From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from smtp3.mail.ru (smtp3.mail.ru [94.100.179.58]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by dev.tarantool.org (Postfix) with ESMTPS id B135D469719 for ; Sat, 29 Feb 2020 17:19:34 +0300 (MSK) References: <20200228170130.81713-1-sergepetrenko@tarantool.org> <468ada69-df04-0564-4c01-1ccab99b535c@tarantool.org> <1582969950.401106116@f137.i.mail.ru> From: Vladislav Shpilevoy Message-ID: <41560ae6-e2fb-1b7e-32d0-d596e73db8c5@tarantool.org> Date: Sat, 29 Feb 2020 15:19:32 +0100 MIME-Version: 1.0 In-Reply-To: <1582969950.401106116@f137.i.mail.ru> Content-Type: text/plain; charset="utf-8" Content-Language: en-US Content-Transfer-Encoding: 8bit Subject: Re: [Tarantool-patches] [PATCH] replication: fix rebootstrap in case the instance is listed in box.cfg.replication List-Id: Tarantool development patches List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: Serge Petrenko Cc: kirichenkoga@gmail.com, tarantool-patches@dev.tarantool.org On 29/02/2020 10:52, Serge Petrenko wrote: > Hi!  > > Суббота, 29 февраля 2020, 2:43 +03:00 от Vladislav Shpilevoy : >   > Thanks for the patch! > > On 28/02/2020 18:01, Serge Petrenko wrote: > > When checking wheter rejoin is needed, replica loops through all the > > instances in box.cfg.replication, which makes it believe that there is a > > master holding files, needed by it, since it accounts itself just like > > all other instances. > > So make replica skip itself when finding an instance which holds files > > needed by it, and determining whether rebootstrap is needed. > > > > We already have a working test for the issue, it missed the issue due to > > replica.lua replication settings. Fix replica.lua to optionally include > > itself in box.cfg.replication, so that the corresponding test works > > correctly. > > > > Closes #4759 > > --- > > https://github.com/tarantool/tarantool/issues/4759 > > https://github.com/tarantool/tarantool/tree/sp/gh-4759-rebootstrap-fix > > > > @ChangeLog > > - fix rebootstrap procedure not working in case replica itself > > is listed in `box.cfg.replication` > > > > src/box/replication.cc | 13 ++++++++++++- > > test/replication/replica.lua | 11 ++++++++++- > > test/replication/replica_rejoin.result | 12 ++++++------ > > test/replication/replica_rejoin.test.lua | 12 ++++++------ > > 4 files changed, 34 insertions(+), 14 deletions(-) > > > > diff --git a/src/box/replication.cc b/src/box/replication.cc > > index e7bfa22ab..01edc0fb2 100644 > > --- a/src/box/replication.cc > > +++ b/src/box/replication.cc > > @@ -768,8 +768,19 @@ replicaset_needs_rejoin(struct replica **master) > > struct replica *leader = NULL; > > replicaset_foreach(replica) { > > struct applier *applier = replica->applier; > > - if (applier == NULL) > > + /* > > + * The function is called right after > > + * box_sync_replication(), which in turn calls > > + * replicaset_connect(), which ensures that > > + * appliers are either stopped (APPLIER OFF) or > > + * connected. > > + * Also ignore self, as self applier might not > > + * have disconnected yet. > > + */ > > + if (applier == NULL || applier->state == APPLIER_OFF || > > + tt_uuid_is_equal(&replica->uuid, &INSTANCE_UUID)) > > continue; > > + assert(applier->state == APPLIER_CONNECTED); > > > > Could you please understand one thing? Below I see this: > > > const struct ballot *ballot = &applier->ballot; > > if (vclock_compare(&ballot->gc_vclock, > > &replicaset.vclock) <= 0) { > > /* > > * There's at least one master that still stores > > * WALs needed by this instance. Proceed to local > > * recovery. > > */ > > return false; > > } > > Question is why do we need rebootstrap if some remote node's > vclock is bigger than ours? It does not mean anything. It doesn't > say whether that remote instance still keeps any xlogs. All it > tells is that the remote instance committed some more data since > our restart. > >   > In the piece above we check for master’s gc vclock, which is vclock of the > oldest WAL the instance still has. If master’s gc vclock is less than our vclock, > no rebootstrap is needed. Master’ll be able to send us all the fresh data. >  Later we check whether our vclock is greater or incompatible with the remote > instance’s vclock. If it is, we cannot rebootstrap from the instance, since we have > some rows, that the instance doesn’t have. Thanks for the explanation. Then I would change the comment to say more about the issue. Currently it says about self only "Also ignore self, as self applier might not have disconnected yet." This does not look related to the actual problem, that we just should not look at our gc vclock. Also why do we need 'applier->state == APPLIER_OFF' check? How is it related to the issue? If this is a separate bug, is it possible to make it a separate patch?