| Current Path : /home/usr.opt/mysql57/mysql-test/suite/innodb/t/ |
FreeBSD hs32.drive.ne.jp 9.1-RELEASE FreeBSD 9.1-RELEASE #1: Wed Jan 14 12:18:08 JST 2015 root@hs32.drive.ne.jp:/sys/amd64/compile/hs32 amd64 |
| Current File : //home/usr.opt/mysql57/mysql-test/suite/innodb/t/innodb_wl6326_big.test |
# This is a script for MTR with hybrid use.
# a) As regression test
# Mostly some brute force attempt to stress the internal sx locks of
# InnoDB which were introduced by WL#6326+WL#6363.
# The file with expected results fits to this variant.
# The impact on code coverage is quite good.
# b) As testbed for attempts to extend or improve the RQG test wl6326_sql.yy.
# The MTR based test uses
# - a table t1 with the same layout
# - the same stored functions
# - the same stored procedure proc_fill_t1 for inserting a configurable
# amount of records into t1
# like the RQG test wl6326_sql.yy.
# Feel free to modify parameters like $max_row_count, $max_con,
# $high_load_duration or switch debugging on (let $test_debug= 1).
# But please be aware that MTR will most probably report that the test
# failed because it got a difference to expected results.
# Reasons:
# - In general: The file with expected results fits to a) only.
# - The actual results might dependend on $max_row_count.
# - Additional result sets might be printed.
#
# WL#6326 is about the sx locks (InnoDB feature only).
--source include/have_innodb.inc
# Runtime properties:
# Notebook i5 dual core with HT, MySQL binaries compiled with debug,
# max_row_count=10000 rows
# vardir on tmpfs : ~ 375
# vardir on disk : ~ 546
--source include/big_test.inc
# We go with "--send" and "--reap" and that fails with the embedded server.
--source include/not_embedded.inc
# Its intentional to not take the risk that a run with valgrind times out.
--source include/not_valgrind.inc
# FIXME:
# Increase the code coverage provided by the current test by
# trying "InnoDB Tablespace Monitor" as soon as some bug is fixed
# or wait till the deprecated "InnoDB Tablespace Monitor" is
# removed.
# Setup of some parameters
# ------------------------
# Number of records within every chunk to be added to t1.
let $load_unit= 10000;
#
# Rough number of records in t1 to achieve.
# We add chunks of $load_unit rows till the actual number
# of rows in the table t1 exceeds $max_row_count.
# let $max_row_count= 1000000;
# let $max_row_count= 300000;
# let $max_row_count= 100000;
# let $max_row_count= 30000;
let $max_row_count= 10000; # ~ 322s on tmpfs (NB)
#
# Determine which variant to run.
let $test_debug= 0;
#
# Number of concurrent sessions to be used in the high load test.
let $max_con= 10;
# Duration of the high load test in seconds.
let $high_load_duration= 60;
# Putting all objects into the SCHEMA my_schema makes the final cleanup easier.
# We simply run than DROP SCHEMA my_schema.
CREATE SCHEMA my_schema;
USE my_schema;
CREATE FUNCTION f_thread_id (i INT) RETURNS CHAR(4) DETERMINISTIC
RETURN CONCAT(LPAD(CAST(i AS CHAR),3,'_'),'_') ;
SELECT CONCAT('->', f_thread_id( 1), '<-');
SELECT CONCAT('->', f_thread_id(12), '<-');
# Definition of parameters used in functions.
# We use here a "1" in order to make the impact on the results of the functions
# good visible.
SET @extra_int = 1;
SET @extra_string = f_thread_id(@extra_int);
SELECT @extra_int , @extra_string;
