From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from [87.239.111.99] (localhost [127.0.0.1]) by dev.tarantool.org (Postfix) with ESMTP id 9FF696EC40; Tue, 8 Jun 2021 21:08:39 +0300 (MSK) DKIM-Filter: OpenDKIM Filter v2.11.0 dev.tarantool.org 9FF696EC40 DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=tarantool.org; s=dev; t=1623175719; bh=K1i2McL+ITkmfkOhfyLtX3zN5KoIuJg5fzrbuB48y8U=; h=To:Date:Subject:List-Id:List-Unsubscribe:List-Archive:List-Post: List-Help:List-Subscribe:From:Reply-To:Cc:From; b=LXv6MjR6bv+PTVwunSLCb6JnvLZTjYomFe5CQvGdm2A7f/Kq8770VPET2ERgzrTwZ kbEf3EP9lextfsm8VWyT2h7LYbKBp12yz012RdWfjMHoutVOgp6jgldgkxoPqYiTfi sVJzORRJsPMJU980013fr9k2P/myjauzc0t8l7zU= Received: from mail-lj1-f170.google.com (mail-lj1-f170.google.com [209.85.208.170]) (using TLSv1.3 with cipher TLS_AES_128_GCM_SHA256 (128/128 bits) key-exchange X25519 server-signature RSA-PSS (2048 bits) server-digest SHA256) (No client certificate requested) by dev.tarantool.org (Postfix) with ESMTPS id 5A04D6EC40 for ; Tue, 8 Jun 2021 21:08:38 +0300 (MSK) DKIM-Filter: OpenDKIM Filter v2.11.0 dev.tarantool.org 5A04D6EC40 Received: by mail-lj1-f170.google.com with SMTP id s22so7800689ljg.5 for ; Tue, 08 Jun 2021 11:08:38 -0700 (PDT) X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20161025; h=x-gm-message-state:from:to:cc:subject:date:message-id:mime-version :content-transfer-encoding; bh=a+BnEKywS8jVGWvQ88D8yfCIFzRT03+cJNKzAmcBg6o=; b=k/ibmSJnlVUdfxz7s0fRQzUgTgIS+tFFXFNgrXdJUpzTvds6HWDWVLDeKDVaWFQb53 kOqJCW6U/ba+8DWGrZpJsldr+CHll/CZSLRxXOAtrf5mE9l0hSHWypCt0eWJDAgs8UtZ ex0NS2+NQwHJGxFIf2pWp7HinxmVxUFtLY2C3sp9aKNv6Cu+Kk7nefubXEvlvXKidk/o E7UapV7fxuDqEYSqu2vS4F5LH4g7RCM108gffuwNMKtWW4hh6ItWXkEaZmw7fAed28/i 0LQfG3TXdFPrd7DZV3sConVrz1NtM3z5J9vdDxRZUcLtUJ+Hl4/0i/BuprQS3bjGNdos JRCA== X-Gm-Message-State: AOAM53323Pxt30PJfzj5HEFLRGPSBMrZgl/vT7Li+4BMYN6aQJTPtkJx uNygwBC4+iuYcrVOO2hRw8Evo8bzypA= X-Google-Smtp-Source: ABdhPJwiylYdHRfMfUkB8AlMeCimP+065wDJ2KrXBvxnMC58eT8RL47KWQp8zPpS9mVzh6KOB1vKWA== X-Received: by 2002:a05:651c:54a:: with SMTP id q10mr19383057ljp.420.1623175716861; Tue, 08 Jun 2021 11:08:36 -0700 (PDT) Received: from grain.localdomain ([5.18.171.94]) by smtp.gmail.com with ESMTPSA id i21sm55081ljb.10.2021.06.08.11.08.35 (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Tue, 08 Jun 2021 11:08:35 -0700 (PDT) Received: by grain.localdomain (Postfix, from userid 1000) id 1046A5A0042; Tue, 8 Jun 2021 21:08:35 +0300 (MSK) To: tml Date: Tue, 8 Jun 2021 21:08:33 +0300 Message-Id: <20210608180833.211678-1-gorcunov@gmail.com> X-Mailer: git-send-email 2.31.1 MIME-Version: 1.0 Content-Transfer-Encoding: 8bit Subject: [Tarantool-patches] [PATCH v5] applier: filter incoming synchro packets via transaction initiator X-BeenThere: tarantool-patches@dev.tarantool.org X-Mailman-Version: 2.1.34 Precedence: list List-Id: Tarantool development patches List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , From: Cyrill Gorcunov via Tarantool-patches Reply-To: Cyrill Gorcunov Cc: Vladislav Shpilevoy Errors-To: tarantool-patches-bounces@dev.tarantool.org Sender: "Tarantool-patches" Currently we use synchro packets filtration based on their contents, in particular by their xrow->replica_id value. Still there was a question if we can optimize this moment and rather filter our all packets coming from non-leader replica. Raft specification requires that only data from a current leader should be applied to local WAL but doesn't put a concrete claim on the data transport, ie how exactly rows are reaching replicas. This implies that data propagation may reach replicas indirectly via transit hops. Thus we drop applier->instance_id filtering and rely on