty low on the require time, with a plan to improve more, I’m going to close this out for now. If folks are still seeing something you want us to look into - please let me know!