Skip to content
Permalink
Browse files
MDEV-8692 prefschema test failures on ARM (on Debian build system)
A few tests assumes that the CYCLE timer is always available,
which is not true on some platforms (e.g. ARM).
Fixing the tests not to reply on the CYCLE availability.
  • Loading branch information
abarkov authored and vuvova committed Jul 17, 2017
1 parent 6efee22 commit d023156
Show file tree
Hide file tree
Showing 6 changed files with 47 additions and 6 deletions.
@@ -0,0 +1,4 @@
if (!`SELECT count(*) FROM performance_schema.performance_timers WHERE timer_name='CYCLE' AND timer_frequency IS NOT NULL`)
{
Skip Need performance timer CYCLE;
}
@@ -554,7 +554,7 @@ ERROR 42000: DROP command denied to user 'pfs_user_4'@'localhost' for table 'eve
#
# Grant access to change tables with the root account
GRANT UPDATE ON performance_schema.setup_consumers TO pfs_user_4;
GRANT UPDATE ON performance_schema.setup_timers TO pfs_user_4;
GRANT UPDATE, SELECT ON performance_schema.setup_timers TO pfs_user_4;
GRANT UPDATE, SELECT ON performance_schema.setup_instruments TO pfs_user_4;
GRANT DROP ON performance_schema.events_waits_current TO pfs_user_4;
GRANT DROP ON performance_schema.events_waits_history TO pfs_user_4;
@@ -565,7 +565,7 @@ UPDATE performance_schema.setup_instruments SET enabled = 'YES'
WHERE name LIKE 'wait/synch/mutex/%'
OR name LIKE 'wait/synch/rwlock/%';
UPDATE performance_schema.setup_consumers SET enabled = 'YES';
UPDATE performance_schema.setup_timers SET timer_name = 'TICK';
UPDATE performance_schema.setup_timers SET timer_name = 'TICK' WHERE name <> "wait";
TRUNCATE TABLE performance_schema.events_waits_history_long;
TRUNCATE TABLE performance_schema.events_waits_history;
TRUNCATE TABLE performance_schema.events_waits_current;
@@ -575,4 +575,6 @@ DROP USER pfs_user_4;
flush privileges;
UPDATE performance_schema.setup_instruments SET enabled = 'YES', timed = 'YES';
UPDATE performance_schema.setup_consumers SET enabled = 'YES';
UPDATE performance_schema.setup_timers SET timer_name = 'CYCLE';
UPDATE performance_schema.setup_timers SET timer_name = 'MICROSECOND' where name="idle";
UPDATE performance_schema.setup_timers SET timer_name = 'NANOSECOND' where name="stage";
UPDATE performance_schema.setup_timers SET timer_name = 'NANOSECOND' where name="statement";
@@ -2,6 +2,7 @@

--source include/not_embedded.inc
--source include/have_perfschema.inc
--source include/have_timer_cycle.inc

--replace_column 2 <frequency> 3 <resolution> 4 <overhead>
select * from performance_schema.performance_timers;
@@ -2,6 +2,7 @@

--source include/not_embedded.inc
--source include/have_perfschema.inc
--source include/have_timer_cycle.inc

select * from performance_schema.setup_timers;

@@ -291,7 +291,7 @@ TRUNCATE TABLE performance_schema.events_waits_current;
--echo # Grant access to change tables with the root account

GRANT UPDATE ON performance_schema.setup_consumers TO pfs_user_4;
GRANT UPDATE ON performance_schema.setup_timers TO pfs_user_4;
GRANT UPDATE, SELECT ON performance_schema.setup_timers TO pfs_user_4;
GRANT UPDATE, SELECT ON performance_schema.setup_instruments TO pfs_user_4;
GRANT DROP ON performance_schema.events_waits_current TO pfs_user_4;
GRANT DROP ON performance_schema.events_waits_history TO pfs_user_4;
@@ -308,7 +308,11 @@ WHERE name LIKE 'wait/synch/mutex/%'

UPDATE performance_schema.setup_consumers SET enabled = 'YES';

UPDATE performance_schema.setup_timers SET timer_name = 'TICK';
# We do not touch "wait", to avoid restoring it at the end of the test,
# as its default value initialized at server startup is ambiguous:
# it can be CYCLE or NANOSECOND depending on platform

UPDATE performance_schema.setup_timers SET timer_name = 'TICK' WHERE name <> "wait";

TRUNCATE TABLE performance_schema.events_waits_history_long;
TRUNCATE TABLE performance_schema.events_waits_history;
@@ -323,5 +327,9 @@ DROP USER pfs_user_4;
flush privileges;
UPDATE performance_schema.setup_instruments SET enabled = 'YES', timed = 'YES';
UPDATE performance_schema.setup_consumers SET enabled = 'YES';
UPDATE performance_schema.setup_timers SET timer_name = 'CYCLE';

# Restore the default values for the timers that we changed.
# Note, we did not touch "wait", see above.
UPDATE performance_schema.setup_timers SET timer_name = 'MICROSECOND' where name="idle";
UPDATE performance_schema.setup_timers SET timer_name = 'NANOSECOND' where name="stage";
UPDATE performance_schema.setup_timers SET timer_name = 'NANOSECOND' where name="statement";
@@ -129,6 +129,31 @@ ulonglong my_timer_cycles_il_x86_64();
clock_gettime(CLOCK_SGI_CYCLE) for Irix platforms,
or on read_real_time for aix platforms. There is
nothing for Alpha platforms, they would be tricky.
On the platforms that do not have a CYCLE timer,
"wait" events are initialized to use NANOSECOND instead of CYCLE
during performance_schema initialization (at the server startup).
Linux performance monitor (see "man perf_event_open") can
provide cycle counter on the platforms that do not have
other kinds of cycle counters. But we don't use it so far.
ARM notes
---------
During tests on ARMv7 Debian, perf_even_open() based cycle counter provided
too low frequency with too high overhead:
MariaDB [performance_schema]> SELECT * FROM performance_timers;
+-------------+-----------------+------------------+----------------+
| TIMER_NAME | TIMER_FREQUENCY | TIMER_RESOLUTION | TIMER_OVERHEAD |
+-------------+-----------------+------------------+----------------+
| CYCLE | 689368159 | 1 | 970 |
| NANOSECOND | 1000000000 | 1 | 308 |
| MICROSECOND | 1000000 | 1 | 417 |
| MILLISECOND | 1000 | 1000 | 407 |
| TICK | 127 | 1 | 612 |
+-------------+-----------------+------------------+----------------+
Therefore, it was decided not to use perf_even_open() on ARM
(i.e. go without CYCLE and have "wait" events use NANOSECOND by default).
*/

ulonglong my_timer_cycles(void)

0 comments on commit d023156

Please sign in to comment.