MDEV-16264 - Fix assertion `m_queue.empty() && !m_tasks_running' in tpool::task_group destructor

This particular assertion happened when shutting down Innodb IO.IO shutdown properly waits for all IOs to finish


However there is a race condition -
right after releasing last IO slot and before decrementing task count
in group, pending_io_count will be 0, but tasks_running will be 1,
leading to assertion.

The fix is to make task_group destructor to wait for last running task
to finish.

1 files changed, 16 insertions, 1 deletions

diff --git a/tpool/task_group.cc b/tpool/task_group.cc
index ec9988ded6e..b52fe7c0f67 100644
--- a/tpool/task_group.cc
+++ b/tpool/task_group.cc
@@ -20,6 +20,9 @@ Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111 - 1301 USA*/
 #include <tpool_structs.h>
 #include <thread>
 #include <assert.h>
+#ifndef _WIN32
+#include <unistd.h> // usleep
+#endif
 namespace tpool
 {
   task_group::task_group(unsigned int max_concurrency) :
@@ -79,6 +82,18 @@ namespace tpool
 
   task_group::~task_group()
   {
-    assert(m_queue.empty() && !m_tasks_running);
+    std::unique_lock<std::mutex> lk(m_mtx);
+    assert(m_queue.empty());
+
+    while (m_tasks_running)
+    {
+      lk.unlock();
+#ifndef _WIN32
+      usleep(1000);
+#else
+      Sleep(1);
+#endif
+      lk.lock();
+    }
   }
 }