- Basho fanboy
- Basho customer
- enStratius product uses Riak
- thus all customers run Riak as part of install
- I am far from a subject matter expert
- Running Riak in production
- Riak itself
- Riak administration or tuning
- Basic definition - "A distributed system consists of multiple computers that communicate through a computer network."
- Jeff Hodges blog post - http://www.somethingsimilar.com/2013/01/14/notes-on-distributed-systems-for-young-bloods/
- which links to "The fallacies" - http://en.wikipedia.org/wiki/Fallacies_of_Distributed_Computing
- Read stuff by Kyle Kingsbury on the topic
- Read Boundary blog posts (I'll say this a lot)
- Single system problems are trivial but even single systems you have properties of distributed systems
- It's not just your systems at play. External APIs. Browsers.
- The network is reliable. (partition tolerance)
- Latency is zero. (eventual consistency)
- Bandwidth is infinite. (max node count)
- The network is secure. (riak provides SSL)
- Topology doesn't change.
- There is one administrator.
- Transport cost is zero.
- The network is homogeneous
- quick note about the logo
- cribbed from Justin Sheehey's Velocity talk
- Each dot contains the same diagram
- a distributed system within your distributed system
- and embraces them
- read the mailing list
- read basho blog posts
- Do you really need a relational model? 5th normal form? Really?
- What is your native format anyway?
- Can you loosen your consistency requirements?
- Querying is a pain in the ass (Riak+Redis)
- Read any Boundary blog post
- Read any mailing list post
- Riak let's you choose (tunable CAP, multiple backends)
- Riak doesn't try to be something it isn't
- Erlang isn't perfect. There are shitty Erlang apps.
- Alan Kay/Smalltalk/Message Passing
- Actor Pattern
- Let It Crash
- Basho contributes back to the Erlang ecosystem to make it more approachable (rebar, lager, webmachine)
- Revisit the fallacies
- Revisit CAP
- Embrace failure
- Fault Tolerance vs. Fault Prevention
- Your entire stack doesn't have the same requirements (strict consistency vs. weak consistency)
- Easy to reason about. (which is easier - let it crash or some complicated recovery scenario?)
- Network partitions happen.
- Be careful with clever
- Each additional component risks unnecessary additional complexity and reduces reasonability