xrow->replica_id matching instead. In the test (inspired by Serge Petrenko's test) we recreate the situation where replica3 obtains master's node data (which is a raft leader) indirectly via replica2 node. Closes #6035 Signed-off-by: Cyrill Gorcunov --- issue https://github.com/tarantool/tarantool/issues/6035 branch gorcunov/gh-6035-applier-filter-5 Guys, take a look please, would such description be verbose enough? Should we allow such configurations at all? src/box/applier.cc | 27 ++-- src/lib/raft/raft.h | 7 - .../replication/gh-6035-applier-filter.result | 147 ++++++++++++++++++ .../gh-6035-applier-filter.test.lua | 66 ++++++++ test/replication/gh6035master.lua | 1 + test/replication/gh6035node.lua | 35 +++++ test/replication/gh6035replica1.lua | 1 + test/replication/gh6035replica2.lua | 1 + 8 files changed, 266 insertions(+), 19 deletions(-) create mode 100644 test/replication/gh-6035-applier-filter.result create mode 100644 test/replication/gh-6035-applier-filter.test.lua create mode 120000 test/replication/gh6035master.lua create mode 100644 test/replication/gh6035node.lua create mode 120000 test/replication/gh6035replica1.lua create mode 120000 test/replication/gh6035replica2.lua diff --git a/src/box/applier.cc b/src/box/applier.cc index 33181fdbf..d3430f582 100644 --- a/src/box/applier.cc +++ b/src/box/applier.cc @@ -1027,21 +1027,24 @@ nopify:; * Return 0 for success or -1 in case of an error. */ static int -applier_apply_tx(struct applier *applier, struct stailq *rows) +applier_apply_tx(struct stailq *rows) { /* - * Rows received not directly from a leader are ignored. That is a - * protection against the case when an old leader keeps sending data - * around not knowing yet that it is not a leader anymore. + * Initially we've been filtering out data if it came from + * an applier which instance_id doesn't match raft->leader, + * but this prevents from obtaining valid leader's data when + * it comes from intermediate node. For example a series of + * replica hops * - * XXX: it may be that this can be fine to apply leader transactions by - * looking at their replica_id field if it is equal to leader id. That - * can be investigated as an 'optimization'. Even though may not give - * anything, because won't change total number of rows sent in the - * network anyway. + * master -> replica 1 -> replica 2 + * + * where each replica carries master's initiated transaction + * in xrow->replica_id field and master's data get propagated + * indirectly. + * + * Finally we dropped such "sender" filtration and use transaction + * "initiator" filtration via xrow->replica_id only. */ - if (!raft_is_source_allowed(box_raft(), applier->instance_id)) - return 0; struct xrow_header *first_row = &stailq_first_entry(rows, struct applier_tx_row, next)->row; struct xrow_header *last_row; @@ -1312,7 +1315,7 @@ applier_subscribe(struct applier *applier) diag_raise(); } applier_signal_ack(applier); - } else if (applier_apply_tx(applier, &rows) != 0) { + } else if (applier_apply_tx(&rows) != 0) { diag_raise(); } diff --git a/src/lib/raft/raft.h b/src/lib/raft/raft.h index a8da564b0..fae30b03d 100644 --- a/src/lib/raft/raft.h +++ b/src/lib/raft/raft.h @@ -236,13 +236,6 @@ raft_is_ro(const struct raft *raft) return raft->is_enabled && raft->state != RAFT_STATE_LEADER; } -/** See if the instance can accept rows from an instance with the given ID. */ -static inline bool -raft_is_source_allowed(const struct raft *raft, uint32_t source_id) -{ - return !raft->is_enabled || raft->leader == source_id; -} - /** Check if Raft is enabled. */ static inline bool raft_is_enabled(const struct raft *raft) diff --git a/test/replication/gh-6035-applier-filter.result