Thank you, ZFS
If you’ve had a technical conversation with anyone at Circonus, there are very likely two technologies that came up: ZFS and DTrace. While we love DTrace, ZFS has literally changed our world and made some personal “whoopsies” out of what could have been otherwise catastrophic, business-ending mistakes. ZFS has been a technology that has changed the way we interact with production computing systems and business problems.
ZFS has a million features that made it over a decade ahead of its time, but that’s not so important today. Today, we’re just thankful.
We want to thank the creators of ZFS and the community that placed such abusive production demands on it before us. We want to thank the community of people that saw the value of ZFS and ported it to… everything: FreeBSD, Linux, Mac OS X, and even Windows. It was a dreadful mistake that many vendors didn’t prop those communities up and assist them in making ZFS ubiquitous and default on their systems; you’ve done a disservice to your customers.
For 5 years, we didn’t bring “snowth” (IRONdb’s internal name) to market as a standalone product for a variety of reasons. One of the primary technical reasons was that our reliance on ZFS and Linux’s lack of adoption of ZFS made our deployable market artificially small. Last year, we decided that ZFS on Linux was “stable enough” to support our customers, and the last gating factor for IRONdb as a product was eliminated. I’m thankful for all of the ridiculously hard work the ZoL (ZFS on Linux) team has put into making ZFS as good on Linux as it is today.
At Circonus, under the torture harness of IRONdb, we’ve pushed on ZFS in ways that are hard to contemplate. Having dialogue with other ZFS developers and users, we know that we push on this filesystem in ways that are simply diabolical. If you look in the sizing section of the documentation of the excellent and capable InfluxDB, IRONdb sits squarely in their “infeasible” performance category; we attribute our ability to shatter these notions because we built out tech atop ZFS.
Yes, we use all the magical features therein: compression, device management (growing pools), online disk replacement, scrubbing, checksumming, snapshots, and many more, but today we want to show appreciation for the most important ZFS characteristic of all… stability. You’re just there for us. You’ve lost less data than any other filesystem we’ve used. You’ve crashed and locked up less than any of your counterparts. You’ve saved us years of filesystem checks, by completely eliminating them. You’ve saved us.