From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Date: Fri, 8 Jun 2018 16:30:01 +0300 From: Vladimir Davydov Subject: Re: [RFC PATCH 06/12] box: refactor hot standby recovery Message-ID: <20180608133001.crnhnkq2jhlstjm3@esperanza> References: <27aabda1063798f83ca4415c01001838bded35c4.1528305232.git.vdavydov.dev@gmail.com> <20180608044045.GO6866@chai> <20180608064302.2xmkbnwwaovy5kg6@esperanza> <20180608131524.GD10854@chai> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <20180608131524.GD10854@chai> To: Konstantin Osipov Cc: tarantool-patches@freelists.org List-ID: On Fri, Jun 08, 2018 at 04:15:24PM +0300, Konstantin Osipov wrote: > * Vladimir Davydov [18/06/08 11:28]: > > No, it doesn't. See, even though it removes recover_remaining_wals() > > from recovery_finalize(), it makes hot_standby_f() call it before > > returning, so that we will rescan the xlog dir just before leaving > > hot standby mode: > > This is quite tricky. And I don't see why you need to do it this > way - why not add recover_remaining_wals to box_init instead, > where it will be more visible. Hmm, I guess you're right. I'll move the invocation of recover_remaining_wals from hot_standby_f to be called right after recovery_stop_local. Just so you understand why I'm doing this: if this patch set is applied, we will init replicaset.vclock before reading snapshot for normal recovery while in case of hot standby we will also have to promote it after recovery_stop_local. I want to separate those two paths so that we don't set replicaset.vclock twice in case of normal recovery. > > > > @@ -490,6 +486,7 @@ hot_standby_f(va_list ap) > > > > > > subscription.events = 0; > > > } > > > + recover_remaining_wals(r, stream, NULL, true); > > > return 0;