MDEV-37494 Diagnostics_area does not always contain apply error info

mysql-test/suite/galera_3nodes/r/galera_vote_majority_dml.result

@@ -0,0 +1,70 @@
+connection node_2;
+connection node_1;
+CREATE TABLE t1 (f1 INTEGER PRIMARY KEY, f2 BLOB) ENGINE=InnoDB;
+connect node_3, 127.0.0.1, root, , test, $NODE_MYPORT_3;
+connection node_3;
+SET GLOBAL wsrep_on=OFF;
+ALTER TABLE t1 MODIFY f2 LONGTEXT;
+SET GLOBAL wsrep_on=ON;
+INSERT INTO t1 VALUES (3, 'a');
+connection node_1;
+SHOW STATUS LIKE 'wsrep_cluster_status';
+Variable_name	Value
+wsrep_cluster_status	Primary
+connection node_2;
+SHOW STATUS LIKE 'wsrep_cluster_status';
+Variable_name	Value
+wsrep_cluster_status	Primary
+INSERT INTO t1 VALUES (2, 'a');
+connection node_3;
+SET SESSION wsrep_sync_wait=0;
+SET SESSION wsrep_on=OFF;
+# restart
+SET SESSION wsrep_on=ON;
+INSERT INTO t1 VALUES (3, 'a');
+connection node_1;
+SHOW CREATE TABLE t1;
+Table	Create Table
+t1	CREATE TABLE `t1` (
+  `f1` int(11) NOT NULL,
+  `f2` blob DEFAULT NULL,
+  PRIMARY KEY (`f1`)
+) ENGINE=InnoDB DEFAULT CHARSET=latin1 COLLATE=latin1_swedish_ci
+SELECT * FROM t1;
+f1	f2
+2	a
+3	a
+connection node_2;
+SHOW CREATE TABLE t1;
+Table	Create Table
+t1	CREATE TABLE `t1` (
+  `f1` int(11) NOT NULL,
+  `f2` blob DEFAULT NULL,
+  PRIMARY KEY (`f1`)
+) ENGINE=InnoDB DEFAULT CHARSET=latin1 COLLATE=latin1_swedish_ci
+SELECT * FROM t1;
+f1	f2
+2	a
+3	a
+connection node_3;
+SHOW CREATE TABLE t1;
+Table	Create Table
+t1	CREATE TABLE `t1` (
+  `f1` int(11) NOT NULL,
+  `f2` blob DEFAULT NULL,
+  PRIMARY KEY (`f1`)
+) ENGINE=InnoDB DEFAULT CHARSET=latin1 COLLATE=latin1_swedish_ci
+SELECT * FROM t1;
+f1	f2
+2	a
+3	a
+DROP TABLE t1;
+connection node_1;
+CALL mtr.add_suppression("Replica SQL: Column 1 of table 'test.t1' cannot be converted from type 'longblob' to type 'blob', Error_code: MY-013146");
+CALL mtr.add_suppression("Event 3 Write_rows_v1 apply failed: 3, seqno");
+connection node_2;
+CALL mtr.add_suppression("Replica SQL: Column 1 of table 'test.t1' cannot be converted from type 'longblob' to type 'blob', Error_code: MY-013146");
+CALL mtr.add_suppression("Event 3 Write_rows_v1 apply failed: 3, seqno");
+connection node_3;
+CALL mtr.add_suppression("Vote 0 \\(success\\) on (.*) is inconsistent with group. Leaving cluster.");
+CALL mtr.add_suppression("Plugin 'InnoDB' will be forced to shutdown");

mysql-test/suite/galera_3nodes/t/galera_vote_majority_dml.cnf

@@ -0,0 +1,5 @@
+!include ../galera_3nodes.cnf
+
+[mysqld]
+wsrep-ignore-apply-errors=0
+

