MDEV-742 XA PREPAREd transaction survive disconnect/server restart

Lifted long standing limitation to the XA of rolling it back at the
transaction's
connection close even if the XA is prepared.

Prepared XA-transaction is made to sustain connection close or server
restart.
The patch consists of

    - binary logging extension to write prepared XA part of
      transaction signified with
      its XID in a new XA_prepare_log_event. The concusion part -
      with Commit or Rollback decision - is logged separately as
      Query_log_event.
      That is in the binlog the XA consists of two separate group of
      events.

      That makes the whole XA possibly interweaving in binlog with
      other XA:s or regular transaction but with no harm to
      replication and data consistency.

      Gtid_log_event receives two more flags to identify which of the
      two XA phases of the transaction it represents. With either flag
      set also XID info is added to the event.

      When binlog is ON on the server XID::formatID is
      constrained to 4 bytes.

    - engines are made aware of the server policy to keep up user
      prepared XA:s so they (Innodb, rocksdb) don't roll them back
      anymore at their disconnect methods.

    - slave applier is refined to cope with two phase logged XA:s
      including parallel modes of execution.

This patch does not address crash-safe logging of the new events which
is being addressed by MDEV-21469.

CORNER CASES: read-only, pure myisam, binlog-*, @@skip_log_bin, etc

Are addressed along the following policies.
1. The read-only at reconnect marks XID to fail for future
   completion with ER_XA_RBROLLBACK.

2. binlog-* filtered XA when it changes engine data is regarded as
   loggable even when nothing got cached for binlog.  An empty
   XA-prepare group is recorded. Consequent Commit-or-Rollback
   succeeds in the Engine(s) as well as recorded into binlog.

3. The same applies to the non-transactional engine XA.

4. @@skip_log_bin=OFF does not record anything at XA-prepare
   (obviously), but the completion event is recorded into binlog to
   admit inconsistency with slave.

The following actions are taken by the patch.

At XA-prepare:
   when empty binlog cache - don't do anything to binlog if RO,
   otherwise write empty XA_prepare (assert(binlog-filter case)).

At Disconnect:
   when Prepared && RO (=> no binlogging was done)
     set Xid_cache_element::error := ER_XA_RBROLLBACK
     *keep* XID in the cache, and rollback the transaction.

At XA-"complete":
   Discover the error, if any don't binlog the "complete",
   return the error to the user.

Kudos
-----
Alexey Botchkov took to drive this work initially.
Sergei Golubchik, Sergei Petrunja, Marko Mäkelä provided a number of
good recommendations.
Sergei Voitovich made a magnificent review and improvements to the code.
They all deserve a bunch of thanks for making this work done!

6 files changed, 245 insertions, 9 deletions

diff --git a/storage/rocksdb/ha_rocksdb.cc b/storage/rocksdb/ha_rocksdb.cc
index 3b2314d3f42..bcf1b18f08e 100644
--- a/storage/rocksdb/ha_rocksdb.cc
+++ b/storage/rocksdb/ha_rocksdb.cc
@@ -3121,6 +3121,8 @@ protected:
     s_tx_list.erase(this);
     RDB_MUTEX_UNLOCK_CHECK(s_tx_list_mutex);
   }
+  virtual bool is_prepared()        { return false; };
+  virtual void detach_prepared_tx() {};
 };
 
