From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mail-lf1-f47.google.com (mail-lf1-f47.google.com [209.85.167.47]) (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 34065469710 for ; Tue, 24 Nov 2020 18:24:11 +0300 (MSK) Received: by mail-lf1-f47.google.com with SMTP id v14so7247401lfo.3 for ; Tue, 24 Nov 2020 07:24:11 -0800 (PST) From: Cyrill Gorcunov Date: Tue, 24 Nov 2020 18:24:02 +0300 Message-Id: <20201124152405.1174898-1-gorcunov@gmail.com> MIME-Version: 1.0 Content-Transfer-Encoding: 8bit Subject: [Tarantool-patches] [PATCH v2 0/3] qsync: evaluate replication_synchro_quorum dynamically List-Id: Tarantool development patches List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: tml Cc: Vladislav Shpilevoy >From the issue description: > The reason is that users will likely not understand how the option > should be configured properly, and will break something accidentally. > So this idea about allowance to write an expression on the cluster > size allows to specify the canonical N/2 + 1 formula, and not update > it manually on all instances, when a new node is added, or an existing > one is deleted. Guys, take a look please once time permit. I decided to not split series too much since otherwise the overall picture disappear. What have been changed since Serge's comments: - keep replication_synchro_quorum been skipped at bootstrap in load_cfg.lua - eliminate redundant say_info calls - call quorum update routine from replica_set_id/replica_clear_id - use replicaset.registered_count directly when evaluating the formula - make quorum evaluation procedure always return value in allowed range, the only error which may happen here is some syntax error or Lua evaluation errors - a test has been added Now the things I don't like, and would prefer to remake somehow but not sure yet about a better way - box_renew_replication_synchro_quorum routine called every time replica_set_id or replica_clear_id is called, this is very weird to see box_ module routine from inside of a replicaset. Maybe we better setup some trigger and simply run it on set_id/clear_id, this way replicaset code will be more self consistent from box? Any comments are highly appreciated! issue https://github.com/tarantool/tarantool/issues/5446 branch gorcunov/gh-5446-eval-quorum-3 Cyrill Gorcunov (3): cfg: add cfg_isnumber helper cfg: support symbolic evaluation of replication_synchro_quorum test: add replication/gh-5446-sqync-eval-quorum.test.lua src/box/box.cc | 115 ++++++++++++- src/box/box.h | 1 + src/box/lua/load_cfg.lua | 2 +- src/box/replication.cc | 4 + src/cfg.c | 9 + src/cfg.h | 6 + .../gh-5446-sqync-eval-quorum.result | 156 ++++++++++++++++++ .../gh-5446-sqync-eval-quorum.test.lua | 62 +++++++ test/replication/replica-quorum-1.lua | 1 + test/replication/replica-quorum-2.lua | 1 + test/replication/replica-quorum-3.lua | 1 + test/replication/replica-quorum-4.lua | 1 + test/replication/replica-quorum.lua | 13 ++ 13 files changed, 367 insertions(+), 5 deletions(-) create mode 100644 test/replication/gh-5446-sqync-eval-quorum.result create mode 100644 test/replication/gh-5446-sqync-eval-quorum.test.lua create mode 120000 test/replication/replica-quorum-1.lua create mode 120000 test/replication/replica-quorum-2.lua create mode 120000 test/replication/replica-quorum-3.lua create mode 120000 test/replication/replica-quorum-4.lua create mode 100644 test/replication/replica-quorum.lua base-commit: 546499c9c001e30cbd2598946c9c0589b8e30f53 -- 2.26.2