Autoscaling Node.js on AWS

A talk on how we autoscaled a simple Node.js microservice (for event ingestion) on AWS. It hopefully shows that autoscaling isn't necessarily hard to do.

Transcript

1. Auto Scaling Node.js @rodpetrovic — LNUG Meetup — November 26th, 2014
2. Ingestion TVbeat Pay-TV Platforms data ingestion API
3. API app.post '/events', (req, res) ->
 
 errors = validateEvent res.body
 
 if errors?
 res.status(400).send errors
 
 else
 writeToKinesis res.body
 res.status(202).send accepted: yes
4. Load Balancing TVbeat Pay-TV Platforms ELB EC2 micro instance
5. Health Check app.get '/status', (req, res) ->
 
 res.status(200).send alive: yes
6. Queue Size queueSize = 0 writeToKinesis = (event) -> queueSize++ params = { ... } kinesis.putRecord params, (err, data) -> queueSize--
7. Taking a Break app.get '/status', (req, res) -> if queueSize < maxQueueSize res.status(200).send alive: yes else res.status(503).send alive: no
8. “The Daily Curve”
9. Auto Scaling TVbeat Pay-TV Platforms auto scaling group
10. Monitoring setInterval (-> sendToCloudWatch 'queueSize', queueSize ), 1000
11. Scaling Policy • Set a threshold < initial maxQueueSize • If average queue size goes over the threshold, add n instances to the auto scaling group • If average queue size goes under the threshold, remove m instances from the auto scaling group • Cool down period in both cases (a few minutes)
12. Dying Gracefully process.on 'SIGTERM', -> maxQueueSize = 0
13. The Result • The auto scaling group had between 7 and 70 micro instances. • We were able to feed Kinesis, event by event, in near real time. • We did it for about 30% of the cost of an instance that would be big enough to do the same job.
14. Thanks @extraordinaire for helping me with this
15. Thank You