The new servicing model for SQL Server is out, with a great explanation from Glenn Berry (there's also a webcast). The summary is that going forward with SQL Server 2017, we won't get Service Packs anymore. I'm somewhat sad, as I was always looking for SPs as a once a year patch if I didn't have any major issues with the platform. This allowed me to keep up to date, but not constantly test smaller patches. I liked that, partly because I've been worried about the quality of CUs in the past. 

We've had Cumulative Updates (CU) for quite some time. While I've seen a few service packs pulled and re-released, that's been very rare with CUs. I do find that SPs might have more issues because building one is an out of process, disruptive project, whereas CUs have become a regular, repeatable part of the development process. That's not to say that the CUs are bug free. A few of them have caused issues, and that's a problem that I hope improves over time. However, I also can't imagine what it's like working on such a large codebase, for installation in a large number of environments. It has to be a tremendous challenge to test all cases and remove bugs. Not that there shouldn't be fewer issues, especially in some areas, but I'm not sure there will ever be a complete absence of bugs.

I've changed my mind on CUs, and I like the new servicing model for it's simplicity as well as the automation and deep testing that CUs get. There is no confusion, no resources at Microsoft maintaining an RTM and SP1 branch with patches. I dislike code merges, and I suspect they're often a source of unintended bugs, so the fewer that are done the better. Now we'll have SQL Server 2017 essentially with one branch of code and patches. We'll have CU1-11 in the first year and then CU 12-15 the next year. Across fix years, I'd expect that we'd go from roughly 16 CUs per SP + SP to potentially 28 CUs in a single stream across the five year mainstream support of the product.

I know there are potentially more issues discovered early in a version's life, and hopefully fewer over time as developers write more tests and learn to cover edge cases better. I suspect that we'll see fewer and fewer items in each patch over time. Maybe not right away in SQL 2017, but it seems as though the number of issues corrected in each CU declines over time, and I'd expect that testing and better coverage of feature corner cases will come in the future. Perhaps SQL 2019 or 2020 will have fewer issues in the first year with even more testing.

I know some people see this as a message that the product isn't well tested. That's a fair view, after all, doubling the number of patches in the first year could be taken either way. I prefer to view this as the new way software is being developed. Build a pipeline, including lots of testing and feedback, and be ready to respond quickly if things are broken. I hope that's what Microsoft is thinking, and based on their push for better security and quality, it's the view I'll adopt.