util: optimize for diagnosis of deadlock #16276

optimize for diagnosis of deadlock it is not intuitive we use pstack/thread backtrace all to get lock information, or the original debug information.

Can you post an example debug log section of the changes?

thank you for your question. let me try to be clear.

(1)while thread lock and unlock cs_main, the log show like this, LOCKCONTENTION: cs_main Locker: src\validation.cpp:2383

but we do not know which state it is, unlocked, or keep waiting for the lock.

(2)after insert of PrintLockContentionOwned(), now we know it is unlocked or keep waiting.

now the log show like this if it is unlocked: LOCKCONTENTION: cs_main Locker: src\validation.cpp:2383 LOCKOWNED after lock contention: cs_main Locker: src\validation.cpp:2383

Another case is that, it is keep waiting: LOCKCONTENTION: cs_main Locker: src\validation.cpp:2383

(3)in our case, there are many threads has locked cs_main, thread1, thread2, thread3, thread4, etc... and there is a deadlock which potential_deadlock_detected() can not decect because another lock is not CCriticalSection, but we could not known which thread keeps waiting for cs_main.

as LOCKOWNED log, now we know thread1, thread4, has owned and unlock cs_main, but thread2 and thread3 keep waiting for cs_main and may be in deadlock state.

Please, change the PR and commit title to something that summarizes the change. "optimize for diagnosis of deadlock" is too general, and this doesn't involve any optimization.

Also: need someone familiar with hunting down deadlock issues (maybe @TheBlueMatt) to comment on this change.

The benefit of this change does not seem to be sufficiently clear to reviewers. Closing this for now.