On Thu, May 17, 2018 at 09:57:59AM -0400, Robert Haas wrote: > I think you and Tom have misunderstood each other somehow. If you > look at CommitTransaction(), you will see a comment that says:
Oh, I understood. You are right. > Also, there is no absolute prohibition on kernel calls in post-commit > cleanup, or in no-fail code in general. Thank you for the explanation! The current approach depends on syscache callbacks anyway. Backend 2 (from the example above) knows is it necessary to unpin segments after syscache callback was called. Tom pointed below that callbacks are occured in various events. So I think I should check the current approach too using CLOBBER_CACHE_ALWAYS. It could show some problems in the current patch. Then if everything is OK I think I'll check another approach (unmapping in TS syscache callback) using CLOBBER_CACHE_ALWAYS. -- Arthur Zakirov Postgres Professional: http://www.postgrespro.com Russian Postgres Company