— /home/elenst/maria-5.5-galera-clean/mysql-test/r/mysqld--help.result 2013-01-29 15:13:53.807815000 +0400
+++ /home/elenst/maria-5.5-galera-clean/mysql-test/r/mysqld--help.reject 2013-01-29 16:33:01.378513264 +0400
@@ -1186,7 +1186,7 @@
wsrep-mysql-replication-bundle 0
wsrep-node-address
wsrep-node-incoming-address AUTO
-wsrep-node-name meddwl
+wsrep-node-name ubuntu11
wsrep-notify-cmd
wsrep-on FALSE
wsrep-provider none

Added wsrep-node-name to the list of skipped variables (@skipvars) in the test file, and removed from the result file.

— /home/elenst/maria-5.5-galera-clean/mysql-test/suite/percona/innodb_sys_index.result 2013-01-29 15:13:53.807815000 +0400
+++ /home/elenst/maria-5.5-galera-clean/mysql-test/suite/percona/innodb_sys_index.reject 2013-01-29 16:45:53.764803052 +0400
@@ -3,7 +3,7 @@
Note 1051 Unknown table 't1'
select @@version_comment limit 1 ;
@@version_comment
-Source distribution
+Source distribution, wsrep_23.7rc1.r3365
SELECT COUNT FROM `information_schema`.`INNODB_SYS_INDEXES` ;
CREATE TABLE test.t1 ( `a` SERIAL NOT NULL , `b` VARCHAR( 255 ) NOT NULL , INDEX ( `b` ) ) ENGINE = InnoDB ;
SHOW TABLE STATUS FROM `information_schema` LIKE 'INNODB_SYS_INDEXES%' ;

Added --replace_regex /wsrep_[0-9]\.[a-zA-Z0-9]\.r[0-9]+/wsrep_X.Y.rZ/ to the test file, and ", wsrep_X.Y.rZ" to the result file.

It turned out to be a primitive job, almost all of the original failures have already been fixed, there are only a couple of tests left. The patch is attached, I'll ask Seppo to confirm he doesn't mind me pushing it into maria-5.5-galera.

My concern about memory issues which wouldn't allow to run MTR tests was also (partially) ungrounded: there would have been memory issues if the tests were run with a full node setup, but since we run MTR tests in buildbot without setting wsrep_provider, the memory problems stay hidden and don't affect the outcome.

I haven't got a confirmation from Seppo, but the change is so trivial, I just pushed it into maria-5.5-galera, he can revert it later if needed.
Meanwhile we'll see if we finally get some green buildbot results.

Got green results for standard tests on bintars, e.g. http://buildbot.askmonty.org/buildbot/builders/kvm-bintar-hardy-amd64/builds/2170
But the percona test still failed, it turned out that sometimes wsrep revno is shown as a number, and sometimes as XXXX, the regex only expected a number. Modified to handle both, pushed into maria-5.5-galera.

Standard tests on bintars pass. fulltest still fails, but it's unrelated to wrong results, caused by memory loss warnings. RPM- and deb tests have issues related to packaging.