# The different functions are used later when filling t1 and also during
# RQG testing. They serve to generate the difference between column values
# in different rows in different areas of the column.
# Fictional example:
# row 1 col_int0=1 colx='1abcdefgh' coly='abcd1efgh' colz='abcdefgh1'
# row 2 col_int0=2 colx='2abcdefgh' coly='abcd2efgh' colz='abcdefgh2'
# The function f_<pattern> is for the column with the name <pattern>.
# There is a function
# - for every column except col_int0
# - even if the SQL for generating the value is simple.
# The reason for this is the architecture of the RQG test.
let $part= AS my_result
FROM (SELECT 1 AS my_col UNION SELECT 12 UNION SELECT 123
UNION SELECT 1234 UNION SELECT 12345) AS tx;
let $function_name= f_col_int1;
eval CREATE FUNCTION $function_name (i INT) RETURNS INT(20) DETERMINISTIC
RETURN i * 1000 + @extra_int ;
eval SELECT $function_name(my_col) $part;
let $function_name= f_col_int2;
eval CREATE FUNCTION $function_name (i INT) RETURNS INT(20) DETERMINISTIC
RETURN @extra_int * 10000000 + i ;
eval SELECT $function_name(my_col) $part;
let $function_name= f_col_int3;
eval CREATE FUNCTION $function_name (i INT) RETURNS INT(20) DETERMINISTIC
RETURN @extra_int ;
eval SELECT $function_name(my_col) $part;
let $function_name= f_col_blob;
eval CREATE FUNCTION $function_name (i INT) RETURNS BLOB DETERMINISTIC
RETURN RPAD(@extra_string,(@@innodb_page_size / 2 ) + 1,'a');
eval SELECT CONCAT('->', SUBSTR($function_name(my_col) FROM 1 FOR 10),
'<-.....->', SUBSTR($function_name(my_col) FROM -10 FOR 10), '<-') $part;
let $function_name= f_col_char0;
eval CREATE FUNCTION $function_name (i INT) RETURNS CHAR(255) DETERMINISTIC
RETURN LPAD(CAST(i AS CHAR),255,' ');
eval SELECT CONCAT('->', $function_name(my_col), '<-') $part;
let $function_name= f_col_char1;
eval CREATE FUNCTION $function_name (i INT) RETURNS CHAR(26) DETERMINISTIC
RETURN
CONCAT('B',
LPAD(SUBSTR(CAST(i AS CHAR),1,(LENGTH(CAST(i AS CHAR)) DIV 2)),10,' '),
@extra_string,
RPAD(SUBSTR(CAST(i AS CHAR), -((LENGTH(CAST(i AS CHAR)) + 1) DIV 2)),10,' '),
'E') ;
eval SELECT CONCAT('->', $function_name(my_col), '<-') $part;
let $function_name= f_col_char2;
eval CREATE FUNCTION $function_name (i INT) RETURNS CHAR(26) DETERMINISTIC
RETURN
CONCAT('B',
RPAD(SUBSTR(CAST(i AS CHAR),1,(LENGTH(CAST(i AS CHAR)) DIV 2)),10,' '),
@extra_string,
LPAD(SUBSTR(CAST(i AS CHAR), -((LENGTH(CAST(i AS CHAR)) + 1) DIV 2)),10,' '),
'E');
eval SELECT CONCAT('->', $function_name(my_col), '<-') $part;
let $function_name= f_col_char3;
eval CREATE FUNCTION $function_name (i INT) RETURNS CHAR(26) DETERMINISTIC
RETURN
CONCAT('B',@extra_string,LPAD(CAST(i AS CHAR),20,' '),'E');
eval SELECT CONCAT('->', $function_name(my_col), '<-') $part;
let $function_name= f_col_char4;
eval CREATE FUNCTION $function_name (i INT) RETURNS CHAR(26) DETERMINISTIC
RETURN
CONCAT('B',RPAD(CAST(i AS CHAR),20,' '),@extra_string,'E');
eval SELECT CONCAT('->', $function_name(my_col), '<-') $part;
