Still not a bug. And no we can't, because concurrent session with older snapshot can subsequently use the index.
If you have a concrete improvement proposal, please start a thread in a suitable list (general or hackers).
On April 24, 2018 6:14:37 PM MDT, "德哥" <digoal@126.com> wrote:
but, there is no other session get any level lock with table a(creating index.). so if concurrenty create index with table a , at the end of the work, we can get a lock for table a and end the creating , like pg_repack?
在 2018-04-24 23:29:52，"David G. Johnston" <david.g.johnston@gmail.com> 写道：
On Tue, Apr 24, 2018 at 8:16 AM, Andres Freund <andres@anarazel.de> wrote:
On 2018-04-24 12:29:47 +0000, PG Bug reporting form wrote:
table a's copy will block table b's CONCURRENTLY index creating.
Yes, concurrently waits for other sessions to complete. I don't see a
bug here?
To be more specific a create index can stop processing as soon as its done its own work because it prevented any concurrent transactions from operating at the same time. With concurrently those other transactions can continue to work and thus the concurrent indexing performed by CREATE INDEX cannot know it is finished while other transactions hold write locks on the affected table/index. Instead it waits "to finish" until those transactions finish. The concurrent index creation has already started and is in progress - there just isn't any communication to the client as to progress.
David J.
--
Sent from my Android device with K-9 Mail. Please excuse my brevity.