Re: [HACKERS] REINDEX CONCURRENTLY 2.0

On Thu, Jan 17, 2019 at 02:11:01PM +0900, Michael Paquier wrote:
> Okay, I have begun digging into the patch, and extracted for now two
> things which can be refactored first, giving a total of three patches:
> - 0001, which creates WaitForOlderSnapshots to snapmgr.c. I think
> that this can be useful for external extensions to have a process wait
> for snapshots older than a minimum threshold hold by other
> transactions.
> - 0002, which moves the concurrent index build into its own routine,
> index_build_concurrent(). At the same time, index_build() has a
> isprimary argument which is not used, so let's remove it. This
> simplifies a bit the refactoring as well.
> - 0003 is the core patch, realigned with the rest, fixing some typos I
> found on the way.

Are there any objections if I commit 0001? Introducing
WaitForOlderSnapshots() is quite independent from the rest, and the
refactoring is obvious. For 0002, I am still not 100% sure if
index_build_concurrent() is the best interface but I am planning to
look more at this stuff next week, particularly the drop portion which
needs more work.
--
Michael