# Auxiliary table for figuring out the impact of scenarios on
# information_schema.innodb_metrics content.
CREATE TABLE my_metrics LIKE information_schema.innodb_metrics;
ALTER TABLE my_metrics ADD COLUMN phase ENUM('after', 'before'),
DROP COLUMN SUBSYSTEM, DROP COLUMN TYPE, DROP COLUMN COMMENT,
ADD PRIMARY KEY (NAME,phase);
let $empty_my_metrics= DELETE FROM my_metrics;
let $before_my_metrics= INSERT INTO my_metrics
SELECT NAME, COUNT, MAX_COUNT, MIN_COUNT, AVG_COUNT,
COUNT_RESET, MAX_COUNT_RESET, MIN_COUNT_RESET, AVG_COUNT_RESET,
TIME_ENABLED, TIME_DISABLED, TIME_ELAPSED, TIME_RESET,
STATUS, 'before'
FROM information_schema.innodb_metrics
WHERE NAME LIKE 'innodb_rwlock_sx_%';
let $after_my_metrics= INSERT INTO my_metrics
SELECT NAME, COUNT, MAX_COUNT, MIN_COUNT, AVG_COUNT,
COUNT_RESET, MAX_COUNT_RESET, MIN_COUNT_RESET, AVG_COUNT_RESET,
TIME_ENABLED, TIME_DISABLED, TIME_ELAPSED, TIME_RESET,
STATUS, 'after'
FROM information_schema.innodb_metrics
WHERE NAME LIKE 'innodb_rwlock_sx_%';
let $print_metrics= SELECT NAME, COUNT, MAX_COUNT, MIN_COUNT, AVG_COUNT,
COUNT_RESET, MAX_COUNT_RESET, MIN_COUNT_RESET, AVG_COUNT_RESET,
TIME_ENABLED, TIME_DISABLED, TIME_ELAPSED, TIME_RESET, STATUS
FROM information_schema.innodb_metrics
WHERE NAME LIKE 'innodb_rwlock_sx_%'
ORDER BY NAME;
# The main table for testing.
CREATE TABLE t1 (
col_int0 BIGINT,
col_int1 BIGINT,
col_int2 BIGINT,
col_int3 BIGINT,
col_blob BLOB,
col_char0 VARCHAR(255),
col_char1 VARCHAR(30),
col_char2 VARCHAR(30),
col_char3 VARCHAR(30),
col_char4 VARCHAR(30)
) ENGINE = InnoDB;
# Use many indexes with mostly significant size in order to cause
# some heavy use of sx locks during data generation.
ALTER TABLE t1 ADD UNIQUE KEY uidx_col_int0 (col_int0),
ADD UNIQUE KEY uidx1 (col_int1, col_char0),
ADD UNIQUE KEY uidx2 (col_int2, col_char0, col_int1),
ADD UNIQUE KEY uidx3 (col_int3, col_int2, col_char0),
ADD UNIQUE KEY uidx4 (col_char1, col_char0),
ADD UNIQUE KEY uidx5 (col_char2, col_char0, col_char1),
ADD UNIQUE KEY uidx6 (col_char3, col_char2, col_char0),
ADD UNIQUE KEY uidx7 (col_int1, col_int2, col_int3, col_char4,
col_char1, col_char2, col_char3, col_char0),
ADD KEY idx8 (col_blob(10), col_char4);
delimiter |;
CREATE PROCEDURE proc_fill_t1 (max_row_count INT, load_unit INT)
BEGIN
DECLARE my_count INTEGER DEFAULT 0;
DECLARE max_load_count INTEGER DEFAULT 0;
DROP TABLE IF EXISTS t0;
CREATE TEMPORARY TABLE t0 (col_int0 BIGINT, PRIMARY KEY(col_int0));
WHILE (my_count < load_unit ) DO
SET my_count = my_count + 1;
INSERT INTO t0 SET col_int0 = my_count;
END WHILE;
SET max_load_count = (SELECT (max_row_count DIV load_unit) + 1 );
SELECT COUNT(col_int0) INTO @val FROM t1;
SET my_count = 0;
REPEAT
INSERT INTO t1 (col_int0, col_int1, col_int2, col_int3, col_blob,
