Hi, I understood it's too late to change the feature. I hope it will be revised in 9.4!
(2013/08/09 4:13), Josh Berkus wrote: > On 08/08/2013 11:01 AM, Andres Freund wrote: > >> I don't think anybody working on related areas of the code thinks it's >> rock solid. >> But anyway, I just don't see the downside of allowing problem >> analysis. You're free to do more testing, review, whatever before the >> release. > > I'm 100% with you that we ought to keep the slow failover code around > and accessible to debugging. What I'm asking is whether it should still > be explicitly available to users, and the default. Based on your > feedback, it's sounding like it should be. > In my opinion, the default behavior "fast promote" is ok. Because we don't have any explicit problem with it for now. And we shouldn't mention about "normal promote" in document. I think it will make user confused if do so. By the way, I'm thinking about when these trigger files(*) are unlinked. (*)Now I treat these three files 1) a file specified in trigger_file 2) ${PGDATA}/promote 3) ${PGDATA}/fast_promote Current source has a problem in some corner cases. It occurs by the timing of detecting these files. for example: ------ case1: createing 1) and executing "pg_ctl promote" occur at the same time. case2: creating 1) and promoting at recovery-end(by recovery_target_xxx) occur at the same time. ------ 1) is created, but promoting completes by another trigger. Both cases 1) remains on the server. If user doesn't know it and make a standby on the server, the standby will promote soon. I think this is not so big problem, but not user-friendly. Against this, I'm thinking unlinking these files before starting recovery. This should be fixed in 9.4 too? --------- Tomonari Katsumata -- Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org) To make changes to your subscription: http://www.postgresql.org/mailpref/pgsql-hackers