BF abort does not work with inserts to table with single unique key

I have seeing no such a problem in 5.5.34-25.9. It's now present in testing. Perhaps, this is а fresh-made bug.

This issue is a side effect of the fix for lp:1280896.

The fix was to both skip check for duplicate keys in secondary indexes *and* also avoid duplicate key error. Fixes pushed in revisions:

wsrep-5.5: http://bazaar.launchpad.net/~codership/codership-mysql/wsrep-5.5/revision/3983
wsrep-5.6: http://bazaar.launchpad.net/~codership/codership-mysql/5.6/revision/4078

Resolved

But now we can have two equal records on the unique key

SELECT id, login FROM users WHERE login LIKE 'andy002715724513%';
30684245 "andy002715724513"
30684248 "andy002715724513"

SELECT DISTINCT login FROM users WHERE login LIKE 'andy002715724513%';
"login"
"andy002715724513"

Table haves
  PRIMARY KEY (`id`),
  UNIQUE KEY `login` (`login`)

see full structure below

CREATE TABLE `users` (
  `id` int(10) unsigned NOT NULL AUTO_INCREMENT,
  `login` varchar(100) DEFAULT NULL,
  `pwd` varchar(100) DEFAULT NULL,
  `s_mail` varchar(50) DEFAULT NULL,
  `s_www` varchar(255) DEFAULT NULL,
  `s_inn` varchar(50) DEFAULT NULL,
  `s_descr` text,
  `s_phone` varchar(100) DEFAULT NULL,
  `offert_accepted` tinyint(3) unsigned NOT NULL DEFAULT '0',
  `s_full_name` varchar(255) DEFAULT NULL,
  `s_first_name` varchar(100) DEFAULT NULL,
  `s_middle_name` varchar(100) DEFAULT NULL,
  `s_last_name` varchar(100) DEFAULT NULL,
  `s_city` varchar(100) DEFAULT NULL,
  `s_address` varchar(255) DEFAULT NULL,
  `last_used` timestamp NOT NULL DEFAULT CURRENT_TIMESTAMP,
  `last_host_id` smallint(5) unsigned DEFAULT NULL,
  `mail_confirmed` tinyint(1) unsigned NOT NULL DEFAULT '0',
  `msisdn_confirmed` tinyint(1) unsigned NOT NULL DEFAULT '0',
  `msisdn_confirm_code` varchar(4) DEFAULT NULL,
  `msisdn_req_time` datetime DEFAULT NULL,
  `java_phone_model` varchar(30) DEFAULT NULL,
  `java_font` varchar(30) DEFAULT NULL,
  `java_safe_mode` tinyint(4) NOT NULL DEFAULT '0',
  `user_pic` varchar(16) DEFAULT NULL,
  `denied_libs` varchar(255) DEFAULT NULL,
  `recenser_type` int(11) DEFAULT NULL,
  `partner_id` int(11) DEFAULT NULL,
  `creat_date` datetime DEFAULT NULL,
  `partner` int(10) unsigned DEFAULT NULL,
  `partner_valid_till` date DEFAULT NULL,
  `partner_pin` varchar(32) DEFAULT NULL,
  `account` decimal(9,2) NOT NULL DEFAULT '0.00',
  `abonement_start` datetime DEFAULT NULL,
  `abonement_expires` date NOT NULL DEFAULT '2006-01-01',
  `abonement_period` smallint(6) DEFAULT NULL,
  `abonement_delay` tinyint(3) unsigned NOT NULL DEFAULT '0',
  `abonement_max_price` decimal(6,2) DEFAULT NULL,
  `abonement_downloads` tinyint(3) unsigned NOT NULL DEFAULT '0',
  `abonement_left_clicks` tinyint(3) unsigned NOT NULL DEFAULT '0',
  `abonement_left_summ` decimal(6,2) DEFAULT NULL,
  `user_pic_height` tinyint(3) unsigned DEFAULT NULL,
  `user_pic_width` tinyint(3) unsigned DEFAULT NULL,
  `show_pay_btn` tinyint(3) unsigned NOT NULL DEFAULT '0',
  `s_puid` varchar(255) DEFAULT NULL,
  `discount` decimal(4,4) NOT NULL DEFAULT '0.0000',
  `money_bonus` decimal(9,2) NOT NULL DEFAULT '0.00',
  `subscr_last_reminded` datetime DEFAULT NULL,
  `subscr_free_arts_given` datetime DEFAULT NULL,
  `subscr_type` tinyint(3) unsigned NOT NULL DEFAULT '2',
  `subscr_period` tinyint(3) unsigned DEFAULT '1',
  `subscr_content` tinyint(3) unsigned DEFAULT '2',
  `subscr_genres` text,
  `s_subscr_text_authors` text,
  `subscr_date` timestamp NOT NULL DEFAULT '0000-00-00 00:00:00',
  `s_subscr_text_pattern` text,
  `subscr_languages` text,
  `prefered_currency` char(3) DEFAULT NULL,
  `last_paymethod` tinyint(3) unsigned DEFAULT NULL,
  `utc_offset` char(6) DEFAULT NULL,
  `last_ip` varchar(15) DEFAUL...

And it only gets worse, yep, with the current release as well. Duplicate entry creation is the most funny thing after we had external keys "on delete cascade" not working.
Any idea on when is this going to be fixed?

The parent bug: lp:1280896 was re-opened and a better fix has been pushed in for further testing. Changing this bug also to fix committed state for now.

Percona now uses JIRA for bug reports so this bug report is migrated to: https://jira.percona.com/browse/PXC-1660