From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mail-lf1-f41.google.com (mail-lf1-f41.google.com [209.85.167.41]) (using TLSv1.2 with cipher ECDHE-RSA-AES128-GCM-SHA256 (128/128 bits)) (No client certificate requested) by dev.tarantool.org (Postfix) with ESMTPS id 5EBEB469710 for ; Fri, 5 Jun 2020 14:33:22 +0300 (MSK) Received: by mail-lf1-f41.google.com with SMTP id e125so5599668lfd.1 for ; Fri, 05 Jun 2020 04:33:22 -0700 (PDT) Date: Fri, 5 Jun 2020 14:33:19 +0300 From: Konstantin Osipov Message-ID: <20200605113319.GA172041@atlas> References: <7407b62139349ee3904a674490a6222b0c960a1c.1591292549.git.korablev@tarantool.org> <20200604225215.GB146885@atlas> <20200605111815.GA26017@tarantool.org> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <20200605111815.GA26017@tarantool.org> Subject: Re: [Tarantool-patches] [PATCH] vinyl: rotate mem during index build on demand List-Id: Tarantool development patches List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: Nikita Pettik Cc: tarantool-patches@dev.tarantool.org, v.shpilevoy@tarantool.org * Nikita Pettik [20/06/05 14:23]: > > > So to get rid of generations mismatch, let's rotate in-memory level > > > after yield on demand. > > > > Could you please also elaborate why generation mismatch is bad > > here? What does it lead to specifically? > > Because right after yield, vy_stmt is allocated using ls region > (in vy_build_insert_stmt()) and inserted into mem. Ls region does > not tolerate non-decreasing ids: otherwise region gc may truncate > data which is still accessible. Right, this is a gotcha I forgot about. The entire ls_region was a temporary crutch to get dumps working. Thanks for explanation. -- Konstantin Osipov, Moscow, Russia https://scylladb.com