[C++] ScanningStress test is flaky in CI

[asfimport]

There is at least one nightly failure: https://github.com/ursacomputing/crossbow/actions/runs/3033965241/jobs/4882574634

Reporter: Weston Pace / @westonpace
Assignee: Percy Camilo Triveño Aucahuasi / @aucahuasi

Related issues:

• [CI][C++] Nightly test-ubuntu-20.04-cpp-thread-sanitizer fails arrow-utility-test around the AsyncTaskScheduler (is duplicated by)

Original Issue Attachments:

• backtrace.log.cpp

_{Note: This issue was originally created as ARROW-17687. Please see the migration documentation for further details.}

[asfimport]

Percy Camilo Triveño Aucahuasi / @aucahuasi:
I got this backtrace.log.cpp.

It seems we are moving the unique_locker and trying to lock some invalid mutex.

Also, I was able to get another issue, this time a deadlock using these values:

constexpr int kNumIters = 1;
constexpr int kNumFragments = 10;
constexpr int kBatchesPerFragment = 10;
constexpr int kNumConcurrentTasks = 2;

I'll try to explore more about where we are getting these errors, so far I was able to reduce and reproduce the test issue using these values:

constexpr int kNumIters = 1;
constexpr int kNumFragments = 2;
constexpr int kBatchesPerFragment = 1;
constexpr int kNumConcurrentTasks = 1;

Given that we can use C++ 17 now, I'll try to use the new std::scoped_lock instead of the other lockers (in the places where it make sense to do so)

[asfimport]

Antoine Pitrou / @pitrou:
@westonpace I marked this as blocker because we would like such crashiness issues to be solved before 10.0.0.

[asfimport]

Weston Pace / @westonpace:
Issue resolved by pull request 14314
#14314