col_char0, col_char1, col_char2,col_char3,col_char4)
SELECT col_int0 + @val,
f_col_int1(col_int0 + @val),
f_col_int2(col_int0 + @val),
f_col_int3(col_int0 + @val),
f_col_blob(col_int0 + @val),
f_col_char0(col_int0 + @val),
f_col_char1(col_int0 + @val),
f_col_char2(col_int0 + @val),
f_col_char3(col_int0 + @val),
f_col_char4(col_int0 + @val)
FROM t0;
COMMIT;
SELECT MAX(col_int0) INTO @val FROM t1;
SET my_count = my_count + 1;
UNTIL( my_count > max_load_count OR @val >= max_row_count )
END REPEAT;
DROP TEMPORARY TABLE t0;
END|
delimiter ;|
delimiter |;
CREATE PROCEDURE proc_dml (max_duration INT, t1_stripe_half INT)
BEGIN
DECLARE aux INTEGER DEFAULT 0;
DECLARE start_time INT;
DECLARE CONTINUE HANDLER FOR SQLEXCEPTION, SQLWARNING, NOT FOUND BEGIN END;
SET @extra_int = CONNECTION_ID();
SET @extra_string = f_thread_id(@extra_int);
SELECT ROUND(MAX(col_int0) / 2 ) INTO @t1_half FROM t1;
# The user lock 'Blocker' should be already set by some other session S1.
# S1 starts the race by releasing that lock.
# Wait till the lock is released and the lock can be obtained.
# In order to prevent endless waiting in case of non foreseen problems
# limit the timespan to 30 seconds.
SELECT GET_LOCK('Blocker', 30) INTO @aux;
# Release the lock immediate so that the other "runner" sessions start too.
SELECT RELEASE_LOCK('Blocker') INTO @aux;
SET start_time = UNIX_TIMESTAMP();
WHILE (UNIX_TIMESTAMP() - start_time < max_duration) DO
SET @aux = @t1_half - t1_stripe_half + ROUND(RAND() * t1_stripe_half * 2);
UPDATE t1 SET
col_int1 = f_col_int1(col_int0),
col_int2 = f_col_int2(col_int0),
col_int3 = f_col_int3(col_int0),
col_blob = f_col_blob(col_int0),
col_char0 = f_col_char0(col_int0),
col_char1 = f_col_char1(col_int0),
col_char2 = f_col_char2(col_int0),
col_char3 = f_col_char3(col_int0),
col_char4 = f_col_char4(col_int0)
WHERE col_int0 = @aux;
COMMIT;
END WHILE;
END|
delimiter ;|
SET GLOBAL innodb_monitor_disable = "innodb_rwlock_sx_%";
SET @pre_reset_ts = NOW();
--sleep 1.1
SET GLOBAL innodb_monitor_reset = "innodb_rwlock_sx_%";
SET @pre_enable_ts = NOW();
--sleep 1.1
SET GLOBAL innodb_monitor_enable = "innodb_rwlock_sx_%";
--sleep 1.1
SET @pre_collect_ts = NOW();
eval $empty_my_metrics;
eval $before_my_metrics;
--echo # TC-01 There are exact three entries "innodb_rwlock_sx_%" with the
--echo # with the name which follow in innodb_metrics.
let $check_statement=
SELECT COUNT(*) <> 3 FROM my_metrics
WHERE NAME IN ('innodb_rwlock_sx_spin_waits',
'innodb_rwlock_sx_spin_rounds',
'innodb_rwlock_sx_os_waits');
if(`$check_statement`)
{
--echo # fail
eval $check_statement;
SELECT NAME FROM my_metrics
ORDER BY NAME;
exit;
}
--echo # pass
SELECT COUNT(*) INTO @sx_count FROM my_metrics;
--echo # TC-02 Counting is now enabled. ALL = @sx_count entries show that.
