MDEV-24845 Oddities around innodb_fatal_semaphore_wait_threshold and global.innodb_disallow_writes

This commit adds a mtr test for reproducing a test scenario where despite of
innodb_disallow_writes blocking, writes to file system can still happen.

The test launches a garbd node, which triggers one of the cluster node to switch to
SST donor state. In this state, all disk activity should be halted, and e.g.
innodb_disallow_writes has been set. The test records md5sum aggregate over mariadb
data directory when the node enters the donor state, and records another md5sum
when the node leaves the donor state. If there is no IO activity in data directory, these
hashes should be equal.

For this test, the Donor state processing, has beeen instrumented so that, SST donor thread can be
stopped when entering the donor state. The test uses this new dbug sync point,
to control when to record the md5sums.

New SST script was added: wsrep_sst_backup, and garbd uses backup method to lauch the donor
node to call this script, and to enter in donor state.

The backup script could be later extended as general purpose backup method for the cluster.

This commit fixes also one race condition happening in wsrep_sst_rsync, like this:
* wsrep_rsync_sst script requests for flush tables,
  and then waits in a loop until mariadbd has created file tables_flushed,
  as confirmation that FLUSH TABLES has completed
* mariadbd's SST donor thread, wakes for the flush table request and then performs FTWRL,
  and after this it creates the tables_flushed file
* note that SST script will now continue to startup rsync sending
* mariadbd's SST donor thread now calls for sst_disallow_writes(),
  so that innodb would setup disk IO blockage, however rsyncing may already be ongoing at this point

This race condition is fixed in this commit, by performing all disk IO blocking before
creating the tables_flushed file.

