main.percona_slave_innodb_stats fails on 12.0 with UNIV_SYNC_DEBUG

[Aleksandr Kuzminsky]

main.percona_slave_innodb_stats [ fail ]
        Test ended at 2010-07-07 21:19:50

CURRENT_TEST: main.percona_slave_innodb_stats
mysqltest: At line 37: failed in 'select master_pos_wait('master-bin.000001', 1296, 300)': 2013: Lost connection to MySQL server during query

The result from queries just before the failure was:
< snip >
drop table if exists t1,t2,t3,t4,t5,t6,t7,t8,t9;
start slave;
DROP TABLE IF EXISTS t;
CREATE TABLE t(id INT,data CHAR(30)) ENGINE=InnoDB;
INSERT INTO t VALUES
(1,"aaaaabbbbbcccccdddddeeeeefffff"),
(2,"aaaaabbbbbcccccdddddeeeeefffff"),
(3,"aaaaabbbbbcccccdddddeeeeefffff"),
(4,"aaaaabbbbbcccccdddddeeeeefffff"),
(5,"aaaaabbbbbcccccdddddeeeeefffff");
INSERT INTO t SELECT t2.id,t2.data from t as t1, t as t2;
INSERT INTO t SELECT t2.id,t2.data from t as t1, t as t2;
STOP SLAVE;
START SLAVE;
INSERT INTO t SELECT t.id,t.data from t;

More results from queries before failure can be found in /mnt/data1/home/buildbot/slaves/percona-server-51-12/sm1u02/build/Percona-Server/mysql-test/var/log/percona_slave_innodb_stats.log

[Aleksandr Kuzminsky (akuzminsky)]

Server log

[Oleg Tsarev (tsarev)]

I made investigation of trouble, and now i think what trouble in out innodb-patches.

First case (series.1 file in attachments):
bzr branch lp:percona-server clean
cd clean
bzr update -r61
cat ../series.1 > series
make all
make tests
cd Percona-Server
export CFLAGS="-O2 -g -fmessage-length=0 -D_FORTIFY_SOURCE=2 -DUNIV_SYNC_DEBUG -DUNIV_DEBUG"
export CXXFLAGS="-O2 -g -fmessage-length=0 -D_FORTIFY_SOURCE=2 -DUNIV_SYNC_DEBUG -DUNIV_DEBUG"
export LIBS=-lrt
./configure --without-plugin-innobase --with-plugin-innodb_plugin
make -j3
cd mysql-test
./mysql-test-run percona_slave_innodb_stats

Result you can see in file log.1 (attachment).
Result is different from expected, because patch innodb_extend_slow.patch doesn't applied.

Second case (series.2 file in attachments):
bzr branch lp:percona-server clean2
cd clean2
bzr update -r61
cat ../series.2 > series
make all
make tests
cd Percona-Server
export CFLAGS="-O2 -g -fmessage-length=0 -D_FORTIFY_SOURCE=2 -DUNIV_SYNC_DEBUG -DUNIV_DEBUG"
export CXXFLAGS="-O2 -g -fmessage-length=0 -D_FORTIFY_SOURCE=2 -DUNIV_SYNC_DEBUG -DUNIV_DEBUG"
export LIBS=-lrt
./configure --without-plugin-innobase --with-plugin-innodb_plugin
make -j3
cd mysql-test
./mysql-test-run percona_slave_innodb_stats

Result you can see in log.2 (attachment).
Crash.

How can you look in series.2 i remove all patches expect innodb_patches.
I can't found what the patch have trouble, because when i exclude some patches from end of series, source code doesn't
build (is bad too, i expect what patches have only forward dependes - for example, i can doesn't apply some last
patches, and source code would build).

[Yasufumi Kinoshita (yasufumi-kinoshita)]

I cannot reproduce it. at least at
mysql-5.1.48 & revno: 74 of release-5.1.48-12....

[Yasufumi Kinoshita (yasufumi-kinoshita)]

Sorry, reproduced.

[Yasufumi Kinoshita (yasufumi-kinoshita)]

Anyway, I have fixed the following UNIV_SYNC_DEBUG problem.
innodb_stats.patch: restart mtr but the dict_hdr page is not re-obtain in exact
innodb_overwrite_relay_log_info.patch: one mtr try to get several x-lock of the same block.

I pushed the fix directly. So please continue to check the bug using the fix.

[Shahriyar Rzayev (rzayev-sehriyar)]

Percona now uses JIRA for bug reports so this bug report is migrated to: https://jira.percona.com/browse/PS-416