let $check_statement=
SELECT COUNT(*) <> @sx_count FROM my_metrics
WHERE STATUS = 'enabled';
if(`$check_statement`)
{
--echo # fail
eval $check_statement;
SELECT NAME, STATUS FROM my_metrics
ORDER BY NAME;
exit;
}
--echo # pass
--echo # TC-03 @pre_reset_ts < TIME_RESET. ALL = @sx_count entries show that.
let $check_statement=
SELECT COUNT(*) <> @sx_count FROM my_metrics
WHERE @pre_reset_ts < TIME_RESET;
if(`$check_statement`)
{
--echo # fail
eval $check_statement;
SELECT NAME, @pre_reset_ts, TIME_RESET FROM my_metrics
ORDER BY NAME;
exit;
}
--echo # pass
--echo # TC-04 @pre_enable_ts < TIME_ENABLED. ALL = @sx_count entries show that.
let $check_statement=
SELECT COUNT(*) <> @sx_count FROM my_metrics
WHERE @pre_enable_ts < TIME_ENABLED;
if(`$check_statement`)
{
--echo # fail
eval $check_statement;
SELECT NAME, @pre_enable_ts, TIME_ENABLED FROM my_metrics
ORDER BY NAME;
exit;
}
--echo # pass
--echo # TC-05 TIME_RESET < TIME_ENABLED AND TIME_ENABLED < @pre_collect_ts
--echo # AND TIME_ELAPSED > 0. ALL = @sx_count entries show that.
let $check_statement=
SELECT COUNT(*) <> @sx_count FROM my_metrics
WHERE TIME_RESET < TIME_ENABLED AND TIME_ENABLED < @pre_collect_ts
AND TIME_ELAPSED > 0;
if(`$check_statement`)
{
--echo # fail
eval $check_statement;
SELECT NAME, @pre_collect_ts, TIME_RESET, TIME_ENABLED, TIME_ELAPSED
FROM my_metrics
ORDER BY NAME;
exit;
}
--echo # pass
--echo # TC-06 COUNT_RESET = MAX_COUNT_RESET. ALL = @sx_count entries show that.
let $check_statement=
SELECT COUNT(*) <> @sx_count FROM my_metrics
WHERE COUNT_RESET = MAX_COUNT_RESET;
if(`$check_statement`)
{
--echo # fail
eval $check_statement;
SELECT NAME, COUNT_RESET, MAX_COUNT_RESET FROM my_metrics
ORDER BY NAME;
exit;
}
--echo # pass
SET GLOBAL innodb_monitor_reset = "innodb_rwlock_sx_%";
--disable_result_log
if($test_debug)
{
--enable_result_log
eval $print_metrics;
}
SHOW ENGINE INNODB STATUS;
--enable_result_log
eval $empty_my_metrics;
eval $before_my_metrics;
# These values (the "0") help to identify later if some record is in its
# initial state or already modified.
SET @extra_string = '__0_';
SET @extra_int = 0;
--echo # TC-07 One session inserts some significant amount of rows into t1.
--echo # The system MUST survive that.
--replace_result $max_row_count <max_row_count>
eval SET @max_row_count = $max_row_count;
--replace_result $load_unit <load_unit>
eval SET @load_unit = $load_unit;
SET @start_time = UNIX_TIMESTAMP();
SET AUTOCOMMIT = OFF;
CALL proc_fill_t1 (@max_row_count, @load_unit);
--echo # pass
SET AUTOCOMMIT = ON;
SELECT col_int0 INTO @t1_half FROM t1
WHERE col_int0 >= (@val DIV 2) ORDER BY col_int0 LIMIT 1;
--disable_result_log
if($test_debug)
{
--enable_result_log
SELECT COUNT(*) AS table_row_count,
UNIX_TIMESTAMP() - @start_time AS fill_run_time
FROM t1;
eval $print_metrics;
}
SHOW ENGINE INNODB STATUS;
