Where Cassandra REALLY shines and is often overlooked is ease of maintenance. Cassandra's ability to bootstrap new nodes, replicate, reshard and handle down nodes (w/ hinted handoff) is almost magical. I use it in production and it works very reliably.
Sure, it's got some cool big data stuff, but try doing any of those "maintenance" operations on other databases without ripping your hair out. For example, even bringing up a new MySQL slave is a huge pain in the ass, let alone doing something non-trivial like promoting a new master.
I agree completely. By the time you need something like Cassandra you have typically had so many problems with your previous database that you just assume that anything you move to will be ultra-complex, but Cassandra is so simple to keep running that you can almost forget that it's running at all.
Even the developers who were resistant due to the different data model have completely embraced it.
Sure, those things can be a pain if you're relatively new to MySQL administration. But luckily there's a huge wealth of automation options released by some of the world's largest web sites.
One of MySQL's greatest strengths is the maturity of its ecosystem. Tons of existing automation, a relatively large number of skilled developers you can hire, multiple dedicated conferences, good free online docs, dozens of books, several top-notch consulting firms for handling the crazy edge cases.
I'll readily admit that MySQL has its flaws, but personally I wouldn't rank difficulty of maintenance operations high on the list... at least from a practical perspective, and relative to other comparable data stores used at a massive scale.
In my experience, no matter what databases you choose, if you "go big" you can't ever escape the need for robust automation and a decent amount of in-house talent.
Agreed, we had a pleasant operational experience, except for some not-so-well documented details around auto-bootstrapping and host names, as briefly mentioned in the blog post.
But cluster elasticity and resiliency has been working really good so far.
Interesting. I thought sysadmin operations like MySQL sharding and master-promotion are usually automated/scripted? Not to say it makes it conceptually easier, but it limits the opportunities to lose hair to 1.
Really? I always heard it was a bit of a pain to manage (update a schema? reboot all nodes). Not sure if any of that has changed, or if it was ever true (never used it myself).
did you try riak? I haven't myself yet, but from what I read it is also very easy to maintain. I'm on a crossroads right now, wandering which one of cassandra or risk to use to replace mongo in our still-in-development project. stability and devops is a very strong factor in the decision.
Sure, it's got some cool big data stuff, but try doing any of those "maintenance" operations on other databases without ripping your hair out. For example, even bringing up a new MySQL slave is a huge pain in the ass, let alone doing something non-trivial like promoting a new master.