MDEV-4506: Parallel replication: intermediate commit.

Fix a bunch of issues found with locking, ordering, and non-thread-safe stuff
in Relay_log_info.

Now able to do a simple benchmark, showing 4.5 times speedup for applying a
binlog with 10000 REPLACE statements.

sql/log.h

@@ -408,7 +408,7 @@ private:
 
 class binlog_cache_mngr;
 struct rpl_gtid;
-class wait_for_commit;
+struct wait_for_commit;
 class MYSQL_BIN_LOG: public TC_LOG, private MYSQL_LOG
 {
  private:

sql/log_event.cc

@@ -9101,7 +9101,7 @@ int Rows_log_event::do_apply_event(struct rpl_group_info *rgi)
     do_apply_event(). We still check here to prevent future coding
     errors.
   */
-  DBUG_ASSERT(rli->sql_thd == thd);
+  DBUG_ASSERT(rgi->thd == thd);
 
   /*
     If there is no locks taken, this is the first binrow event seen

sql/log_event_old.cc

@@ -68,7 +68,7 @@ Old_rows_log_event::do_apply_event(Old_rows_log_event *ev, struct rpl_group_info
     do_apply_event(). We still check here to prevent future coding
     errors.
   */
-  DBUG_ASSERT(rli->sql_thd == ev_thd);
+  DBUG_ASSERT(rgi->thd == ev_thd);
 
   /*
     If there is no locks taken, this is the first binrow event seen
@@ -1481,7 +1481,7 @@ int Old_rows_log_event::do_apply_event(struct rpl_group_info *rgi)
     do_apply_event(). We still check here to prevent future coding
     errors.
   */
-  DBUG_ASSERT(rli->sql_thd == thd);
+  DBUG_ASSERT(rgi->thd == thd);
 
   /*
     If there is no locks taken, this is the first binrow event seen

sql/rpl_parallel.cc

@@ -38,15 +38,16 @@
      everything needs to be correctly rolled back and stopped in all threads,
      to ensure a consistent slave replication state.
 
-   - We need some knob on the master to allow the user to deliberately delay
-     commits waiting for more transactions to join group commit, to increase
-     potential for parallel execution on the slave.
-
    - Handle the case of a partial event group. This occurs when the master
      crashes in the middle of writing the event group to the binlog. The
      slave rolls back the transaction; parallel execution needs to be able
      to deal with this wrt. commit_orderer and such.
 
+   - Relay_log_info::is_in_group(). This needs to be handled correctly in all
+     callers. I think it needs to be split into two, one version in
+     Relay_log_info to be used from next_event() in slave.cc, one to be used in
+     per-transaction stuff.
+
    - We should fail if we connect to the master with opt_slave_parallel_threads
      greater than zero and master does not support GTID. Just to avoid a bunch
      of potential problems, we won't be able to do any parallel replication
@@ -58,12 +59,12 @@ struct rpl_parallel_thread_pool global_rpl_thread_pool;
 
 static void
 rpt_handle_event(rpl_parallel_thread::queued_event *qev,
-                 THD *thd,
                  struct rpl_parallel_thread *rpt)
 {
   int err;
   struct rpl_group_info *rgi= qev->rgi;
   Relay_log_info *rli= rgi->rli;
+  THD *thd= rgi->thd;
 
   thd->rli_slave= rli;
   thd->rpl_filter = rli->mi->rpl_filter;
@@ -143,6 +144,7 @@ handle_rpl_parallel_thread(void *arg)
       rpl_group_info *rgi= events->rgi;
       rpl_parallel_entry *entry= rgi->parallel_entry;
       uint64 wait_for_sub_id;
+      uint64 wait_start_sub_id;
       bool end_of_group;
 
       if (event_type == GTID_EVENT)
@@ -155,14 +157,28 @@ handle_rpl_parallel_thread(void *arg)
         /* Save this, as it gets cleared once event group commits. */
         event_gtid_sub_id= rgi->gtid_sub_id;
 
