Changelog category (leave one):

• Not for changelog (changelog entry is not required)

Changelog entry (a user-readable short description of the changes that goes to CHANGELOG.md):

Don't hold NamedSessionsStorage::mutex while destroying stale sessions. Fixes #71307

Logs: https://pastila.nl/?000c4a4b/384edaa6b888a3d8fcfbf275b1e1fac2#YGtn/olp6K1dM4cz+3D8/w==

Thread 123121 got stuck somewhere for 6 seconds while getting a session for tx_async 19 "alter table test update value = 18 where id = 2". In the meantime, tx_async 19 "commit" was waiting for the previous query in tx_wait, but timeout 5 tail --pid=$query_pid -f /dev/null timed out in 5 seconds. Then we also look for the previous query in system.processes, but it wasn't there yet because thread 123121 was still stuck on getting session and didn't start processing the query. So tx_async 19 "commit" started executing and got "Session is locked by a concurrent client"

Looks like thread 123121 was waiting on the mutex in acquireSession. The lock was held by thread 10168 while cleaning old sessions and rolling back unfinished transactions: https://pastila.nl/?00f522cd/99e913c3fd765a0354d6279d59d086bc#iKrL5stckRaL46ceM1WEGg==

Let's destroy session objects without holding the lock

Documentation entry for user-facing changes

• Documentation is written (mandatory for new features)