--enable_result_log
# Show that the value distribution is according to the plan.
--vertical_results
SELECT col_int0, col_int1, col_int2, col_int3,
CONCAT('->', SUBSTR(col_blob FROM 1 FOR 10),
'<-.....->', SUBSTR(col_blob FROM -10 FOR 10), '<-') AS col_blobx,
CONCAT('->',col_char0,'<-') AS col_char0x,
CONCAT('->',col_char1,'<-') AS col_char1x,
CONCAT('->',col_char2,'<-') AS col_char2x,
CONCAT('->',col_char3,'<-') AS col_char3x,
CONCAT('->',col_char4,'<-') AS col_char4x
FROM t1 WHERE col_int0 between 98 AND 102;
--horizontal_results
# For experiments/interest only. Please do not remove that.
if (0)
{
ANALYZE TABLE t1;
SELECT n_rows, clustered_index_size, sum_of_other_index_sizes
FROM mysql.innodb_table_stats;
# SELECT * FROM mysql.innodb_index_stats;
# idx_col_int3_int0 n_diff_pfx01 1 col_int3
# idx_col_int3_int0 n_diff_pfx02 10000 col_int3,col_int0
# idx_col_int3_int0 n_diff_pfx03 10000 col_int3,col_int0,DB_ROW_ID
# idx_col_int3_int0 n_leaf_pages 19 Number of leaf pages in the index
# idx_col_int3_int0 size 20 Number of pages in the index
--vertical_results
SELECT t1.index_name, t1.stat_value AS idx_pages, t2.stat_value AS idx_leaf_pages,
(t1.stat_value - t2.stat_value - 1) / t1.stat_value AS sx_page_ratio
FROM mysql.innodb_index_stats t1, mysql.innodb_index_stats t2
WHERE t1.index_name = t2.index_name
AND t1.stat_name = 'size' AND t2.stat_name = 'n_leaf_pages'
ORDER BY t1.index_name;
--horizontal_results
}
--echo # TC-11 Several concurrent sessions perform updates in t1 like mad.
--echo # The system MUST survive this.
--echo # Printing of statements is partially suppressed.
SET @start_time = UNIX_TIMESTAMP();
SELECT 1 FROM t1 WHERE col_int0 = @t1_half FOR UPDATE;
SELECT GET_LOCK('Blocker', 1000) ;
--disable_query_log
let $num= $max_con;
while ($num)
{
--connect (con$num,localhost,root,,)
USE my_schema;
# The second parameter of the procedure is size of the affected stripe / 2.
# A smaller stripe causes some smaller counter growth but most probably
# also more stress around locking in general.
# Example # (nnnn) = half stripe size
# NAME | COUNT_RESET (5000) | COUNT_RESET (100)
# -----------------------------+--------------------+----------------
# innodb_rwlock_sx_os_waits | 1412 | 486
# innodb_rwlock_sx_spin_rounds | 44061 | 17031
# innodb_rwlock_sx_spin_waits | 996 | 515
--send
eval CALL proc_dml($high_load_duration,@t1_half);
dec $num;
}
--connection default
SELECT RELEASE_LOCK('Blocker') ;
--sleep 3
COMMIT;
let $num= $max_con;
while ($num)
{
--connection con$num
--reap
dec $num;
}
--echo # pass
--connection default
--enable_query_log
# let $wait_timeout= 181;
# --source include/wait_condition.inc
# eval $after_my_metrics;
--disable_result_log
if($test_debug)
{
--enable_result_log
SELECT UNIX_TIMESTAMP() - @start_time AS update_battle_run_time;
eval $print_metrics;
}
SHOW ENGINE INNODB STATUS;
--enable_result_log
--echo # TC-13 One session performs ALTER TABLE t1 ADD KEY ... on the fat table t1.
--echo # The system MUST survive this.
SET @start_time = UNIX_TIMESTAMP();
ALTER TABLE t1 ADD KEY idx_col_char4_col_char0 (col_char4,col_char0);
--disable_result_log
if($test_debug)
{
--enable_result_log
SELECT UNIX_TIMESTAMP() - @start_time AS add_key_run_time;
eval $print_metrics;
}
SHOW ENGINE INNODB STATUS;
--enable_result_log
--echo # pass
--echo # TC-15 One session performs a fat update on the fat table t1.
--echo # The system MUST survive this.
SET @start_time = UNIX_TIMESTAMP();
SET @extra_int = 13;
SET @extra_string = f_thread_id(@extra_int);
eval UPDATE t1 SET
col_int1 = f_col_int1(col_int0), col_int2 = f_col_int2(col_int0),
col_int3 = f_col_int3(col_int0), col_blob = f_col_blob(col_int0),
col_char0 = f_col_char0(col_int0), col_char1 = f_col_char1(col_int0),
col_char2 = f_col_char2(col_int0), col_char3 = f_col_char3(col_int0),
col_char4 = f_col_char4(col_int0)
WHERE col_int0 BETWEEN @t1_half - 2500 AND @t1_half + 2500;
COMMIT;
--disable_result_log
if($test_debug)
{
--enable_result_log
SELECT UNIX_TIMESTAMP() - @start_time AS total_update_run_time;
eval $print_metrics;
}
SHOW ENGINE INNODB STATUS;
--enable_result_log
--echo # pass
# Basically every of the big activities causes some counter growth.
# But caused by
# - the architecture of InnoDB (certain things happen asynchronous)
# - the actual test configuration (server/InnoDB options)
# - conditions like parallel (./mtr --parallel=auto?) load on the testing box
# this might be not fulfilled per single big activity every time except
# we go with huge waits or similar.
# Observation:
# - non debug binaries: expectation frequent not fulfilled
# - debug binaries: expectation rare not fulfilled
#
let $wait_timeout= 121;
let $wait_condition=
SELECT COUNT(*) = @sx_count
FROM information_schema.innodb_metrics t_after
JOIN my_metrics t_before
ON t_after.COUNT_RESET > t_before.COUNT_RESET AND t_after.NAME = t_before.NAME;
--source include/wait_condition.inc
eval $after_my_metrics;
--echo # TC-16 The following activities happend after reset in innodb_metrics
--echo # - Insert some significant amount of rows into t1.
--echo # - Several concurrent users perform excessive updates in t1.
--echo # - ALTER TABLE ... ADD KEY <sufficient big enough structure>
--echo # - One UPDATE statement modifying a huge slice of t1.
--echo # Any of them causes heavy use of SX lock and therefore COUNT_RESET
--echo # must have grown for ALL = @sx_count entries.
# The former testcases TC-10 and TC12 had to be made a part of this testcase
# because their results were unstable.
let $check_statement=
SELECT COUNT(*) <> @sx_count FROM my_metrics t_after JOIN my_metrics t_before
ON t_after.COUNT_RESET > t_before.COUNT_RESET AND t_after.NAME = t_before.NAME
WHERE t_after.phase = 'after' AND t_before.phase = 'before';
if(`$check_statement`)
{
--echo # fail
eval $check_statement;
SELECT * FROM my_metrics
ORDER BY NAME, phase;
exit;
}
--echo # pass
--echo # TC-09 Heavy activity after reset.
