From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from smtp38.i.mail.ru (smtp38.i.mail.ru [94.100.177.98]) (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 CC39B469719 for ; Mon, 9 Nov 2020 13:16:45 +0300 (MSK) References: <7ef30635b68970d001065f233ff83e6e292bada1.1604767356.git.v.shpilevoy@tarantool.org> From: Serge Petrenko Message-ID: <5ae0a854-46db-8ea4-4054-15b0319c7407@tarantool.org> Date: Mon, 9 Nov 2020 13:16:43 +0300 MIME-Version: 1.0 In-Reply-To: <7ef30635b68970d001065f233ff83e6e292bada1.1604767356.git.v.shpilevoy@tarantool.org> Content-Type: text/plain; charset="utf-8"; format="flowed" Content-Transfer-Encoding: 8bit Content-Language: en-GB Subject: Re: [Tarantool-patches] [PATCH 1/2] raft: fix crash on sm restart during WAL write List-Id: Tarantool development patches List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: Vladislav Shpilevoy , tarantool-patches@dev.tarantool.org, avtikhon@tarantool.org 07.11.2020 19:45, Vladislav Shpilevoy пишет: > Raft state machine crashed if was restarted during a WAL write > being in progress. When the machine was started, it didn't assume > there still can be a not finished WAL write from the time it was > enabled earlier. > > The patch makes it continue waiting for the write end. Hi!  Thanks  for the patch! LGTM. > > Part of #5506 > --- > src/box/raft.c | 13 ++++- > .../gh-5506-election-on-off.result | 55 +++++++++++++++++++ > .../gh-5506-election-on-off.test.lua | 31 +++++++++++ > test/replication/suite.cfg | 1 + > test/replication/suite.ini | 2 +- > 5 files changed, 98 insertions(+), 4 deletions(-) > create mode 100644 test/replication/gh-5506-election-on-off.result > create mode 100644 test/replication/gh-5506-election-on-off.test.lua > > diff --git a/src/box/raft.c b/src/box/raft.c > index 914b0d68f..3a99a0f26 100644 > --- a/src/box/raft.c > +++ b/src/box/raft.c > @@ -857,13 +857,20 @@ raft_sm_start(void) > { > say_info("RAFT: start state machine"); > assert(!ev_is_active(&raft.timer)); > - assert(!raft.is_write_in_progress); > assert(!raft.is_enabled); > assert(raft.state == RAFT_STATE_FOLLOWER); > raft.is_enabled = true; > raft.is_candidate = raft.is_cfg_candidate; > - if (!raft.is_candidate) { > - /* Nop. */; > + if (raft.is_write_in_progress) { > + /* > + * Nop. If write is in progress, the state machine is frozen. It > + * is continued when write ends. > + */ > + } else if (!raft.is_candidate) { > + /* > + * Nop. When a node is not a candidate, it can't initiate > + * elections anyway, so it does not need to monitor the leader. > + */ > } else if (raft.leader != 0) { > raft_sm_wait_leader_dead(); > } else { > diff --git a/test/replication/gh-5506-election-on-off.result b/test/replication/gh-5506-election-on-off.result > new file mode 100644 > index 000000000..1a718396f > --- /dev/null > +++ b/test/replication/gh-5506-election-on-off.result > @@ -0,0 +1,55 @@ > +-- test-run result file version 2 > +test_run = require('test_run').new() > + | --- > + | ... > + > +old_election_mode = box.cfg.election_mode > + | --- > + | ... > +old_replication_timeout = box.cfg.replication_timeout > + | --- > + | ... > + > +-- > +-- gh-5506: Raft state machine crashed in case there was a WAL write in > +-- progress, and Raft was disabled + enabled back immediately. It didn't assume > +-- that there can be a not finished WAL write when Raft is just enabled. > +-- > + > +-- Start a WAL write and wait when it starts. > +box.error.injection.set("ERRINJ_WAL_DELAY_COUNTDOWN", 0) > + | --- > + | - ok > + | ... > +box.cfg{ \ > + election_mode = 'candidate', \ > + replication_timeout = 0.1, \ > +} > + | --- > + | ... > +test_run:wait_cond(function() \ > + return box.error.injection.get("ERRINJ_WAL_DELAY") \ > +end) > + | --- > + | - true > + | ... > + > +-- Restart the state machine. It should notice the not finished WAL write and > +-- continue waiting. > +box.cfg{election_mode = 'off'} > + | --- > + | ... > +box.cfg{election_mode = 'candidate'} > + | --- > + | ... > +box.error.injection.set("ERRINJ_WAL_DELAY", false) > + | --- > + | - ok > + | ... > + > +box.cfg{ \ > + election_mode = old_election_mode, \ > + replication_timeout = old_replication_timeout, \ > +} > + | --- > + | ... > diff --git a/test/replication/gh-5506-election-on-off.test.lua b/test/replication/gh-5506-election-on-off.test.lua > new file mode 100644 > index 000000000..290408f06 > --- /dev/null > +++ b/test/replication/gh-5506-election-on-off.test.lua > @@ -0,0 +1,31 @@ > +test_run = require('test_run').new() > + > +old_election_mode = box.cfg.election_mode > +old_replication_timeout = box.cfg.replication_timeout > + > +-- > +-- gh-5506: Raft state machine crashed in case there was a WAL write in > +-- progress, and Raft was disabled + enabled back immediately. It didn't assume > +-- that there can be a not finished WAL write when Raft is just enabled. > +-- > + > +-- Start a WAL write and wait when it starts. > +box.error.injection.set("ERRINJ_WAL_DELAY_COUNTDOWN", 0) > +box.cfg{ \ > + election_mode = 'candidate', \ > + replication_timeout = 0.1, \ > +} > +test_run:wait_cond(function() \ > + return box.error.injection.get("ERRINJ_WAL_DELAY") \ > +end) > + > +-- Restart the state machine. It should notice the not finished WAL write and > +-- continue waiting. > +box.cfg{election_mode = 'off'} > +box.cfg{election_mode = 'candidate'} > +box.error.injection.set("ERRINJ_WAL_DELAY", false) > + > +box.cfg{ \ > + election_mode = old_election_mode, \ > + replication_timeout = old_replication_timeout, \ > +} > diff --git a/test/replication/suite.cfg b/test/replication/suite.cfg > index 8fd62fdb8..f2addebda 100644 > --- a/test/replication/suite.cfg > +++ b/test/replication/suite.cfg > @@ -16,6 +16,7 @@ > "gh-4424-misc-orphan-on-reconfiguration-error.test.lua": {}, > "gh-5426-election-on-off.test.lua": {}, > "gh-5433-election-restart-recovery.test.lua": {}, > + "gh-5506-election-on-off.test.lua": {}, > "once.test.lua": {}, > "on_replace.test.lua": {}, > "status.test.lua": {}, > diff --git a/test/replication/suite.ini b/test/replication/suite.ini > index 6136c934f..34ee32550 100644 > --- a/test/replication/suite.ini > +++ b/test/replication/suite.ini > @@ -3,7 +3,7 @@ core = tarantool > script = master.lua > description = tarantool/box, replication > disabled = consistent.test.lua > -release_disabled = catch.test.lua errinj.test.lua gc.test.lua gc_no_space.test.lua before_replace.test.lua qsync_advanced.test.lua qsync_errinj.test.lua quorum.test.lua recover_missing_xlog.test.lua sync.test.lua long_row_timeout.test.lua gh-4739-vclock-assert.test.lua gh-4730-applier-rollback.test.lua gh-5140-qsync-casc-rollback.test.lua gh-5144-qsync-dup-confirm.test.lua gh-5167-qsync-rollback-snap.test.lua > +release_disabled = catch.test.lua errinj.test.lua gc.test.lua gc_no_space.test.lua before_replace.test.lua qsync_advanced.test.lua qsync_errinj.test.lua quorum.test.lua recover_missing_xlog.test.lua sync.test.lua long_row_timeout.test.lua gh-4739-vclock-assert.test.lua gh-4730-applier-rollback.test.lua gh-5140-qsync-casc-rollback.test.lua gh-5144-qsync-dup-confirm.test.lua gh-5167-qsync-rollback-snap.test.lua gh-5506-election-on-off.test.lua > config = suite.cfg > lua_libs = lua/fast_replica.lua lua/rlimit.lua > use_unix_sockets = True -- Serge Petrenko