If something went really wrong with thread pool, e.g. it used up all threads (due to bad configuration) and further flow got locked up, we need to print meaningful error messages in the server error log. The messages should

• be real errors – that is, not warnings, as warnings are often suppressed;
• be printed in a reasonable manner, without flooding the error log;
• identify problem as clearly as possible and maybe suggest further actions.

Additional warnings might also make sense, for example if thread pool approaches max_threads. Same requirements about not flooding the error log apply to them, too.