--echo # COUNT_RESET = MAX_COUNT_RESET for ALL = @sx_count entries
--echo # needs to stay valid though he counters will have grown.
let $check_statement=
SELECT COUNT(*) <> @sx_count FROM my_metrics
WHERE phase = 'after' AND COUNT_RESET = MAX_COUNT_RESET;
if(`$check_statement`)
{
--echo # fail
eval $check_statement;
SELECT * FROM my_metrics
ORDER BY NAME, phase;
exit;
}
--echo # pass
eval $empty_my_metrics;
eval $before_my_metrics;
SET GLOBAL innodb_monitor_reset = "innodb_rwlock_sx_%";
eval $after_my_metrics;
--echo # TC-08 There was a reset. COUNT_RESET = MAX_COUNT_RESET for ALL
--echo # = @sx_count entries.
let $check_statement=
SELECT COUNT(*) <> @sx_count FROM my_metrics
WHERE phase = 'before' AND COUNT_RESET = MAX_COUNT_RESET;
if(`$check_statement`)
{
--echo # fail
eval $check_statement;
SELECT * FROM my_metrics
ORDER BY NAME, phase;
exit;
}
--echo # pass
--echo # TC-17 We had heavy activity causing big counters and after that a reset.
--echo # Reset causes COUNT > COUNT_RESET AND MAX_COUNT > MAX_COUNT_RESET
--echo # for ALL @sx_count entries.
let $check_statement=
SELECT COUNT(*) <> @sx_count FROM my_metrics
WHERE phase = 'after'
AND COUNT > COUNT_RESET
AND MAX_COUNT > MAX_COUNT_RESET;
if(`$check_statement`)
{
--echo # fail
eval $check_statement;
SELECT * FROM my_metrics
ORDER BY NAME, phase;
exit;
}
--echo # pass
--echo # TC-18 We had some reset but this must not decrease COUNT or MAX_COUNT
--echo # after.COUNT >= before.COUNT AND
--echo # after.MAX_COUNT >= before.MAX_COUNT for ALL @sx_count entries.
let $check_statement=
SELECT COUNT(*) <> @sx_count FROM my_metrics t_after JOIN my_metrics t_before
ON t_after.COUNT >= t_before.COUNT AND t_after.MAX_COUNT >= t_before.MAX_COUNT
AND t_after.NAME = t_before.NAME
WHERE t_after.phase = 'after' AND t_before.phase = 'before';
if(`$check_statement`)
{
--echo # fail
eval $check_statement;
SELECT * FROM my_metrics
ORDER BY NAME, phase;
exit;
}
--echo # pass
--echo # TC-19 We had some reset after heavy activity and this must cause
--echo # after.COUNT_RESET < before.COUNT_RESET
--echo # AND after.MAX_COUNT_RESET < before.MAX_COUNT_RESET AND
--echo # for ALL @sx_count entries.
let $check_statement=
SELECT COUNT(*) <> @sx_count FROM my_metrics t_after JOIN my_metrics t_before
ON t_after.COUNT_RESET < t_before.COUNT_RESET
AND t_after.MAX_COUNT_RESET < t_before.MAX_COUNT_RESET
AND t_after.NAME = t_before.NAME
WHERE t_after.phase = 'after' AND t_before.phase = 'before';
if(`$check_statement`)
{
--echo # fail
eval $check_statement;
SELECT * FROM my_metrics
ORDER BY NAME, phase;
exit;
}
--echo # pass
# Cleanup
let $num= $max_con;
while ($num)
{
--connection con$num
--disconnect con$num
--source include/wait_until_disconnected.inc
dec $num;
}
--connection default
USE test;
DROP SCHEMA my_schema;
SET GLOBAL innodb_monitor_disable = all;
SET GLOBAL innodb_monitor_reset_all = all;
--disable_warnings
SET GLOBAL innodb_monitor_enable = default;
SET GLOBAL innodb_monitor_disable = default;
SET GLOBAL innodb_monitor_reset = default;
SET GLOBAL innodb_monitor_reset_all = default;
--enable_warnings
SET GLOBAL innodb_monitor_disable = "innodb_rwlock_sx_%";
SET GLOBAL innodb_monitor_reset = "innodb_rwlock_sx_%";