On Fri, May 12, 2017 at 6:40 PM, Josh Berkus <josh@berkus.org> wrote:
> * WAL support for Hash Indexes > > Crash safe Hash Indexes or ACID compliant Hash Indexes.
Crash safe is good.
Also worth having something there about them now being replication safe? "Crash and replication safe" sounds a bit weird though, but something along that line?
> Version 10 will have an unusually high number of backwards-incompatible > changes. It is critical that all users test it against their > applications and platforms as soon as possible. > > I don't think we need to say anything more than: > > Version 10 has a high number of backwards-incompatible changes. For a > list of these changes please see the [Release Notes](link to release > notes).
I disagree. As a rule, we don't break backwards compatibility so pg 10 is going to be a shock to a lot of people. We really haven't seen this quantity of breakage since 8.3, which was released nine years ago, long before the majority of our current users were using Postgres. Given that -- because of partiitoning and logical replication -- many users will want to upgrade to 10 the month it comes out, I think we need to point out *in detail* why they will want to do extra testing. At a minimum, this includes the change in version numbering, the renaming of xlog to wal, and dropping support for FEBE 1.0.
None of those are very likely to break *applications*.
Most applications don't talk to the xlog/wal functions. DBA tools and scripts and whatnot *do*, but it is probably worth distinguishing that from applications, because they're also usually different ownership within organisations. It is likely to break more or less all of those though.
And I really doubt protocol v1.0 dropping is going to hit a lot of people.