 /*
@@ -3157,7 +3159,16 @@ class Rdb_transaction_impl : public Rdb_transaction {
 
   virtual bool is_writebatch_trx() const override { return false; }
 
- private:
+  bool is_prepared() {
+    return m_rocksdb_tx && rocksdb::Transaction::PREPARED == m_rocksdb_tx->GetState();
+  }
+
+  void detach_prepared_tx() {
+    DBUG_ASSERT(rocksdb::Transaction::PREPARED == m_rocksdb_tx->GetState());
+    m_rocksdb_tx = nullptr;
+  }
+
+private:
   void release_tx(void) {
     // We are done with the current active transaction object.  Preserve it
     // for later reuse.
@@ -3798,7 +3809,8 @@ static int rocksdb_close_connection(handlerton *const hton, THD *const thd) {
           "disconnecting",
           rc);
     }
-
+    if (tx->is_prepared())
+      tx->detach_prepared_tx();
     delete tx;
   }
   return HA_EXIT_SUCCESS;
@@ -5301,7 +5313,7 @@ static int rocksdb_init_func(void *const p) {
 #ifdef MARIAROCKS_NOT_YET
   rocksdb_hton->update_table_stats = rocksdb_update_table_stats;
 #endif // MARIAROCKS_NOT_YET
-  
+
   /*
   Not needed in MariaDB:
   rocksdb_hton->flush_logs = rocksdb_flush_wal;
diff --git a/storage/rocksdb/mysql-test/rocksdb/r/xa.result b/storage/rocksdb/mysql-test/rocksdb/r/xa.result
index 12ae2b474b6..8cb6f39bbac 100644
--- a/storage/rocksdb/mysql-test/rocksdb/r/xa.result
+++ b/storage/rocksdb/mysql-test/rocksdb/r/xa.result
@@ -1,6 +1,7 @@
-# 
-# MDEV-13155: XA recovery not supported for RocksDB (Just a testcase)
 #
+# MDEV-742 fixes
+#   MDEV-13155: XA recovery not supported for RocksDB
+# as well.
 call mtr.add_suppression("Found .* prepared XA transactions");
 connect  con1,localhost,root,,test;
 DROP TABLE IF EXISTS t1;
@@ -15,19 +16,55 @@ INSERT INTO t1 (a) VALUES (3);
 INSERT INTO t1 (a) VALUES (4);
 XA END 'xa2';
 XA PREPARE 'xa2';
+connect  con3,localhost,root,,test;
+XA START 'xa3';
+INSERT INTO t1 (a) VALUES (5);
+INSERT INTO t1 (a) VALUES (6);
+XA END 'xa3';
+XA PREPARE 'xa3';
+disconnect con3;
 connection default;
 SELECT * FROM t1;
 a
+Must be all three XA:s in
+XA RECOVER;
+formatID	gtrid_length	bqual_length	data
+1	3	0	xa3
+1	3	0	xa1
+1	3	0	xa2
 # restart
 connect  con3,localhost,root,,test;
 XA RECOVER;
 formatID	gtrid_length	bqual_length	data
+1	3	0	xa3
 1	3	0	xa1
 1	3	0	xa2
 XA ROLLBACK 'xa1';
 XA COMMIT 'xa2';
+XA ROLLBACK 'xa3';
+SELECT a FROM t1;
+a
+3
+4
+connect  con4,localhost,root,,test;
+XA START 'xa4';
+INSERT INTO t1 (a) VALUES (7);
+INSERT INTO t1 (a) VALUES (8);
+XA END 'xa4';
+XA PREPARE 'xa4';
+connection default;
+# Now restart through graceful shutdown
+# restart
+connect  con5,localhost,root,,test;
+Must have 'xa4'
+XA RECOVER;
+formatID	gtrid_length	bqual_length	data
+1	3	0	xa4
+XA COMMIT 'xa4';
 SELECT a FROM t1;
 a
 3
 4
+7
+8
 DROP TABLE t1;
diff --git a/storage/rocksdb/mysql-test/rocksdb/t/xa.test b/storage/rocksdb/mysql-test/rocksdb/t/xa.test
index f8f381f0580..0c23e71df8c 100644
--- a/storage/rocksdb/mysql-test/rocksdb/t/xa.test
+++ b/storage/rocksdb/mysql-test/rocksdb/t/xa.test
@@ -1,6 +1,7 @@
---echo # 
---echo # MDEV-13155: XA recovery not supported for RocksDB (Just a testcase)
 --echo #
+--echo # MDEV-742 fixes
+--echo #   MDEV-13155: XA recovery not supported for RocksDB
+--echo # as well.
  
 call mtr.add_suppression("Found .* prepared XA transactions");
 
@@ -22,17 +23,51 @@ INSERT INTO t1 (a) VALUES (3);
 INSERT INTO t1 (a) VALUES (4);
 XA END 'xa2';
 XA PREPARE 'xa2';
- 
+
+--connect (con3,localhost,root,,test)
+XA START 'xa3';
+INSERT INTO t1 (a) VALUES (5);
+INSERT INTO t1 (a) VALUES (6);
+XA END 'xa3';
+XA PREPARE 'xa3';
+--disconnect con3
+
 --connection default
 SELECT * FROM t1;
 
+--echo Must be all three XA:s in
+XA RECOVER;
+
 --let $shutdown_timeout= 0
 --source include/restart_mysqld.inc
  
