Into the Wild with Node.js and Web Apps

Node.js Talk at Klout, 2012
Chris Bumgardner
At "Beyond the Closed Beta, Node.js from code to production (at scale)"
Node.js ++ (Serverside Javascripters) Club SF

Transcript

1. INTO THE WILD WITH NODE.JS AND WEB APPS Node.js Talk at Klout, 2012 Chris Bumgardner Friday, April 20, 12
2. ME • Co-founder and Chief-Programmer of AskYourUsers.com • @chrisbumgardner and cbumgard@gmail.com • github.com/cbumgard • My first ‘real’ production web app in Node.js • Invite only beta, going live this week. Friday, April 20, 12
3. THE STACK • Node.js • Express • MongoDB • Redis • Apache Solr • Nginx / Ubuntu 11 / 2 x Linode VPS Friday, April 20, 12
4. CONFIGURATION • Partition your configuration by environment. • E.g. Config[process.env.NODE_ENV] • We have 3 environments: 2 dev and 1 production. • Environment is top-level namespace (‘dev’, ‘production’, etc) for everything, including db. Keeps worlds from colliding. • Makes life easy Friday, April 20, 12
5. CONFIGURATION 2 • dev env 1: sessions in memory, clean db each start. • dev env 2: persistent sessions & db, ssl. • production env: aggressive caching, ssl, live payments, clustered, fancier logging (more later), email/sms notifications. Friday, April 20, 12
6. LOGGING • Log early and often. • Using Winston with the MongoDB transport. • MongoDB logging: fast writes; pass json objs in log msgs; easy ad-hoc queries; easy lru via capped collections. • Client logging from browser: caught and logged to Google Analytics as custom event. Easy monitoring/aggregation. Friday, April 20, 12
7. DEPLOYMENT • Custom CI system in node that responds pushes to github • Post-Receive service hook. Github POSTs to us after a push. • CI will pull code if it’s the right branch (say ‘staging’), restart, and send SMS. • Easy and low touch. Texts from your node are nice. • Refactoring into a more modular/decoupled hook.io system. Friday, April 20, 12
8. ALSO • In development: use ‘nodemon’ to start server. • Watches all files in project and restarts node if anything changes. • Great with persistent sessions, test changes immediately. • In production: use ‘forever’ to start server as daemon. Restarts node automatically if it crashes. Runs in background. Nice CLI. Manages stdout/stderr logs. Tracks multiple processes. Friday, April 20, 12
9. CLUSTERING • In production, use ‘cluster’ module to fork separate node per core • App-wide initialization: if cluster.isMaster • Otherwise per-proc initialization Friday, April 20, 12
10. DEPENDENCIES • Found out the hard way to control node, npm, and dependent module versions. You don’t want to push a change to production and have npm barf. No bueno. • Keep them the same on all machines. • Use npm in development. • In production I recommend bundling/controlling the deps and versions explicitly. Friday, April 20, 12
11. PROFILING • ‘ab’ is your friend. • don’t forget ulimit. • Joyent has great video on profiling with dtrace. • ‘nodetime’ module: great in development, but my server got hosed when I tried it in production. Probably user error. Friday, April 20, 12
12. THANKS! (you can email me at cbumgard@gmail.com) Beyond the Closed Beta, Node.js from code to production (at scale) Node.js ++ (Serverside Javascripters) Club SF Friday, April 20, 12