mleich provided rr replay traces of a 10.6 based testing branch where InnoDB shutdown would hang after shrinking the InnoDB buffer pool was initiated.

This might have been caused by MDEV-26827 in 10.6.13. The fix of MDEV-32029 did refactor some related code, but I don’t think the hang should be related to that. The reason for the hang is that buf_flush_wait_LRU_batch_end() (invoked by buf_pool_t::withdraw_blocks() when shrinking the buffer pool) would wait indefinitely for buf_pool.done_flush_LRU, and the buf_flush_page_cleaner() would wait indefinitely for buf_pool.do_flush_list. I think that the fix (to be tested by mleich) would be as follows:

diff --git a/storage/innobase/buf/buf0flu.cc b/storage/innobase/buf/buf0flu.cc

index 9826320a358..06670e317a4 100644

--- a/storage/innobase/buf/buf0flu.cc

+++ b/storage/innobase/buf/buf0flu.cc

@@ -2295,6 +2295,7 @@ static void buf_flush_page_cleaner()

     set_idle:

       buf_pool.page_cleaner_set_idle(true);

     set_almost_idle:

+      pthread_cond_broadcast(&buf_pool.done_flush_LRU);

       pthread_cond_broadcast(&buf_pool.done_flush_list);

       if (UNIV_UNLIKELY(srv_shutdown_state > SRV_SHUTDOWN_INITIATED))

         break;