b/test/replication/gh-6035-applier-filter.result new file mode 100644 index 000000000..cf032c332 --- /dev/null +++ b/test/replication/gh-6035-applier-filter.result @@ -0,0 +1,147 @@ +-- test-run result file version 2 +-- +-- gh-6035: verify synchronous rows filtration in applier, +-- we need to be sure that filtering synchronous rows is +-- done via transaction initiator not sender (iow via +-- xrow->replica_id). +-- + +local SOCKET_DIR = require('fio').cwd() + | --- + | ... +local function unix_socket(name) \ + return SOCKET_DIR .. "/" .. name .. '.sock' \ +end + | --- + | ... + +env = require('test_run') + | --- + | ... +test_run = env.new() + | --- + | ... +engine = test_run:get_cfg('engine') + | --- + | ... + +test_run:cmd('create server gh6035master with script="replication/gh6035master.lua"') + | --- + | - true + | ... +test_run:cmd('start server gh6035master with args="gh6035master"') + | --- + | - true + | ... + +test_run:cmd('create server gh6035replica1 with script="replication/gh6035replica1.lua"') + | --- + | - true + | ... +test_run:cmd('start server gh6035replica1 with args="gh6035replica1"') + | --- + | - true + | ... + +test_run:cmd('create server gh6035replica2 with script="replication/gh6035replica2.lua"') + | --- + | - true + | ... +test_run:cmd('start server gh6035replica2 with args="gh6035replica2"') + | --- + | - true + | ... + +-- +-- The gh6035master node connected to gh6035replica1 and gh6035replica2, +-- where each of gh6035replicaX connected to the gh6035master only. +-- +-- Lets reroute gh6035replica2 to gh6035replica1 so gh6035replica1 gonna +-- be sending data to gh6035replica2 as a transit hop from gh6035master. +test_run:cmd('switch gh6035replica2') + | --- + | - true + | ... +box.cfg{replication = {unix_socket("gh6035replica1")}} + | --- + | ... + +-- +-- Make the gh6035master to be RAFT leader. +test_run:cmd('switch gh6035master') + | --- + | - true + | ... +box.cfg{ \ + replication = { \ + unix_socket("gh6035master"), \ + unix_socket("gh6035replica1"), \ + }, \ + replication_synchro_quorum = 2, \ + election_mode = 'manual', \ +} + | --- + | ... +box.ctl.promote() + | --- + | ... +_ = box.schema.space.create("sync", {is_sync = true}) + | --- + | ... +_ = box.space.sync:create_index("pk") + | --- + | ... +box.space.sync:insert{1} + | --- + | - [1] + | ... + +-- The first hop is gh6035replica1. +test_run:cmd('switch gh6035replica1') + | --- + | - true + | ... +box.space.sync:select{} + | --- + | - - [1] + | ... + +-- And the second hop is gh6035replica2 where +-- gh6035replica1 replicated the data to us. +test_run:cmd('switch gh6035replica2') + | --- + | - true + | ... +box.space.sync:select{} + | --- + | - - [1] + | ... + +test_run:cmd('switch default') + | --- + | - true + | ... +test_run:cmd('stop server gh6035master') + | --- + | - true + | ... +test_run:cmd('delete server gh6035master') + | --- + | - true + | ... +test_run:cmd('stop server gh6035replica1') + | --- + | - true + | ... +test_run:cmd('delete server gh6035replica1') + | --- + | - true + | ... +test_run:cmd('stop server gh6035replica2') + | --- + | - true + | ... +test_run:cmd('delete server gh6035replica2') + | --- + | - true + | ... diff --git a/test/replication/gh-6035-applier-filter.test.lua b/test/replication/gh-6035-applier-filter.test.lua new file mode 100644 index 000000000..280ae4bae --- /dev/null +++ b/test/replication/gh-6035-applier-filter.test.lua @@ -0,0 +1,66 @@ +-- +-- gh-6035: verify synchronous rows filtration in applier, +-- we need to be sure that filtering synchronous rows is +-- done via transaction initiator not sender (iow via +-- xrow->replica_id). +-- + +local SOCKET_DIR = require('fio').cwd() +local function