test: fix flaky router-luatest/router_2_2_test #625
Conversation
This commit fixes the flakiness of the router_2_2_test. Most failures were caused by low timeouts. In order to fix this, timeouts are increased, and `t.helpers.retrying` is used in test_replicaset_services_do_not_spam_same_errors and test_log_ratelimiter_is_dropped_on_replica_delete to fix other failures. Fixes tarantool#535 NO_DOC=testfix
| end, {g.replica_2_a:replicaset_uuid()}) | ||
| msg = "Error during master search" | ||
| g.router:wait_log_exactly_once(msg, {timeout = 0.1, on_yield = function() | ||
| g.router:wait_log_exactly_once(msg, {timeout = 0.5, on_yield = function() |
There was a problem hiding this comment.
How increase of waiting timeout makes the test less flaky? wait_log_exactly_once waits for timeout seconds and constantly checks the logs, so that the same message doesn't appear. By making timeout greater we just prolong the time of tests and do not make it less flaky
The same comment for all other places
| g.router:wait_log_exactly_once(msg, {timeout = 0.1, on_yield = function() | ||
| _G.failover_wakeup() | ||
| end}) | ||
| t.helpers.retrying({}, function() |
There was a problem hiding this comment.
We cannot retry here, the whole sense of wait_log_exactly_once is that logs are not duplicated. And if it failed once, it'll fail on all iterations, since the whole log after restart of the instance is checked
| _G.sleep_cond:wait() | ||
| end | ||
| end) | ||
| g.router:exec(function(bid, uuid) |
There was a problem hiding this comment.
Please, update the issue #535 with fails you encountered and fixing in the current patch. Otherwise, it's impossible to understand, what we're fixing here
2 participants
This commit fixes the flakiness of the router_2_2_test. Most failures were caused by low timeouts.
In order to fix this, timeouts are increased, and
t.helpers.retryingis used in test_replicaset_services_do_not_spam_same_errors and test_log_ratelimiter_is_dropped_on_replica_delete to fix other failures.Fixes #535
NO_DOC=testfix