Reviewed-by: Jan Lindström <jan.lindstrom@mariadb.com>

3 files changed, 188 insertions, 0 deletions

diff --git a/mysql-test/suite/galera_3nodes/r/galera_garbd_backup.result b/mysql-test/suite/galera_3nodes/r/galera_garbd_backup.result
new file mode 100644
index 00000000000..f176ef1dd7f
--- /dev/null
+++ b/mysql-test/suite/galera_3nodes/r/galera_garbd_backup.result
@@ -0,0 +1,41 @@
+connection node_1;
+connection node_1;
+connection node_2;
+connection node_3;
+connection node_1;
+SET GLOBAL innodb_max_dirty_pages_pct=99;
+SET GLOBAL innodb_max_dirty_pages_pct_lwm=99;
+connection node_1;
+CREATE TABLE t1 (f1 INTEGER, f2 varchar(1024)) Engine=InnoDB;
+CREATE TABLE ten (f1 INTEGER) ENGINE=InnoDB;
+INSERT INTO ten VALUES (1),(2),(3),(4),(5),(6),(7),(8),(9),(10);
+INSERT INTO t1 (f2) SELECT REPEAT('x', 1024) FROM ten AS a1, ten AS a2, ten AS a3, ten AS a4;
+connection node_2;
+Killing node #3 to free ports for garbd ...
+connection node_3;
+connection node_1;
+SET GLOBAL debug_dbug = "+d,sync.wsrep_donor_state";
+Starting garbd ...
+SET SESSION debug_sync = "now WAIT_FOR sync.wsrep_donor_state_reached";
+SET GLOBAL innodb_max_dirty_pages_pct_lwm=0;
+SET GLOBAL innodb_max_dirty_pages_pct=0;
+SET SESSION debug_sync = "now SIGNAL signal.wsrep_donor_state";
+SET GLOBAL debug_dbug = "";
+SET debug_sync='RESET';
+connection node_2;
+Killing garbd ...
+connection node_1;
+connection node_2;
+DROP TABLE t1;
+DROP TABLE ten;
+Restarting node #3 to satisfy MTR's end-of-test checks
+connection node_3;
+connection node_1;
+SET GLOBAL innodb_max_dirty_pages_pct = 75.000000;
+SET GLOBAL innodb_max_dirty_pages_pct_lwm = 0.000000;
+connection node_1;
+CALL mtr.add_suppression("WSREP: Protocol violation\. JOIN message sender 1\.0 \(.*\) is not in state transfer \(SYNCED\)");
+connection node_2;
+CALL mtr.add_suppression("WSREP: Protocol violation\. JOIN message sender 1\.0 \(.*\) is not in state transfer \(SYNCED\)");
+connection node_3;
+CALL mtr.add_suppression("WSREP: Protocol violation\. JOIN message sender 1\.0 \(.*\) is not in state transfer \(SYNCED\)");
diff --git a/mysql-test/suite/galera_3nodes/t/galera_garbd_backup.cnf b/mysql-test/suite/galera_3nodes/t/galera_garbd_backup.cnf
new file mode 100644
index 00000000000..8b7cb948a87
--- /dev/null
+++ b/mysql-test/suite/galera_3nodes/t/galera_garbd_backup.cnf
@@ -0,0 +1,13 @@
+!include ../galera_3nodes.cnf
+
+[mysqld]
+wsrep_sst_method=rsync
+
+[mysqld.1]
+wsrep_node_name=node1
+
+[mysqld.2]
+wsrep_node_name=node2
+
+[mysqld.3]
+wsrep_node_name=node3
diff --git a/mysql-test/suite/galera_3nodes/t/galera_garbd_backup.test b/mysql-test/suite/galera_3nodes/t/galera_garbd_backup.test
new file mode 100644
index 00000000000..302bf430dde
--- /dev/null
+++ b/mysql-test/suite/galera_3nodes/t/galera_garbd_backup.test
@@ -0,0 +1,134 @@
+#
+# A very basic test for the galera arbitrator. We shut down node #3 and use its port allocation to start garbd.
+# As MTR does not allow multiple servers to be down at the same time, we are limited as to what we can test.
+#
+
+--source include/galera_cluster.inc
+--source include/have_innodb.inc
+--source include/have_garbd.inc
+--source include/big_test.inc
+--source include/have_debug.inc
+--source include/have_debug_sync.inc
+
+--connection node_1
+# Save original auto_increment_offset values.
+--let $node_1=node_1
+--let $node_2=node_2
+--let $node_3=node_3
+
+--let $galera_connection_name = node_3
+--let $galera_server_number = 3
+--source include/galera_connect.inc
+--source suite/galera/include/galera_base_port.inc
+--let $NODE_GALERAPORT_3 = $_NODE_GALERAPORT
+
+--source ../galera/include/auto_increment_offset_save.inc
+
+# Save galera ports
+--connection node_1
+--source suite/galera/include/galera_base_port.inc
+--let $NODE_GALERAPORT_1 = $_NODE_GALERAPORT
+--let $datadir= `SELECT @@datadir`
+