unix_socket(name) \ + return SOCKET_DIR .. "/" .. name .. '.sock' \ +end + +env = require('test_run') +test_run = env.new() +engine = test_run:get_cfg('engine') + +test_run:cmd('create server gh6035master with script="replication/gh6035master.lua"') +test_run:cmd('start server gh6035master with args="gh6035master"') + +test_run:cmd('create server gh6035replica1 with script="replication/gh6035replica1.lua"') +test_run:cmd('start server gh6035replica1 with args="gh6035replica1"') + +test_run:cmd('create server gh6035replica2 with script="replication/gh6035replica2.lua"') +test_run:cmd('start server gh6035replica2 with args="gh6035replica2"') + +-- +-- The gh6035master node connected to gh6035replica1 and gh6035replica2, +-- where each of gh6035replicaX connected to the gh6035master only. +-- +-- Lets reroute gh6035replica2 to gh6035replica1 so gh6035replica1 gonna +-- be sending data to gh6035replica2 as a transit hop from gh6035master. +test_run:cmd('switch gh6035replica2') +box.cfg{replication = {unix_socket("gh6035replica1")}} + +-- +-- Make the gh6035master to be RAFT leader. +test_run:cmd('switch gh6035master') +box.cfg{ \ + replication = { \ + unix_socket("gh6035master"), \ + unix_socket("gh6035replica1"), \ + }, \ + replication_synchro_quorum = 2, \ + election_mode = 'manual', \ +} +box.ctl.promote() +_ = box.schema.space.create("sync", {is_sync = true}) +_ = box.space.sync:create_index("pk") +box.space.sync:insert{1} + +-- The first hop is gh6035replica1. +test_run:cmd('switch gh6035replica1') +box.space.sync:select{} + +-- And the second hop is gh6035replica2 where +-- gh6035replica1 replicated the data to us. +test_run:cmd('switch gh6035replica2') +box.space.sync:select{} + +test_run:cmd('switch default') +test_run:cmd('stop server gh6035master') +test_run:cmd('delete server gh6035master') +test_run:cmd('stop server gh6035replica1') +test_run:cmd('delete server gh6035replica1') +test_run:cmd('stop server gh6035replica2') +test_run:cmd('delete server gh6035replica2') diff --git a/test/replication/gh6035master.lua b/test/replication/gh6035master.lua new file mode 120000 index 000000000..43b20c2cc --- /dev/null +++ b/test/replication/gh6035master.lua @@ -0,0 +1 @@ +gh6035node.lua \ No newline at end of file diff --git a/test/replication/gh6035node.lua b/test/replication/gh6035node.lua new file mode 100644 index 000000000..2fb1ce86f --- /dev/null +++ b/test/replication/gh6035node.lua @@ -0,0 +1,35 @@ +local SOCKET_DIR = require('fio').cwd() +local node_name = arg[1] + +function unix_socket(name) + return SOCKET_DIR .. "/" .. name .. '.sock'; +end + +require('console').listen(os.getenv('ADMIN')) + +if node_name == "gh6035master" then + box.cfg({ + listen = unix_socket("gh6035master"), + }) +elseif node_name == "gh6035replica1" then + box.cfg({ + listen = unix_socket("gh6035replica1"), + replication = { + unix_socket("gh6035master"), + unix_socket("gh6035replica1") + }, + election_mode = 'voter' + }) +else + assert(node_name == "gh6035replica2") + box.cfg({ + replication = { + unix_socket("gh6035master"), + }, + election_mode = 'voter' + }) +end + +box.once("bootstrap", function() + box.schema.user.grant('guest', 'super') +end) diff --git a/test/replication/gh6035replica1.lua b/test/replication/gh6035replica1.lua new file mode 120000 index 000000000..43b20c2cc --- /dev/null +++ b/test/replication/gh6035replica1.lua @@ -0,0 +1 @@ +gh6035node.lua \ No newline at end of file diff --git a/test/replication/gh6035replica2.lua b/test/replication/gh6035replica2.lua new file mode 120000 index 000000000..43b20c2cc --- /dev/null +++ b/test/replication/gh6035replica2.lua @@ -0,0 +1 @@ +gh6035node.lua \ No newline at end of file -- 2.31.1