From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Date: Fri, 28 Dec 2018 10:58:12 +0300 From: Vladimir Davydov Subject: Re: [tarantool-patches] Re: [PATCH v3 1/3] box: get rid of atexit() for calling cleanup routine Message-ID: <20181228075812.nhtchuubtnxkxyuj@esperanza> References: <98b93e4fd2d6d6affb4dc191ad5d4e8432ffefaa.1544533138.git.sergepetrenko@tarantool.org> <20181227124355.hmoioqgodmkcjclu@esperanza> <20181227125058.GH18166@chai> <20181227125620.2o7xgjts5pmljesm@esperanza> <20181228074326.GA8619@chai> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <20181228074326.GA8619@chai> To: Konstantin Osipov Cc: tarantool-patches@freelists.org, Serge Petrenko List-ID: On Fri, Dec 28, 2018 at 10:43:26AM +0300, Konstantin Osipov wrote: > * Vladimir Davydov [18/12/27 16:00]: > > > > On Tue, Dec 11, 2018 at 04:24:38PM +0300, Serge Petrenko wrote: > > > > > Move a call to tarantool_free() to the end of main(). > > > > > Also instead of breaking the event loop directly when processing a > > > > > signal start a fiber which can do some work before event loop break and > > > > > then break the event loop (this will be helpful when on_shutdown > > > > > triggers are implemented to run them while ev loop active). > > > > > Only wal_thread_stop() is left in atexit() to make sure we close the > > > > > journal in case the user exits by typing os.exit() to the console. > > > > > > > > I don't understand why you need to do it in the scope of the issue > > > > you're intending to fix. You can start a fiber from SITERM/SIGINT signal > > > > handler and run on_shutdown triggers from it without atexit rework, > > > > right? > > > > > > I asked Sergey to refactor our atexit machinery. I want atexit > > > handlers to be able to run the event loop, all such handlers, not > > > just the exit trigger. > > > > But this isn't what this patch does - it just moves some code from > > atexit() to the end of main, after the event loop is stopped. > > This is OK for existing handlers, since they don't require an > event loop today. Then why move them from atexit() to the end of main()? It doesn't achieve anything, it doesn't have anything to do with on_shutdown trigger. For some reason, we just stop freeing memory if the program terminates by exit()-ing.