From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: From: Serge Petrenko Subject: [PATCH v4 0/3] replication: do not ignore replication_connect_quorum. Date: Tue, 14 Aug 2018 13:02:56 +0300 Message-Id: To: tarantool-patches@freelists.org Cc: georgy@tarantool.org, vdavydov.dev@gmail.com, Serge Petrenko List-ID: https://github.com/tarantool/tarantool/issues/3428 https://github.com/tarantool/tarantool/tree/sergepetrenko/gh-3428-replication-connect-quorum Previously replication_connect_quorum setting was ignored during initial bootstrap and during replication reconfiguration: the instance tried to connect to every other instance listed in box.cfg.replication, and threw an error, if it faild to do so. Now instance tries to connect to every other instance possible during replication_connect_timeout, but doesn't throw an error, if it was able to connect to replication_connect_quorum instances. First patch updates test-run to allow to pass arguments to instances started with create_cluster. Second patch utilizes the new functionality of test-run to update replication tests with easy to control start options for instances. Also alter on_replace.lua instance file to eliminate possible timing error, when executing code before box.once could cause an error "user 'cluster' is not found". Third patch introduces new replication_connect_quorum behaviour and adds a test case to check that replication_conenct_quorum isn't ignored anymore. Changes in v4: - split the patch into 3 separate patches. - update test-run to allow passing start arguments to create_cluster(). - update comments to match new behaviour. - in test instances when no timeout option is passed, set a default option. - be consistent when choosing timeouts for tests. Changes in v3: - Added a documentation request to TarantoolBot to commit message. - removed timeout parameter from box_sync_replication() and replicaset_connect() - rewritten replication tests to start instances with different replication_connect_timeout and replication_timeout parameters. This made tests run faster and more stable. - added a new test case to replication/quorum.test.lua to check that replication_connect_quorum isn't ignored anymore during bootstrap and reconfiguration. Changes in v2: - change test/replication/ddl.lua instance file to fix test failure on Travis. Serge Petrenko (3): test: update test-run Add arguments to replication test instances. replication: do not ignore replication_connect_quorum. src/box/box.cc | 35 +++++++++----- src/box/replication.cc | 11 +++-- src/box/replication.h | 7 +-- test-run | 2 +- test/replication-py/init_storage.test.py | 2 +- test/replication-py/master.lua | 1 + test/replication-py/replica.lua | 1 + test/replication/autobootstrap.lua | 6 ++- test/replication/autobootstrap.result | 4 +- test/replication/autobootstrap.test.lua | 4 +- test/replication/autobootstrap_guest.lua | 7 ++- test/replication/autobootstrap_guest.result | 2 +- test/replication/autobootstrap_guest.test.lua | 2 +- test/replication/before_replace.result | 8 ++-- test/replication/before_replace.test.lua | 8 ++-- test/replication/catch.result | 2 +- test/replication/catch.test.lua | 2 +- test/replication/ddl.lua | 7 ++- test/replication/ddl.result | 2 +- test/replication/ddl.test.lua | 2 +- test/replication/errinj.result | 6 +-- test/replication/errinj.test.lua | 6 +-- test/replication/master.lua | 1 + test/replication/master_quorum.lua | 7 ++- test/replication/misc.result | 2 +- test/replication/misc.test.lua | 2 +- test/replication/on_replace.lua | 14 ++++-- test/replication/on_replace.result | 2 +- test/replication/on_replace.test.lua | 2 +- test/replication/quorum.lua | 8 +++- test/replication/quorum.result | 65 ++++++++++++++++++++++---- test/replication/quorum.test.lua | 36 ++++++++++---- test/replication/rebootstrap.lua | 8 +++- test/replication/rebootstrap.result | 6 +-- test/replication/rebootstrap.test.lua | 6 +-- test/replication/recover_missing_xlog.result | 4 +- test/replication/recover_missing_xlog.test.lua | 4 +- test/replication/replica_no_quorum.lua | 3 +- test/replication/replica_quorum.lua | 24 ++++++++++ test/replication/replica_timeout.lua | 3 +- test/replication/replica_uuid_ro.lua | 7 ++- test/replication/replicaset_ro_mostly.result | 8 ++-- test/replication/replicaset_ro_mostly.test.lua | 8 ++-- 43 files changed, 251 insertions(+), 96 deletions(-) create mode 100644 test/replication/replica_quorum.lua -- 2.15.2 (Apple Git-101.1)