+        rgi->thd= thd;
+
         /*
           Register ourself to wait for the previous commit, if we need to do
           such registration _and_ that previous commit has not already
           occured.
+
+          Also do not start parallel execution of this event group until all
+          prior groups have committed that are not safe to run in parallel with.
         */
-        if ((wait_for_sub_id= rgi->wait_commit_sub_id))
+        wait_for_sub_id= rgi->wait_commit_sub_id;
+        wait_start_sub_id= rgi->wait_start_sub_id;
+        if (wait_for_sub_id || wait_start_sub_id)
         {
           mysql_mutex_lock(&entry->LOCK_parallel_entry);
+          if (wait_start_sub_id)
+          {
+            while (wait_start_sub_id > entry->last_committed_sub_id)
+              mysql_cond_wait(&entry->COND_parallel_entry,
+                              &entry->LOCK_parallel_entry);
+          }
+          rgi->wait_start_sub_id= 0;            /* No need to check again. */
           if (wait_for_sub_id > entry->last_committed_sub_id)
           {
             wait_for_commit *waitee=
@@ -176,7 +192,7 @@ handle_rpl_parallel_thread(void *arg)
         thd->wait_for_commit_ptr= &rgi->commit_orderer;
       }
 
-      rpt_handle_event(events, thd, rpt);
+      rpt_handle_event(events, rpt);
 
       end_of_group=
         in_event_group &&
@@ -376,6 +392,7 @@ err:
       while (new_free_list->running)
         mysql_cond_wait(&new_free_list->COND_rpl_thread,
                         &new_free_list->LOCK_rpl_thread);
+      mysql_mutex_unlock(&new_free_list->LOCK_rpl_thread);
       my_free(new_free_list);
       new_free_list= next;
     }
@@ -503,8 +520,7 @@ rpl_parallel::wait_for_done()
 
 
 bool
-rpl_parallel::do_event(struct rpl_group_info *serial_rgi, Log_event *ev,
+rpl_parallel::do_event(struct rpl_group_info *serial_rgi, Log_event *ev)
-                       THD *parent_thd)
 {
   rpl_parallel_entry *e;
   rpl_parallel_thread *cur_thread;
@@ -530,13 +546,34 @@ rpl_parallel::do_event(struct rpl_group_info *serial_rgi, Log_event *ev,
     Gtid_log_event *gtid_ev= static_cast<Gtid_log_event *>(ev);
 
     if (!(e= find(gtid_ev->domain_id)) ||
-        !(e->current_group_info= rgi= new rpl_group_info(rli)) ||
+        !(rgi= new rpl_group_info(rli)) ||
         event_group_new_gtid(rgi, gtid_ev))
     {
       my_error(ER_OUT_OF_RESOURCES, MYF(MY_WME));
       return true;
     }
 
