Move deletion of old GTID rows to slave background thread

This patch changes how old rows in mysql.gtid_slave_pos* tables are deleted.
Instead of doing it as part of every replicated transaction in
record_gtid(), it is done periodically (every @@gtid_cleanup_batch_size
transaction) in the slave background thread.

This removes the deletion step from the replication process in SQL or worker
threads, which could speed up replication with many small transactions. It
also decreases contention on the global mutex LOCK_slave_state. And it
simplifies the logic, eg. when a replicated transaction fails after having
deleted old rows.

With this patch, the deletion of old GTID rows happens asynchroneously and
slightly non-deterministic. Thus the number of old rows in
mysql.gtid_slave_pos can temporarily exceed @@gtid_cleanup_batch_size. But
all old rows will be deleted eventually after sufficiently many new GTIDs
have been replicated.

1 files changed, 34 insertions, 1 deletions

diff --git a/sql/slave.cc b/sql/slave.cc
index 2a8c977451d..03b730b2b00 100644
--- a/sql/slave.cc
+++ b/sql/slave.cc
@@ -465,6 +465,8 @@ static struct slave_background_gtid_pos_create_t {
   void *hton;
 } *slave_background_gtid_pos_create_list;
 
+static volatile bool slave_background_gtid_pending_delete_flag;
+
 
 pthread_handler_t
 handle_slave_background(void *arg __attribute__((unused)))
@@ -499,6 +501,7 @@ handle_slave_background(void *arg __attribute__((unused)))
   {
     slave_background_kill_t *kill_list;
     slave_background_gtid_pos_create_t *create_list;
+    bool pending_deletes;
 
     thd->ENTER_COND(&COND_slave_background, &LOCK_slave_background,
                     &stage_slave_background_wait_request,
@@ -508,13 +511,15 @@ handle_slave_background(void *arg __attribute__((unused)))
       stop= abort_loop || thd->killed || slave_background_thread_stop;
       kill_list= slave_background_kill_list;
       create_list= slave_background_gtid_pos_create_list;
-      if (stop || kill_list || create_list)
+      pending_deletes= slave_background_gtid_pending_delete_flag;
+      if (stop || kill_list || create_list || pending_deletes)
         break;
       mysql_cond_wait(&COND_slave_background, &LOCK_slave_background);
     }
 
     slave_background_kill_list= NULL;
     slave_background_gtid_pos_create_list= NULL;
+    slave_background_gtid_pending_delete_flag= false;
     thd->EXIT_COND(&old_stage);
 
     while (kill_list)
@@ -541,6 +546,17 @@ handle_slave_background(void *arg __attribute__((unused)))
       create_list= next;
     }
 
+    if (pending_deletes)
+    {
+      rpl_slave_state::list_element *list;
+
+      slave_background_gtid_pending_delete_flag= false;
+      list= rpl_global_gtid_slave_state->gtid_grab_pending_delete_list();
+      rpl_global_gtid_slave_state->gtid_delete_pending(thd, &list);
+      if (list)
+        rpl_global_gtid_slave_state->put_back_list(list);
+    }
+
     mysql_mutex_lock(&LOCK_slave_background);
   } while (!stop);
 
@@ -615,6 +631,23 @@ slave_background_gtid_pos_create_request(
 
 
 /*
+  Request the slave background thread to delete no longer used rows from the
+  mysql.gtid_slave_pos* tables.
+
+  This is called from time-critical rpl_slave_state::update(), so we avoid
+  taking any locks here. This means we may race with the background thread
+  to occasionally lose a signal. This is not a problem; any pending rows to
+  be deleted will just be deleted a bit later as part of the next batch.
+*/
+void
+slave_background_gtid_pending_delete_request(void)
+{
+  slave_background_gtid_pending_delete_flag= true;
+  mysql_cond_signal(&COND_slave_background);
+}
+
+
+/*
   Start the slave background thread.
 
   This thread is currently used for two purposes: