Sergei Golubchik added a comment - 2017-01-05 13:29 - edited Looks ok. On xenial it doesn't fail, it just takes a lot of time: parts.partition_float_myisam w6 [ pass ] 2212982 It's more than half an hour. On a slower box (with an old valgrind, etc) it could time out, I suppose. There's no bug here, but perhaps this test should be considered "big" or skipped for valgrind runs completely?

Elena Stepanova added a comment - 2017-01-05 14:01 It doesn't take long on a regular builder, so maybe not "big", but we can make it "no_valgrind_without_big" (include/no_valgrind_without_big.inc)

Elena Stepanova added a comment - 2017-02-18 01:31 - edited UPD: see later comments and MDEV-12084 about this. Here is a timeout without valgrind: http://buildbot.askmonty.org/buildbot/builders/kvm-bintar-centos5-amd64/builds/4623/steps/test/logs/stdio parts.partition_float_myisam w2 [ fail ] timeout after 900 seconds Test ended at 2017-02-08 22:14:47   Test case timeout after 900 seconds   == /usr/local/mariadb-10.2.4-linux-x86_64/mysql-test/var/2/log/partition_float_myisam.log == -2.2250738585072014e-208 0 1.5 1234.567 2.2250738585072016e208 select * from t2 where a=1234.567; a 1234.567 delete from t2 where a=1234.567; select * from t2; a -2.2250738585072016e208 -1.5 -1 -2.2250738585072014e-208 0 1.5 2.2250738585072016e208 delete from t2; 16384*3 inserts; And retry took only 20 seconds: parts.partition_float_myisam w2 [ retry-pass ] 20540 So, there might be more to this than just a slow machine. On the other hand, the timeout on parts.partition_float_myisam happened 5 times over the last year, twice on work-valgrind and 3 times on centos5-amd64. For parts.partition_int_myisam (it's a longer test) – 11 times over the last year, 8 times on work-valgrind and 3 times on centos5-amd64. So, it is somehow specific to the machines. So, maybe it's something specific to the machines, after all.

Elena Stepanova added a comment - 2017-02-18 17:41 - edited Assuming it's just a slow machine, I think a reasonable solution for these tests is splitting them into logically separate parts. partition_float_myisam is in fact two tests: --source suite/parts/inc/partition_float.inc --source suite/parts/inc/partition_double.inc partition_int_myisam is 5 tests: --source suite/parts/inc/partition_tinyint.inc --source suite/parts/inc/partition_smallint.inc --source suite/parts/inc/partition_int.inc --source suite/parts/inc/partition_mediumint.inc --source suite/parts/inc/partition_bigint.inc There is literally no gain at combining these parts under one mega-test – there is no preparation work, no even test-specific server parameters which would cause a server restart. The test simply sets several variables and then runs the sequence of include files. From every point of view, it's more correct to split it. I'm going to do just that.

Elena Stepanova added a comment - 2017-02-18 18:38 Comparative timing for combined vs split tests (shows no loss either at server restarts or test execution time) worker[1] Using MTR_BUILD_THREAD 300, with reserved ports 16000..16019 parts.partition_float_innodb 'innodb_plugin' [ pass ] 1453 parts.partition_int_innodb 'innodb_plugin' [ pass ] 1232 worker[1] > Restart [mysqld.1 - pid: 445, winpid: 445] - running with different options '--innodb --innodb-cmpmem --innodb-trx --innodb-buffer-pool-stats --innodb-buffer-page --innodb-buffer-page-lru --enable-partition' != '--ignore-builtin-innodb --plugin-load=ha_innodb.so --innodb --innodb-cmpmem --innodb-trx --innodb-buffer-pool-stats --innodb-buffer-page --innodb-buffer-page-lru --enable-partition' parts.partition_float_innodb 'xtradb' [ pass ] 1497 parts.partition_int_innodb 'xtradb' [ pass ] 1183 worker[1] > Restart [mysqld.1 - pid: 499, winpid: 499] - running with different options '--enable-partition' != '--innodb --innodb-cmpmem --innodb-trx --innodb-buffer-pool-stats --innodb-buffer-page --innodb-buffer-page-lru --enable-partition' parts.partition_float_myisam [ pass ] 28321 parts.partition_int_myisam [ pass ] 69986 -------------------------------------------------------------------------- The servers were restarted 2 times Spent 103.672 of 117 seconds executing testcases worker[1] Using MTR_BUILD_THREAD 300, with reserved ports 16000..16019 parts.partition_bigint_innodb 'innodb_plugin' [ pass ] 284 parts.partition_double_innodb 'innodb_plugin' [ pass ] 738 parts.partition_float_innodb 'innodb_plugin' [ pass ] 752 parts.partition_int_innodb 'innodb_plugin' [ pass ] 287 parts.partition_mediumint_innodb 'innodb_plugin' [ pass ] 268 parts.partition_smallint_innodb 'innodb_plugin' [ pass ] 416 parts.partition_tinyint_innodb 'innodb_plugin' [ pass ] 239 worker[1] > Restart [mysqld.1 - pid: 1339, winpid: 1339] - running with different options '--innodb --innodb-cmpmem --innodb-trx --innodb-buffer-pool-stats --innodb-buffer-page --innodb-buffer-page-lru --enable-partition' != '--ignore-builtin-innodb --plugin-load=ha_innodb.so --innodb --innodb-cmpmem --innodb-trx --innodb-buffer-pool-stats --innodb-buffer-page --innodb-buffer-page-lru --enable-partition' parts.partition_bigint_innodb 'xtradb' [ pass ] 408 parts.partition_double_innodb 'xtradb' [ pass ] 649 parts.partition_float_innodb 'xtradb' [ pass ] 652 parts.partition_int_innodb 'xtradb' [ pass ] 359 parts.partition_mediumint_innodb 'xtradb' [ pass ] 448 parts.partition_smallint_innodb 'xtradb' [ pass ] 343 parts.partition_tinyint_innodb 'xtradb' [ pass ] 186 worker[1] > Restart [mysqld.1 - pid: 1472, winpid: 1472] - running with different options '--enable-partition' != '--innodb --innodb-cmpmem --innodb-trx --innodb-buffer-pool-stats --innodb-buffer-page --innodb-buffer-page-lru --enable-partition' parts.partition_bigint_myisam [ pass ] 15853 parts.partition_double_myisam [ pass ] 13525 parts.partition_float_myisam [ pass ] 13687 parts.partition_int_myisam [ pass ] 16200 parts.partition_mediumint_myisam [ pass ] 16396 parts.partition_smallint_myisam [ pass ] 16064 parts.partition_tinyint_myisam [ pass ] 143 -------------------------------------------------------------------------- The servers were restarted 2 times Spent 97.897 of 117 seconds executing testcases InnoDB tests are much smaller than MyISAM (they insert less data), but there has been one occasion of a timeout on {[partition_float_innodb}} on a P8 builder, so I'll split them as well.

Elena Stepanova added a comment - 2017-02-18 18:44 https://github.com/MariaDB/server/commit/6364adb199f8adbc5adfe0c276bdf2d3dd17454c

Elena Stepanova added a comment - 2017-02-19 23:20 - edited Apparently the problems on valgrind builders and on CentOS 5 are completely different. Valgrind is just a slow builder, but CentOS has some issue which affects communication between the server and the client. In scope of this JIRA entry the slow builder is being solved, splitting the test into logical parts should help with that. The CentOS problem is filed as MDEV-12084 and should be handled separately. https://github.com/MariaDB/server/commit/6364adb199f8adbc5adfe0c276bdf2d3dd17454c