Test failures with MariaDB 10.3.9

[loqs]

"/usr/bin/perl" -MExtUtils::Command::MM -e 'cp_nonempty' -- mysql.bs blib/arch/auto/DBD/mysql/mysql.bs 644
PERL_DL_NONLAZY=1 "/usr/bin/perl" "-MExtUtils::Command::MM" "-MTest::Harness" "-e" "undef *Test::Harness::Switches; test_harness(0, 'blib/lib', 'blib/arch')" t/*.t
# Driver version is 4.048
t/00base.t .............................. ok
# DBI connect('','builduser',...) failed: Can't connect to local MySQL server through socket '/run/mysqld/mysqld.sock' (2) at t/05dbcreate.t line 16.
t/05dbcreate.t .......................... skipped: no database connection
# mysql_clientinfo is: 10.3.9
# mysql_clientversion is: 100309
# mysql_serverversion is: 100309
# mysql_hostinfo is: Localhost via UNIX socket
# mysql_serverinfo is: 10.3.9-MariaDB-log
# mysql_stat is: Uptime: 10  Threads: 8  Questions: 2  Slow queries: 0  Opens: 17  Flush tables: 1  Open tables: 11  Queries per second avg: 0.200
# mysql_protoinfo is: 10
# SQL_DBMS_VER is 10.3.9-MariaDB-log
# Default storage engine is: InnoDB
2018-10-02 16:08:18 10 [Warning] Access denied for user '4yZ73s9qeECdWi'@'localhost' (using password: YES)
t/10connect.t ........................... ok
t/15reconnect.t ......................... 
Failed 2/16 subtests 
t/16dbi-get_info.t ...................... ok
t/20createdrop.t ........................ ok
t/25lockunlock.t ........................ ok
t/29warnings.t .......................... ok
t/30insertfetch.t ....................... ok
t/31insertid.t .......................... ok
t/32insert_error.t ...................... ok
t/35limit.t ............................. ok
t/35prepare.t ........................... ok
t/40bindparam.t ......................... ok
t/40bindparam2.t ........................ ok
t/40bit.t ............................... ok
t/40blobs.t ............................. ok
t/40catalog.t ........................... ok
t/40keyinfo.t ........................... ok
t/40listfields.t ........................ ok
t/40nulls.t ............................. ok
t/40nulls_prepare.t ..................... ok
t/40numrows.t ........................... ok
t/40server_prepare.t .................... ok
t/40server_prepare_crash.t .............. ok
t/40server_prepare_error.t .............. ok
t/40types.t ............................. ok
t/41bindparam.t ......................... ok
t/41blobs_prepare.t ..................... ok
t/41int_min_max.t ....................... ok
t/42bindparam.t ......................... ok
t/43count_params.t ...................... ok
t/50chopblanks.t ........................ ok
t/50commit.t ............................ ok
t/51bind_type_guessing.t ................ ok
t/52comment.t ........................... ok
t/53comment.t ........................... ok
t/55utf8.t .............................. ok
t/55utf8mb4.t ........................... ok
t/56connattr.t .......................... skipped: performance schema not enabled
t/60leaks.t ............................. skipped: Skip $ENV{EXTENDED_TESTING} is not set
t/65segfault.t .......................... ok
t/65types.t ............................. ok
2018-10-02 12:08:22 63 [Warning] Aborted connection 63 to db: 'test' user: 'builduser' host: 'localhost' (Got an error reading communication packets)
t/70takeimp.t ........................... ok
t/71impdata.t ........................... ok
t/75supported_sql.t ..................... ok
t/76multi_statement.t ................... ok
t/80procs.t ............................. skipped: Your test user does not have ALTER_ROUTINE privileges.
t/81procs.t ............................. skipped: Your test user does not have ALTER_ROUTINE privileges.
t/85init_command.t ...................... ok
t/86_bug_36972.t ........................ ok
t/87async.t ............................. ok
t/88async-multi-stmts.t ................. ok
t/89async-method-check.t ................ ok
t/91errcheck.t .......................... ok
t/92ssl_backronym_vulnerability.t ....... ok
t/92ssl_optional.t ...................... ok
t/92ssl_riddle_vulnerability.t .......... ok
t/99_bug_server_prepare_blob_null.t ..... ok
t/manifest.t ............................ skipped: these tests are for release testing
t/pod.t ................................. skipped: Test::Pod 1.00 required for testing POD
t/rt110983-valid-mysqlfd.t .............. ok

#   Failed test at t/rt118977-zerofill.t line 22.
#          got: '1'
#     expected: '00001'
# Looks like you failed 1 test of 8.
t/rt118977-zerofill.t ................... 
Dubious, test returned 1 (wstat 256, 0x100)
Failed 1/8 subtests 
t/rt25389-bin-case.t .................... ok
t/rt50304-column_info_parentheses.t ..... ok
t/rt61849-bind-param-buffer-overflow.t .. ok
t/rt75353-innodb-lock-timeout.t ......... ok
t/rt83494-quotes-comments.t ............. ok
2018-10-02 12:08:46 94 [Warning] Aborted connection 94 to db: 'test' user: 'builduser' host: 'localhost' (Got timeout reading communication packets)
t/rt85919-fetch-lost-connection.t ....... ok
t/rt86153-reconnect-fail-memory.t ....... skipped: $ENV{EXTENDED_TESTING} is not set
t/rt88006-bit-prepare.t ................. ok
t/rt91715.t ............................. ok
t/version.t ............................. ok

Test Summary Report
-------------------
t/15reconnect.t                       (Wstat: 139 Tests: 14 Failed: 0)
  Non-zero wait status: 139
  Parse errors: Bad plan.  You planned 16 tests but ran 14.
t/rt118977-zerofill.t                 (Wstat: 256 Tests: 8 Failed: 1)
  Failed test:  8
  Non-zero exit status: 1
Files=73, Tests=2223, 31 wallclock secs ( 0.31 usr  0.06 sys +  4.25 cusr  0.45 csys =  5.07 CPU)
Result: FAIL
Failed 2/73 test programs. 1/2223 subtests failed.

Are these failures expected?

[pali]

This is one of bugs which cannot be (easily) fixed for MariaDB in DBD::mysql.
It is already fixed in DBD::MariaDB: https://metacpan.org/pod/DBD::MariaDB

[loqs]

The issue with switching to DBD:MariaDB is that Arch Linux packages applications that hard code the user of DBD::mysql such as xtrabackup and perconna-toolkit [1]. Currently Arch is packaging MariaDB 10.1.36 but ideally it would package the latest stable release. Could DBD::MariaDB provide compatibility support for DBD::mysql?
[1] https://www.archlinux.org/packages/extra/x86_64/perl-dbd-mysql/

[pali]

DBD::MariaDB is DBI driver, so applications which uses DBI API and do not depend on internals, driver specific or undocumented features should work. DBD::MariaDB fixed lot of DBD::mysql bugs (Unicode, SSL, security, MariaDB related), but some applications depend on those bugs -- so these parts are for obvious reason not backward compatible.

[gregoa]

We're seeing this in Debian as well: https://bugs.debian.org/917303
Any ideas for fixing this issues would be appreciated, otherwise either MariaDB or DBD::mysql will get into troubles in Debian …

[pali]

Testing with MariaDB 10.3 was disabled:
c27bfd5

Maybe MariaDB 10.3 is not supported by DBD::mysql at all?

[dveeden]

I'll try to fix both this issue and see if we can work-around the issue that prevented testing from working.

[dveeden]

Update:

• I've found a workaround to get testing working for MariaDB 10.3
• t/57trackgtid.t didn't properly exclude MariaDB (GTID tracking is MySQL specific). Fixed.
• t/rt118977-zerofill.t breaks because zerofil handling. This is due to this bug: https://jira.mariadb.org/browse/MDEV-18143
• t/15reconnect.t had failures. Not diagnosed yet.

Commits are in https://github.com/dveeden/DBD-mysql/commits/issue_275

[dveeden]

• I've added a workaround for t/rt118977-zerofill.t.
• t/15reconnect.t hits a segfault in mariadb_reconnect(). This only seems to be the case when running against libmariadb.so from the server, not the latest MariaDB Connector/C. Looks like mysql_stmt_init() changes mysql->extension->conn_hdlr (from 0x0 to something). To be continued.

[dveeden]

Looks like the segfault happens when libmariadb was build against GnuTLS (either dynamic or static) and not if it was build against OpenSSL.

[gregoa]

Looks like the segfault happens when libmariadb was build against GnuTLS (either dynamic or static) and not if it was build against OpenSSL.

Thanks for all your work! libmariadb3 in Debian is built/linked against OpenSSL but t/15reconnect.t still fails. - Cf. https://bugs.debian.org/917303 I've now built the package locally 4 times; two times I got the same failure, but two times t/15reconnect.t ended after the 6th test: ``` t/15reconnect.t ......................... 1..16 ok 1 - Connected to database ok 2 - checking for active handle ok 3 - enabling reconnect ok 4 - enabling autocommit ok 5 - disconnecting active handle ok 6 - checking for inactive handle Failed 10/16 subtests ```

[loqs]

Just checked https://github.com/dveeden/DBD-mysql/commits/issue_275 all tests are now passing.
I echo gregoa's sentiments many thanks for your work in resolving this.
Perhaps I am not seeing a failure with t/15reconnect.t because the connection is using a Unix domain socket?

[dveeden]

I think t/15reconnect.t fails because of a libmariadb bug.
This fixes that: mariadb-corporation/mariadb-connector-c#95

[dveeden]

Looks like t/15reconnect.t was already discussed in #247
That discussion and https://jira.mariadb.org/browse/CONC-289 suggests that forcing a reconnect/init is a better solution.

[dveeden]

A new fix: 667d63c

If this passes w/o issues on Travis and when there are no objections then I will merge this and release a new version.

[snowman]

If you encounter this extract from above:

t/rt110983-valid-mysqlfd.t .............. ok

#   Failed test at t/rt118977-zerofill.t line 22.
#          got: '1'
#     expected: '00001'
# Looks like you failed 1 test of 8.

Try cpanm --notest DBD::mysql