NodeSummit Slides

The slides from the RTDB presentation @ NodeSummit 2013

Transcript

1. Disrupting traditional database architectures with node.js Jim Mittler
2. Introduction !  Jim is the founder of Rheosoft, a software and professional services company. (The Greek word rhéos means “something flowing”). !  Currently serves as lead architect for web services and enterprise application integration (EAI) for BMW of North America. !  Some “notable” accomplishments…
3. 1979 - 11 year old runner-up DB’s Delight quiz show. – St. Louis Mo Note: Not me pictured.
4. 1997 – inventor of an adaptor to connect dictation microphone to PC via joystick port Note: Quickly made obsolete by USB and downfall of joystick port.
5. 2011 – builder/pilot of $500 race car for “24 hours of Lemons”; the “Pikachubaru”
6. Here is a (simplified) traditional database architecture Publisher Table Query Client Client
7. Traditional database architecture with Business Intelligence Publisher Table Query BI Application Client
8. According to Wikipedia…
9. Familiar “polling” option... http://www.klipfolio.com/solutions/real-time-dashboard
10. Reboot… !  Start with a requirement that analytics should be delivered in real-time. !  Introduce a web based “subscriber” modeled after message push. !  But instead of delivering individual messages – deliver the aggregated analytic results. !  Highly useful for executive dashboards, retail analytics, mobile real-time updates and many more.
11. A node.js database solution ! RTDB is a real-time JSON document database implemented entirely in node.js. ! Documents are inserted via REST API. ! Analytics are made available via map/reduce. ! When data changes, queries are updated in real- time and subscribers are updated immediately using HTML5 server-sent events.
12. RTDB architecture Publisher Collection Query Client Subscriber
13. RTDB protocols Publisher Collection Query Client Subscriber REST HTML5 Server Sent Event
14. Comparison SQL/Document DB !  Relational/JSON !  SQL or Map/Reduce !  DB specific Driver !  Polling !  Dynamic/Prepared !  Separate BI/Analytics RTDB !  JSON !  Map/Reduce !  REST !  Push !  Prepared !  Integrated Analytics
15. RTDB implementation Publisher Express Event Emitter Client Subscriber JSON over REST JSON via HTML5 Server Sent Event Configurable File System
16. Map Reduce Finalize Map/Reduce Pipeline Personalize organize by “key” Rollup/ aggregate Sorting/filter Apply user specific filter/ function Note: RTDB will perform an “incremental” map/ reduce when possible. This is very powerful. Add one record to a million record collection? No need to “remap” the entire dataset.
17. Reduce emit(values.reduce(function (a, b) { return a + b; })); Map emit(item.artist,1); Finalize reduction.sort(function(a,b) {return b[1] -a[1];}); if (reduction.length > 30) reduction.length = 30; emit(reduction); { artist: “Neil Diamond”, song: “Forever in Blue Jeans”, album: “You Don’t Bring me flowers”, year: 1978 } [[‘Neil Diamond’,1]]
18. Why node.js? ! node.js was the initial choice for prototyping Server Sent Events. ! As the prototype matured into an application, the application never outgrew node.js. ! node.js is interpreted so the map/reduce scripts can be changed at run- time. ! Very fast; very stable. ! Scalable I/O. Used heavily for files and subscribers. ! Async model is very well-suited for the internal architecture. ! NPM provides an excellent set of API building blocks for rapid assembly. ! Extremely PaaS friendly. (Heroku, Amazon EBS, Modulus, Cloudnode and more).
19. APIs leveraged via NPM •  Express – used for REST API and web admin •  Jade – templates for web admin •  AWS-SDK – access to the S3 file system •  Async – concurrency framework •  Symmetry – JSON delta processing •  Winston – logging Note: NPM offers several overlapping APIs. You are free to choose the best fit for your needs.
20. Configurable File System (CFS) •  RTDB can use a variety of backing stores; even other databases. •  Loads all javascript modules in specified directory via “requires” var cfslist = fs.readdirSync('./cfs'); var cfsTypes = {}; cfslist.forEach(function(file) { var cfs = require('./cfs/' + file); cfsTypes[cfs.name] = cfs; }); self.cfs = new cfsTypes[self.globalSettings.cfs](); self.cfs.init(self.globalSettings.cfsinit);
21. Configurable File System •  Small set of required methods: function name() - return a unique name for this provider function init(parms) - initialize with params from settings.json function exists(dir, callback) - does this exist? function get(key, callback) - return object by key function del(key, callback) - delete object by key function put(prefix, item, callback, expires) - put object function list(prefix, callback) - list objects
22. JSON simplifies everything. Here is all the code for inserting documents. app.post('/db/collections/:id/documents', function(req, res) { var c = database.collectionAt(req.params.id); var docs = []; if (!Array.isArray(req.body)) docs.push(req.body); else docs = req.body; c.put(docs, function(err) { if (!err) res.send(201); else res.send(500,err); }); }); Note: REST and JSON make it very easy to interact with the database using command line tools such as CURL.
23. Node.js events are the real-time glue •  Create emitter _emitter = new events.EventEmitter(); •  Register reduce function _emitter.once('change', doReduce); •  When there is work… emitter.emit('change'); •  Use “once” versus “on” to manage flow.
24. Async framework expedites concurent programming •  Async.each – process in parallel •  Async.eachSeries – process sequentially •  Load collections in priority order •  Async.eachLimit – parallel, but with limit •  Load files from file system without running out of system resources
25. Symmetry for wire protocol performance a = { x: 3, y: 5, z: 1 }; b = { x: 3, y: 8, z: 1 }; Symmetry.diff(a, b) # => { t: 'o', s: { y: 8 } } obj = { x: 3, y: 5, z: 1 }; diff = { t: 'o', s: { y: 8 } }; Symmetry.patch(obj, diff); obj # => { x: 3, y: 8, z: 1 } Example from https://github.com/Two-Screen/symmetry Symmetry will “delta” the JSON result set and can significantly reduce the bytes transferred.
26. GUIDs •  Every persistent object gets a GUID. •  Easy to share data between implementations. var uuid = require('node-uuid'); this._id = uuid.v4();
27. Demo https://rdtb.herokuapp.com •  Standalone •  Integration with Salesforce •  Constant streaming
28. For more info (I would love to hear from you.) jim@rheosoft.com Twitter @rheosoft http://facebook.com/rheosoft