 --connect (con3,localhost,root,,test)
 --disable_abort_on_error
-XA RECOVER;
+XA RECOVER; # like above
 XA ROLLBACK 'xa1';
 XA COMMIT 'xa2';
+XA ROLLBACK 'xa3';
 SELECT a FROM t1;
+
+--connect (con4,localhost,root,,test)
+XA START 'xa4';
+INSERT INTO t1 (a) VALUES (7);
+INSERT INTO t1 (a) VALUES (8);
+XA END 'xa4';
+XA PREPARE 'xa4';
+
+--connection default
+--echo # Now restart through graceful shutdown
+--source include/restart_mysqld.inc
+
+
+--connect (con5,localhost,root,,test)
+--disable_abort_on_error
+
+--echo Must have 'xa4'
+XA RECOVER;
+XA COMMIT 'xa4';
+
+SELECT a FROM t1;
+
 DROP TABLE t1;
diff --git a/storage/rocksdb/mysql-test/rocksdb_rpl/r/rpl_xa.result b/storage/rocksdb/mysql-test/rocksdb_rpl/r/rpl_xa.result
new file mode 100644
index 00000000000..86f73f2fc9d
--- /dev/null
+++ b/storage/rocksdb/mysql-test/rocksdb_rpl/r/rpl_xa.result
@@ -0,0 +1,61 @@
+include/master-slave.inc
+[connection master]
+connection master;
+create table ti (a int, b int) engine=innodb;
+create table t1 (a int, b int) engine=rocksdb;
+insert into ti values(0, 0);
+insert into t1 values(0, 0);
+xa start 't';
+insert into ti values(1, 2);
+insert into t1 values(1, 2);
+xa end 't';
+xa prepare 't';
+xa commit 't';
+connection slave;
+include/diff_tables.inc [master:t1, slave:t1]
+connection master;
+xa start 't';
+insert into ti values(3, 4);
+insert into t1 values(3, 4);
+xa end 't';
+xa prepare 't';
+xa rollback 't';
+connection slave;
+include/diff_tables.inc [master:t1, slave:t1]
+connection master;
+SET pseudo_slave_mode=1;
+create table t2 (a int) engine=rocksdb;
+xa start 't';
+insert into ti values (5, 6);
+insert into t1 values (5, 6);
+xa end 't';
+xa prepare 't';
+xa start 's';
+insert into ti values (7, 8);
+insert into t2 values (0);
+xa end 's';
+xa prepare 's';
+include/save_master_gtid.inc
+connection slave;
+include/sync_with_master_gtid.inc
+xa recover;
+formatID	gtrid_length	bqual_length	data
+1	1	0	t
+1	1	0	s
+connection master;
+xa commit 't';
+xa commit 's';
+SET pseudo_slave_mode=0;
+Warnings:
+Warning	1231	Slave applier execution mode not active, statement ineffective.
+xa start 'r';
+insert into t1 values(7, 8);
+xa end 'r';
+xa prepare 'r';
+xa commit 'r';
+connection slave;
+include/diff_tables.inc [master:t1, slave:t1]
+include/diff_tables.inc [master:t2, slave:t2]
+connection master;
+drop table ti, t1, t2;
+include/rpl_end.inc
diff --git a/storage/rocksdb/mysql-test/rocksdb_rpl/t/rpl_xa.inc b/storage/rocksdb/mysql-test/rocksdb_rpl/t/rpl_xa.inc
new file mode 100644
index 00000000000..253d9f16316
--- /dev/null
+++ b/storage/rocksdb/mysql-test/rocksdb_rpl/t/rpl_xa.inc
@@ -0,0 +1,84 @@
+#
+# This "body" file checks general properties of XA transaction replication
+# as of MDEV-7974, including XA of mixed engine branches.
+# Parameters:
+# --let rpl_xa_check= SELECT ...
+#
+connection master;
+create table ti (a int, b int) engine=innodb;
+create table t1 (a int, b int) engine=rocksdb;
+insert into ti values(0, 0);
+insert into t1 values(0, 0);
+xa start 't';
+insert into ti values(1, 2);
+insert into t1 values(1, 2);
+xa end 't';
+xa prepare 't';
+xa commit 't';
+
+sync_slave_with_master;
+let $diff_tables= master:t1, slave:t1;
+source include/diff_tables.inc;
+
+connection master;
+
+xa start 't';
+insert into ti values(3, 4);
+insert into t1 values(3, 4);
+xa end 't';
+xa prepare 't';
+xa rollback 't';
+
+sync_slave_with_master;
+let $diff_tables= master:t1, slave:t1;
+source include/diff_tables.inc;
+
+connection master;
+SET pseudo_slave_mode=1;
+create table t2 (a int) engine=rocksdb;
+xa start 't';
+insert into ti values (5, 6);
+insert into t1 values (5, 6);
+xa end 't';
+xa prepare 't';
+xa start 's';
+insert into ti values (7, 8);
+insert into t2 values (0);
+xa end 's';
+xa prepare 's';
+--source include/save_master_gtid.inc
+
+connection slave;
+source include/sync_with_master_gtid.inc;
+if ($rpl_xa_check)
+{
+  --eval $rpl_xa_check
+  if ($rpl_xa_verbose)
+  {
+    --eval SELECT $rpl_xa_check_lhs
+    --eval SELECT $rpl_xa_check_rhs
+  }
+}
+xa recover;
+
+connection master;
+xa commit 't';
+xa commit 's';
+SET pseudo_slave_mode=0;
+
+# pure rocksdb xa
+xa start 'r';
+insert into t1 values(7, 8);
+xa end 'r';
+xa prepare 'r';
+xa commit 'r';
+
+
+sync_slave_with_master;
+let $diff_tables= master:t1, slave:t1;
+source include/diff_tables.inc;
+let $diff_tables= master:t2, slave:t2;
+source include/diff_tables.inc;
+
+connection master;
+drop table ti, t1, t2;
diff --git a/storage/rocksdb/mysql-test/rocksdb_rpl/t/rpl_xa.test b/storage/rocksdb/mysql-test/rocksdb_rpl/t/rpl_xa.test
new file mode 100644
index 00000000000..34384c74ca9
--- /dev/null
+++ b/storage/rocksdb/mysql-test/rocksdb_rpl/t/rpl_xa.test
@@ -0,0 +1,7 @@
+source include/have_rocksdb.inc;
+source include/have_innodb.inc;
+source include/master-slave.inc;
+source include/have_binlog_format_row.inc;
+
+source rpl_xa.inc;
+source include/rpl_end.inc;