+    if ((gtid_ev->flags2 & Gtid_log_event::FL_GROUP_COMMIT_ID) &&
+        e->last_commit_id == gtid_ev->commit_id)
+    {
+      /*
+        We are already executing something else in this domain. But the two
+        event groups were committed together in the same group commit on the
+        master, so we can still do them in parallel here on the slave.
+
+        However, the commit of this event must wait for the commit of the prior
+        event, to preserve binlog commit order and visibility across all
+        servers in the replication hierarchy.
+      */
+      rpl_parallel_thread *rpt= global_rpl_thread_pool.get_thread(e);
+      rgi->wait_commit_sub_id= e->current_sub_id;
+      rgi->wait_commit_group_info= e->current_group_info;
+      rgi->wait_start_sub_id= e->prev_groupcommit_sub_id;
+      e->rpl_thread= cur_thread= rpt;
+      /* get_thread() returns with the LOCK_rpl_thread locked. */
+    }
+    else
+    {
       /* Check if we already have a worker thread for this entry. */
       cur_thread= e->rpl_thread;
       if (cur_thread)
@@ -557,7 +594,24 @@ rpl_parallel::do_event(struct rpl_group_info *serial_rgi, Log_event *ev,
           thread to do this event group in parallel with anything else that might
           be running in other domains.
         */
-      if (gtid_ev->flags & Gtid_log_event::FL_GROUP_COMMIT_ID)
+        cur_thread= e->rpl_thread= global_rpl_thread_pool.get_thread(e);
+        /* get_thread() returns with the LOCK_rpl_thread locked. */
+      }
+      else
+      {
+        /*
+          We are still executing the previous event group for this replication
+          domain, and we have to wait for that to finish before we can start on
+          the next one. So just re-use the thread.
+        */
+      }
+
+      rgi->wait_commit_sub_id= 0;
+      rgi->wait_start_sub_id= 0;
+      e->prev_groupcommit_sub_id= e->current_sub_id;
+    }
+
+    if (gtid_ev->flags2 & Gtid_log_event::FL_GROUP_COMMIT_ID)
     {
       e->last_server_id= gtid_ev->server_id;
       e->last_seq_no= gtid_ev->seq_no;
@@ -569,38 +623,8 @@ rpl_parallel::do_event(struct rpl_group_info *serial_rgi, Log_event *ev,
       e->last_seq_no= 0;
       e->last_commit_id= 0;
     }
-      cur_thread= e->rpl_thread= global_rpl_thread_pool.get_thread(e);
-      rgi->wait_commit_sub_id= 0;
-      /* get_thread() returns with the LOCK_rpl_thread locked. */
-    }
-    else if ((gtid_ev->flags & Gtid_log_event::FL_GROUP_COMMIT_ID) &&
-             e->last_commit_id == gtid_ev->commit_id)
-    {
-      /*
-        We are already executing something else in this domain. But the two
-        event groups were committed together in the same group commit on the
-        master, so we can still do them in parallel here on the slave.
-
-        However, the commit of this event must wait for the commit of the prior
-        event, to preserve binlog commit order and visibility across all
-        servers in the replication hierarchy.
-      */
-      rpl_parallel_thread *rpt= global_rpl_thread_pool.get_thread(e);
-      rgi->wait_commit_sub_id= e->current_sub_id;
-      rgi->wait_commit_group_info= e->current_group_info;
-      e->rpl_thread= cur_thread= rpt;
-      /* get_thread() returns with the LOCK_rpl_thread locked. */
-    }
-    else
-    {
-      /*
-        We are still executing the previous event group for this replication
-        domain, and we have to wait for that to finish before we can start on
-        the next one. So just re-use the thread.
-      */
-      rgi->wait_commit_sub_id= 0;
-    }
 
+    e->current_group_info= rgi;
     e->current_sub_id= rgi->gtid_sub_id;
     current= rgi->parallel_entry= e;
   }
@@ -612,7 +636,7 @@ rpl_parallel::do_event(struct rpl_group_info *serial_rgi, Log_event *ev,
       but they might be from an old master).
     */
     qev->rgi= serial_rgi;
-    rpt_handle_event(qev, parent_thd, NULL);
+    rpt_handle_event(qev, NULL);
     delete_or_keep_event_post_apply(rli, typ, qev->ev);
 
     return false;

sql/rpl_parallel.h

@@ -60,6 +60,15 @@ struct rpl_parallel_entry {
   mysql_cond_t COND_parallel_entry;
   uint64 current_sub_id;
   struct rpl_group_info *current_group_info;
+  /*
+    The sub_id of the last event group in the previous batch of group-committed
+    transactions.
+
+    When we spawn parallel worker threads for the next group-committed batch,
+    they first need to wait for this sub_id to be committed before it is safe
+    to start executing them.
+  */
+  uint64 prev_groupcommit_sub_id;
 };
 struct rpl_parallel {
   HASH domain_hash;
@@ -69,7 +78,7 @@ struct rpl_parallel {
   ~rpl_parallel();
   rpl_parallel_entry *find(uint32 domain_id);
   void wait_for_done();
-  bool do_event(struct rpl_group_info *serial_rgi, Log_event *ev, THD *thd);
+  bool do_event(struct rpl_group_info *serial_rgi, Log_event *ev);
 };
 
