Shaving 2 Minutes Off Our Graphcool Deployment Times
Photo by Erik Eastman on Unsplash
Graphcool is an awesome GraphQL-powered backend we use at Orchard. One of the features they offer is custom functions (resolvers) a la AWS Lambda.
The thing is, deploying to any environment includes both prod AND dev dependencies inside your node_modules folder. That means eslint and anything else you have in that directory.
I’m sure this will change soon enough, but for now there’s a few simple steps we can follow to minimize our bundle size & time spent deploying:
- Remove any files & folders that aren’t related to Graphcool. That goes for any scripts, utilities, functions, etc that don’t need to live here.
/src only includes the functions we require. Nothing else.
├── README.md
├── graphcool.yml
├── node_modules
├── package.json
├── src
├── types.graphql
└── yarn.lock
- Create npm commands in your package.json that will delete node_modules and only install packages required for production. Now you’ll know for sure that none of your eslint stuff will be bundled and deployed.
"deploy:dev": "rm -rf node_modules && yarn --prod && graphcool deploy -t dev && yarn"
- Lastly, YMMV but yarn includes a
—-flat
flag you can use to only install one version of a package across ALL your dependencies. We didn’t run into any issues using this, but you might so give it a go on dev first:
"deploy:dev": "rm -rf node_modules && yarn --prod --flat && graphcool deploy -t dev && yarn"
We were going to go as far as removing lodash and installing only the modules we use (lodash.chunk, lodash.get) but after running — flat, we discovered some of our node modules already require lodash and saved ourselves the hassle by just forcing that version. Our yarn.lock file is like 100x smaller!
We went from 120s deployments down to 19.1s!
So it was worth writing about :)