Fix incorrect ThreadPool usage after ThreadPool introspection

[azat]

$ gg 'ThreadPool[^()]([A-Za-z_]\+,' src/
src/Interpreters/Context.cpp:        shared->load_marks_threadpool = std::make_unique<ThreadPool>(pool_size, pool_size, queue_size);
src/Interpreters/Context.cpp:        shared->prefetch_threadpool = std::make_unique<ThreadPool>(pool_size, pool_size, queue_size);
src/Interpreters/Context.cpp:        shared->threadpool_writer = std::make_unique<ThreadPool>(pool_size, pool_size, queue_size);

Fixes: #47880
Closes: #48234

Changelog category (leave one):

• Not for changelog (changelog entry is not required)

[tavplubix]

Do we have other incorrect usages? Let's make the compiler find them (it's okay to to in a separate PR)

[tavplubix]

Style check is broken in master, so it's impossible to merge anything...

[azat]

Do we have other incorrect usages? Let's make the compiler find them (it's okay to to in a separate PR)

The problem is that before #47880 there was the following ctor:

explicit ThreadPoolImpl(size_t max_threads_); /* 1 */

ThreadPoolImpl(
    size_t max_threads_,
    size_t max_free_threads_,
    size_t queue_size_,
    bool shutdown_on_exception_ = true); /* 2 */

After:

explicit ThreadPoolImpl(
    Metric metric_threads_,
    Metric metric_active_threads_,
    size_t max_threads_); /* 1 */

ThreadPoolImpl(
    Metric metric_threads_,
    Metric metric_active_threads_,
    size_t max_threads_,
    size_t max_free_threads_,
    size_t queue_size_,
    bool shutdown_on_exception_ = true); /* 2 */

And Metric is size_t, so compiler does not show any error when max_threads, max_free_threads, queue_size was passed as metric_threads, metric_active_threads, max_threads and ctor 1 was used instead of 2

[azat]

Yep, seems that this helps - https://s3.amazonaws.com/clickhouse-test-reports/0/46e85357ce2da2a99f56ee83a079e892d7ec3726/stress_test__asan_.html

[tavplubix]

And Metric is size_t, so compiler does not show any error when max_threads, max_free_threads, queue_size was passed as metric_threads, metric_active_threads, max_threads and ctor 1 was used instead of 2

Yes, I understand the problem. I mean that we can change ctors somehow, so it will be impossible to pass invalid arguments. It's great when the compiler can validate code. For example, we can create a structure like ThreadPoolSettings or try using StrongTypedef or something else.

[azat]

@tavplubix yeah, I thought about this, but this will require recompile lots of stuff, so decided to postpone this, anyway here is a PR, that uncovers another, last, bug - #48314