 

sql/rpl_rli.cc

@@ -1226,7 +1226,7 @@ void Relay_log_info::stmt_done(my_off_t event_master_log_pos,
     middle of the "transaction". START SLAVE will resume at BEGIN
     while the MyISAM table has already been updated.
   */
-  if ((sql_thd->variables.option_bits & OPTION_BEGIN) && opt_using_transactions)
+  if ((rgi->thd->variables.option_bits & OPTION_BEGIN) && opt_using_transactions)
     inc_event_relay_log_pos();
   else
   {
@@ -1267,7 +1267,7 @@ void Relay_log_info::cleanup_context(THD *thd, bool error)
 {
   DBUG_ENTER("Relay_log_info::cleanup_context");
 
-  DBUG_ASSERT(sql_thd == thd);
+  DBUG_ASSERT(opt_slave_parallel_threads > 0 || sql_thd == thd);
   /*
     1) Instances of Table_map_log_event, if ::do_apply_event() was called on them,
     may have opened tables, which we cannot be sure have been closed (because
@@ -1534,8 +1534,8 @@ end:
 
 
 rpl_group_info::rpl_group_info(Relay_log_info *rli_)
-  : rli(rli_), gtid_sub_id(0), wait_commit_sub_id(0), wait_commit_group_info(0),
+  : rli(rli_), thd(0), gtid_sub_id(0), wait_commit_sub_id(0),
-    parallel_entry(0)
+    wait_commit_group_info(0), wait_start_sub_id(0), parallel_entry(0)
 {
   bzero(&current_gtid, sizeof(current_gtid));
 }

sql/rpl_rli.h

@@ -604,6 +604,7 @@ private:
 struct rpl_group_info
 {
   Relay_log_info *rli;
+  THD *thd;
   /*
     Current GTID being processed.
     The sub_id gives the binlog order within one domain_id. A zero sub_id
@@ -630,10 +631,19 @@ struct rpl_group_info
   */
   uint64 wait_commit_sub_id;
   struct rpl_group_info *wait_commit_group_info;
+  /*
+    If non-zero, the event group must wait for this sub_id to be committed
+    before the execution of the event group is allowed to start.
+
+    (When we execute in parallel the transactions that group committed
+    together on the master, we still need to wait for any prior transactions
+    to have commtted).
+  */
+  uint64 wait_start_sub_id;
 
   struct rpl_parallel_entry *parallel_entry;
 
-  rpl_group_info(Relay_log_info *rli);
+  rpl_group_info(Relay_log_info *rli_);
   ~rpl_group_info() { };
 };
 

sql/slave.cc

@@ -3246,7 +3246,7 @@ static int exec_relay_log_event(THD* thd, Relay_log_info* rli,
     }
 
     if (opt_slave_parallel_threads > 0)
-      DBUG_RETURN(rli->parallel.do_event(serial_rgi, ev, thd));
+      DBUG_RETURN(rli->parallel.do_event(serial_rgi, ev));
 
     /*
       For GTID, allocate a new sub_id for the given domain_id.
@@ -3995,6 +3995,7 @@ pthread_handler_t handle_slave_sql(void *arg)
   thd = new THD; // note that contructor of THD uses DBUG_ !
   thd->thread_stack = (char*)&thd; // remember where our stack is
   thd->rpl_filter = mi->rpl_filter;
+  serial_rgi.thd= thd;
 
   DBUG_ASSERT(rli->inited);
   DBUG_ASSERT(rli->mi == mi);