For those of you not familiar with PuSH, it is a simple, open, server-to-server web-hook-based extension to Atom (and RSS) pubsub (publish/subscribe) protocol. Servers communicating via the PubSubHubbub protocol can get near-instant notifications (via webhook callbacks) when a topic (feed URL) they're interested in is updated.
In a post to Techcrunch Josh Fraser described PuSH saying "Currently, getting updates on the web requires constant polling. Subscribers are forced to act like nagging children asking, “Are we there yet?” Subscribers must constantly ping the publisher to ask if there are new updates even if the answer is “no” 99% of the time. This is terribly inefficient, wastes resources, and makes it incredibly hard to find new content in as soon as it appears. Both protocols flip the current model on its head so that updates are event driven rather than request driven. By that I mean that both protocols eliminate the need for polling by essentially telling subscribers, “Don’t ask us if there’s anything new. We’ll tell you.”"
Recently it occured to me that PuSH seems ideally suited as a low cost (free) decentralized notification system similar to Amazon's CloudWatch service. Amazon's CloudWatch is a web service that provides monitoring for AWS cloud resources, such as Amazon EC2 . It provides customers with visibility into resource utilization, operational performance, and overall demand patterns—including metrics such as CPU utilization, disk reads and writes, and network traffic. Amazon offers a the service at a rate of $0.015 per hour for each Amazon EC2 instance you monitor or about $18/day for a 50 node EC2 cluster. Yes it's cheap, but this approach is even cheaper and works across both existing internal infrastructure as well as cloud based IaaS such as Amazon EC2 Enomaly ECP, VMware, etc.
Below is a rough outline of how you might use PuSH as a basis for monitoring and scaling your cloud infrastructure.
Step 1
- Find a cluster monitor that supports a RSS or Atom output. My recommendation is Ganglia, a scalable distributed monitoring system targeted at federations of clusters. Ganglia includes the option of an RSS output for notification / syndication of system metrics such as load, i/o, dead machines, etc.
Step 2
- Create a feed URL (a "topic") that declares its Hub server(s) using an Atom or RSS XML file, via . (Atom and RssFeeds are supported) Topic's could include new machines, dead machines, 5 min average load, average network usage or any other system metric you deem valuable. The hub(s) can be run by the publisher of the feed, or can be a community hub that anybody can use.
Step 3
- A subscriber (a server or application that's interested in a topic), initially fetches the RSS or Atom URL and requests to the feed's hub(s) and subscribes to any subsequent updates pushed from it. Basically being notified upon certain changes. A subscriber could be a system administration tool such as Puppet that automatically adjusts based on near realtime notifications sent over the PubSubHubbub protocol.
Step 4
- When the Publisher (Load Monitor) next updates the Topic URL (i.e. Cluster ABC Load Average), the publisher pings the Hub(s) saying that there's an update.
Step 5
- The hub efficiently fetches the published feed and multicasts the new/changed content out to all registered subscribers where the subscribers then automatically can scale or adapt the infrastructure. Think of announcing a dead server to a cluster of hundreds of slaves. Or the addition of a new server.
The possibilities are quite interesting to say the least.