+--let $innodb_max_dirty_pages_pct = `SELECT @@innodb_max_dirty_pages_pct`
+--let $innodb_max_dirty_pages_pct_lwm = `SELECT @@innodb_max_dirty_pages_pct_lwm`
+
+SET GLOBAL innodb_max_dirty_pages_pct=99;
+SET GLOBAL innodb_max_dirty_pages_pct_lwm=99;
+
+--connection node_1
+CREATE TABLE t1 (f1 INTEGER, f2 varchar(1024)) Engine=InnoDB;
+CREATE TABLE ten (f1 INTEGER) ENGINE=InnoDB;
+INSERT INTO ten VALUES (1),(2),(3),(4),(5),(6),(7),(8),(9),(10);
+INSERT INTO t1 (f2) SELECT REPEAT('x', 1024) FROM ten AS a1, ten AS a2, ten AS a3, ten AS a4;
+ 
+--connection node_2
+--source suite/galera/include/galera_base_port.inc
+--let $NODE_GALERAPORT_2 = $_NODE_GALERAPORT
+
+--echo Killing node #3 to free ports for garbd ...
+--connection node_3
+--source include/shutdown_mysqld.inc
+
+--connection node_1
+--let $wait_condition = SELECT VARIABLE_VALUE = 2 FROM INFORMATION_SCHEMA.GLOBAL_STATUS WHERE VARIABLE_NAME = 'wsrep_cluster_size'
+--source include/wait_condition.inc
+
+# stop SST donor thread when node is in donor state
+SET GLOBAL debug_dbug = "+d,sync.wsrep_donor_state";
+
+--echo Starting garbd ...
+--exec $MTR_GARBD_EXE --address "gcomm://127.0.0.1:$NODE_GALERAPORT_1" --group my_wsrep_cluster --donor node1 --sst backup --options 'base_port=$NODE_GALERAPORT_3' > $MYSQL_TMP_DIR/garbd.log 2>&1 &
+
+SET SESSION debug_sync = "now WAIT_FOR sync.wsrep_donor_state_reached";
+
+#
+# get hash of data directory contents before BP dirty page flushing
+#
+--exec find $datadir -type f ! -name tables_flushed ! -name backup_sst_complete -exec md5sum {} \; | md5sum >$MYSQLTEST_VARDIR/tmp/innodb_before
+
+# this should force buffer pool flushing, if not already done by donor state change transfer
+SET GLOBAL innodb_max_dirty_pages_pct_lwm=0;
+SET GLOBAL innodb_max_dirty_pages_pct=0;
+
+--disable_query_log
+--disable_result_log
+select f1 from t1;
+select * from ten;
+--enable_result_log
+--enable_query_log
+
+#
+#
+# record the hash of data directory contents after BP dirty page flushing
+#
+--exec find $datadir -type f ! -name tables_flushed ! -name backup_sst_complete -exec md5sum {} \; | md5sum >$MYSQLTEST_VARDIR/tmp/innodb_after
+
+# there should be no disk writes
+--diff_files $MYSQLTEST_VARDIR/tmp/innodb_before $MYSQLTEST_VARDIR/tmp/innodb_after
+
+SET SESSION debug_sync = "now SIGNAL signal.wsrep_donor_state";
+SET GLOBAL debug_dbug = "";
+SET debug_sync='RESET';
+
+--connection node_2
+
+#
+# garbd will die automatically, because of the backup SST script
+# but just to be sure, sending explicit kill here, as well
+#
+--echo Killing garbd ...
+# FreeBSD's /bin/pkill only supports short versions of the options:
+# -o Select only the oldest (least recently started)
+# -f Match against full argument lists
+--error 0,1
+--exec pkill -o -f garbd.*$NODE_GALERAPORT_3
+
+--connection node_1
+--let $wait_condition = SELECT VARIABLE_VALUE = 2 FROM INFORMATION_SCHEMA.GLOBAL_STATUS WHERE VARIABLE_NAME = 'wsrep_cluster_size'
+--source include/wait_condition.inc
+
+--connection node_2
+
+DROP TABLE t1;
+DROP TABLE ten;
+
+--echo Restarting node #3 to satisfy MTR's end-of-test checks
+--connection node_3
+let $restart_noprint=2;
+--source include/start_mysqld.inc
+
+--connection node_1
+--eval SET GLOBAL innodb_max_dirty_pages_pct = $innodb_max_dirty_pages_pct
+--eval SET GLOBAL innodb_max_dirty_pages_pct_lwm = $innodb_max_dirty_pages_pct_lwm
+
+--source ../galera/include/auto_increment_offset_restore.inc
+
+--connection node_1
+CALL mtr.add_suppression("WSREP: Protocol violation\. JOIN message sender 1\.0 \(.*\) is not in state transfer \(SYNCED\)");
+
+--connection node_2
+CALL mtr.add_suppression("WSREP: Protocol violation\. JOIN message sender 1\.0 \(.*\) is not in state transfer \(SYNCED\)");
+
+--connection node_3
+CALL mtr.add_suppression("WSREP: Protocol violation\. JOIN message sender 1\.0 \(.*\) is not in state transfer \(SYNCED\)");