MDEV-24278 InnoDB page cleaner keeps waking up on idle server

The purpose of the InnoDB page cleaner subsystem is to write out
modified pages from the buffer pool to data files. When the
innodb_max_dirty_pages_pct_lwm is not exceeded or
innodb_adaptive_flushing=ON decides not to write out anything,
the page cleaner should keep sleeping indefinitely until the state
of the system changes: a dirty page is added to the buffer pool such
that the page cleaner would no longer be idle.

buf_flush_page_cleaner(): Explicitly note when the page cleaner is idle.
When that happens, use mysql_cond_wait() instead of mysql_cond_timedwait().

buf_flush_insert_into_flush_list(): Wake up the page cleaner if needed.

innodb_max_dirty_pages_pct_update(),
innodb_max_dirty_pages_pct_lwm_update():
Wake up the page cleaner just in case.

Note: buf_flush_ahead(), buf_flush_wait_flushed() and shutdown are
already waking up the page cleaner thread.

3 files changed, 52 insertions, 4 deletions

diff --git a/storage/innobase/buf/buf0flu.cc b/storage/innobase/buf/buf0flu.cc
index 316c1822f0b..dc7ae3af70e 100644
--- a/storage/innobase/buf/buf0flu.cc
+++ b/storage/innobase/buf/buf0flu.cc
@@ -126,6 +126,20 @@ static void buf_flush_validate_skip()
 }
 #endif /* UNIV_DEBUG */
 
+/** Wake up the page cleaner if needed */
+inline void buf_pool_t::page_cleaner_wakeup()
+{
+  if (page_cleaner_idle() &&
+      (srv_max_dirty_pages_pct_lwm == 0.0 ||
+       srv_max_dirty_pages_pct_lwm <=
+       double(UT_LIST_GET_LEN(buf_pool.flush_list)) * 100.0 /
+       double(UT_LIST_GET_LEN(buf_pool.LRU) + UT_LIST_GET_LEN(buf_pool.free))))
+  {
+    page_cleaner_is_idle= false;
+    mysql_cond_signal(&do_flush_list);
+  }
+}
+
 /** Insert a modified block into the flush list.
 @param[in,out]	block	modified block
 @param[in]	lsn	oldest modification */
@@ -145,6 +159,7 @@ void buf_flush_insert_into_flush_list(buf_block_t* block, lsn_t lsn)
 
 	UT_LIST_ADD_FIRST(buf_pool.flush_list, &block->page);
 	ut_d(buf_flush_validate_skip());
+	buf_pool.page_cleaner_wakeup();
 	mysql_mutex_unlock(&buf_pool.flush_list_mutex);
 }
 
@@ -2067,8 +2082,12 @@ furious_flush:
     else if (srv_shutdown_state > SRV_SHUTDOWN_INITIATED)
       break;
 
-    mysql_cond_timedwait(&buf_pool.do_flush_list, &buf_pool.flush_list_mutex,
-                         &abstime);
+    if (buf_pool.page_cleaner_idle())
+      mysql_cond_wait(&buf_pool.do_flush_list, &buf_pool.flush_list_mutex);
+    else
+      mysql_cond_timedwait(&buf_pool.do_flush_list, &buf_pool.flush_list_mutex,
+                           &abstime);
+
     set_timespec(abstime, 1);
 
     lsn_limit= buf_flush_sync_lsn;
@@ -2091,6 +2110,8 @@ furious_flush:
         /* wake up buf_flush_wait_flushed() */
         mysql_cond_broadcast(&buf_pool.done_flush_list);
       }
+unemployed:
+      buf_pool.page_cleaner_set_idle(true);
       continue;
     }
 
@@ -2101,13 +2122,14 @@ furious_flush:
       double(UT_LIST_GET_LEN(buf_pool.LRU) + UT_LIST_GET_LEN(buf_pool.free));
 
     if (dirty_pct < srv_max_dirty_pages_pct_lwm && !lsn_limit)
-      continue;
+      goto unemployed;
 
     const lsn_t oldest_lsn= buf_pool.get_oldest_modification(0);
 
     if (UNIV_UNLIKELY(lsn_limit != 0) && oldest_lsn >= lsn_limit)
       buf_flush_sync_lsn= 0;
 
+    buf_pool.page_cleaner_set_idle(false);
     mysql_mutex_unlock(&buf_pool.flush_list_mutex);
 
     ulint n_flushed;
@@ -2159,6 +2181,11 @@ do_checkpoint:
         goto do_checkpoint;
       }
     }
+    else
+    {
+      mysql_mutex_lock(&buf_pool.flush_list_mutex);
+      goto unemployed;
+    }
 
 #ifdef UNIV_DEBUG
     while (innodb_page_cleaner_disabled_debug && !buf_flush_sync_lsn &&
diff --git a/storage/innobase/handler/ha_innodb.cc b/storage/innobase/handler/ha_innodb.cc
index b12d6cfd695..9a47fb6b30f 100644
--- a/storage/innobase/handler/ha_innodb.cc
+++ b/storage/innobase/handler/ha_innodb.cc
@@ -17207,6 +17207,7 @@ innodb_max_dirty_pages_pct_update(
 				    in_val);
 
 		srv_max_dirty_pages_pct_lwm = in_val;
+		mysql_cond_signal(&buf_pool.do_flush_list);
 	}
 
 	srv_max_buf_pool_modified_pct = in_val;
@@ -17240,6 +17241,7 @@ innodb_max_dirty_pages_pct_lwm_update(
 	}
 
 	srv_max_dirty_pages_pct_lwm = in_val;
+	mysql_cond_signal(&buf_pool.do_flush_list);
 }
 
 /*************************************************************//**
diff --git a/storage/innobase/include/buf0buf.h b/storage/innobase/include/buf0buf.h
index df52d6b572c..e255f6db056 100644
--- a/storage/innobase/include/buf0buf.h
+++ b/storage/innobase/include/buf0buf.h
@@ -1937,10 +1937,29 @@ public:
   FlushHp flush_hp;
   /** modified blocks (a subset of LRU) */
   UT_LIST_BASE_NODE_T(buf_page_t) flush_list;
-
+private:
+  /** whether the page cleaner needs wakeup from indefinite sleep */
+  bool page_cleaner_is_idle;
+public:
   /** signalled to wake up the page_cleaner; protected by flush_list_mutex */
   mysql_cond_t do_flush_list;
 
+  /** @return whether the page cleaner must sleep due to being idle */
+  bool page_cleaner_idle() const
+  {
+    mysql_mutex_assert_owner(&flush_list_mutex);
+    return page_cleaner_is_idle;
+  }
+  /** Wake up the page cleaner if needed */
+  inline void page_cleaner_wakeup();
+
+  /** Register whether an explicit wakeup of the page cleaner is needed */
+  void page_cleaner_set_idle(bool deep_sleep)
+  {
+    mysql_mutex_assert_owner(&flush_list_mutex);
+    page_cleaner_is_idle= deep_sleep;
+  }
+
   // n_flush_LRU + n_flush_list is approximately COUNT(io_fix()==BUF_IO_WRITE)
   // in flush_list