tarantool / tarantool / 13628208179

88%

push

github

test: fix flaky replication/qsync_errinj

Firstly, in gh-5100 test it's impossible to change
`replication_timeout` to big value reliably, since it may
cause TimedOut error on replica, which won't reconnect,
since replication_timeout is 1000. For that we can use
`replication_reconnect_timeout`, introduced in the
previous commit.

Almost every line of the following test is flaky too, it is
caused by the first test, which sets `replication_timeout`
to 1000 and never resets that. The problem here is that
`replication_timeout` on master is 1000 and replica restarts
in the subsequent (gh-5123) test, which resets her timeout
to 0.1 second. It constantly times out and restarts replication,
but master sets a lot of error injections (e.g. ERRINJ_WAL_IO),
which may lead to critical error during subscribe (master
tries to create gc_consumer, so it needs to write during
subscribe). After that replica's upstream goes into stopped
state and never recovers, test times out. In order to
fix that `replication_timeout` is reset after first test.

Closes tarantool/tarantool-qa#333

NO_CHANGELOG=flaky test
NO_DOC=flaky test

69913 of 123395 branches covered (56.66%)

103011 of 117792 relevant lines covered (87.45%)

2223857.67 hits per line