mysql-test/suite/galera_3nodes/t/galera_vote_majority_dml.test

@@ -0,0 +1,67 @@
+#
+#  MDEV-37494: Inconsistency voting: create conditions where applying would
+#              fail on replicas in table_def::compatible_with() and check that
+#              that replicas survive and the primary (trx source) bails out.
+#
+--source include/galera_cluster.inc
+
+CREATE TABLE t1 (f1 INTEGER PRIMARY KEY, f2 BLOB) ENGINE=InnoDB;
+
+--connect node_3, 127.0.0.1, root, , test, $NODE_MYPORT_3
+--connection node_3
+SET GLOBAL wsrep_on=OFF;
+ALTER TABLE t1 MODIFY f2 LONGTEXT; # Introducing schema inconsistency
+SET GLOBAL wsrep_on=ON;
+INSERT INTO t1 VALUES (3, 'a'); # Nodes 1 and 2 should fail to apply this
+
+--connection node_1
+# Wait until node #3 leaves the cluster
+--let $wait_condition = SELECT VARIABLE_VALUE = 2 FROM performance_schema.global_status WHERE VARIABLE_NAME = 'wsrep_cluster_size'
+--source include/wait_condition.inc
+SHOW STATUS LIKE 'wsrep_cluster_status';
+
+--connection node_2
+--let $wait_condition = SELECT VARIABLE_VALUE = 2 FROM performance_schema.global_status WHERE VARIABLE_NAME = 'wsrep_cluster_size'
+# Wait until node #3 leaves the cluster
+--source include/wait_condition.inc
+SHOW STATUS LIKE 'wsrep_cluster_status';
+
+INSERT INTO t1 VALUES (2, 'a'); # Nodes 1 and 2 should successfully apply this
+
+--connection node_3
+SET SESSION wsrep_sync_wait=0;
+--let $wait_condition = SELECT VARIABLE_VALUE = 'Disconnected' FROM performance_schema.global_status WHERE VARIABLE_NAME = 'wsrep_cluster_status';
+--source include/wait_condition.inc
+SET SESSION wsrep_on=OFF;
+--source include/restart_mysqld.inc
+--source include/wait_wsrep_ready.inc
+SET SESSION wsrep_on=ON;
+
+INSERT INTO t1 VALUES (3, 'a'); # All nodes should successfully apply this
+
+# Check that consistency is restored
+--connection node_1
+SHOW CREATE TABLE t1;
+SELECT * FROM t1;
+
+--connection node_2
+SHOW CREATE TABLE t1;
+SELECT * FROM t1;
+
+--connection node_3
+SHOW CREATE TABLE t1;
+SELECT * FROM t1;
+
+DROP TABLE t1;
+
+--connection node_1
+CALL mtr.add_suppression("Replica SQL: Column 1 of table 'test.t1' cannot be converted from type 'longblob' to type 'blob', Error_code: MY-013146");
+CALL mtr.add_suppression("Event 3 Write_rows_v1 apply failed: 3, seqno");
+
+--connection node_2
+CALL mtr.add_suppression("Replica SQL: Column 1 of table 'test.t1' cannot be converted from type 'longblob' to type 'blob', Error_code: MY-013146");
+CALL mtr.add_suppression("Event 3 Write_rows_v1 apply failed: 3, seqno");
+
+--connection node_3
+CALL mtr.add_suppression("Vote 0 \\(success\\) on (.*) is inconsistent with group. Leaving cluster.");
+CALL mtr.add_suppression("Plugin 'InnoDB' will be forced to shutdown");

sql/wsrep_applier.cc

@@ -83,9 +83,35 @@ wsrep_get_apply_format(THD* thd)
   return thd->wsrep_rgi->rli->relay_log.description_event_for_exec;
 }
 
-void wsrep_store_error(const THD* const thd,
-                       wsrep::mutable_buffer& dst,
-                       bool const include_msg)
+/* store error from rli */
+static void wsrep_store_error_rli(const THD* const thd,
+                                  wsrep::mutable_buffer& dst,
+                                  bool const include_msg)
+{
+  Slave_reporting_capability* const rli= thd->wsrep_rgi->rli;
+  if (rli && rli->last_error().number != 0)
+  {
+    auto error= rli->last_error();
+    std::ostringstream os;
+    if (include_msg)
+    {
+      os << error.message << ",";
+    }
+    os << " Error_code: " << error.number << ';';
+    std::string const err_str= os.str();
+    dst.resize(err_str.length() + 1);
+    sprintf(dst.data(), "%s", err_str.c_str());
+
+    WSREP_DEBUG("Error buffer (RLI) for thd %u seqno %lld, %zu bytes: '%s'",
+                thd->thread_id, (long long)wsrep_thd_trx_seqno(thd),
+                dst.size(), dst.size() ? dst.data() : "(null)");
+  }
+}
+
+/* store error from diagnostic area */
+static void wsrep_store_error_da(const THD* const thd,
+                                 wsrep::mutable_buffer& dst,
+                                 bool const include_msg)
 {
   Diagnostics_area::Sql_condition_iterator it=
     thd->get_stmt_da()->sql_conditions();
@@ -123,11 +149,35 @@ void wsrep_store_error(const THD* const thd,
 
   dst.resize(slider - dst.data());
 
-  WSREP_DEBUG("Error buffer for thd %llu seqno %lld, %zu bytes: '%s'",
+  WSREP_DEBUG("Error buffer (DA) for thd %llu seqno %lld, %zu bytes: '%s'",
               thd->thread_id, (long long)wsrep_thd_trx_seqno(thd),
               dst.size(), dst.size() ? dst.data() : "(null)");
 }
 
+/* store error info after applying error */
+void wsrep_store_error(const THD* const thd,
+                       wsrep::mutable_buffer& dst,
+                       bool const include_msg)
+{
+  dst.clear();
+  wsrep_store_error_da(thd, dst, include_msg);
+  if (dst.size() == 0)
+  {
+    wsrep_store_error_rli(thd, dst, include_msg);
+  }
+  if (dst.size() == 0)
+  {
+    WSREP_WARN("Failed to get apply error description from either "
+               "Relay_log_info or Diagnostics_area, will use random data.");
+    DBUG_ASSERT(0);
+    uintptr_t const n1= reinterpret_cast<uintptr_t>(&dst);
+    uintptr_t const n2= reinterpret_cast<uintptr_t>(thd);
+    uintptr_t const data= n1 ^ (n2 < 1);
+    const char* const data_ptr= reinterpret_cast<const char*>(&data);
+    dst.push_back(data_ptr, data_ptr + sizeof(data));
+  }
+}
+
 int wsrep_apply_events(THD*        thd,
                        Relay_log_info* rli,
                        const void* events_buf,