(like in 1)

In this example Stress tests have found a real issue: after #42222 removal outdated parts from DiskS3 slowed down significantly. However, the only reason why we call MergeTreeData::clearOldPartsFromFilesystem in StorageMergeTree::shutdown is to avoid resurrection of parts dropped by DROP PARTITION after server restart. Of course it does not work and parts still can resurrect if server was killed. After #41145 we will be able to simply remove this call to clearOldPartsFromFilesystem.

I would prefer to fix the issue instead of increasing timeout. But maybe it's okay temporarily increase it and then decrease back after removing clearOldPartsFromFilesystem.

I would prefer to fix the issue instead of increasing timeout.

Yep, this is just a workaround to avoid noisy CI, and I think eventually it will be a problem even with increased timeout, but just not that frequent.

Why 5 functional tests failed????

Let's see is it stable

Why 5 functional tests failed????

Some of these tests are known to be flaky in debug build (they are just too slow), and usually they fail all together

@azat std::exception. Code: 1001, type: boost::wrapexcept<boost::program_options::unknown_option>, e.what() = unrecognised option '--timeout' (version 22.11.1.1360 (official build))

std::exception. Code: 1001, type: boost::wrapexceptboost::program_options::unknown_option, e.what() = unrecognised option '--timeout' (version 22.11.1.1360 (official build))

Yes, this is because it was used with the binary from previous version. Fixed.

FWIW Not a lot of difference after merging this (play).

It did not help.

@alexey-milovidov It hasn't been used at that time, since it was broken in #44197, , and was gotten back in minor cleanup - #45136