I've observed a case where pt-online-schema-change fails to nibble from the old to new table because the new table doesn't exist. This happens because it's trying to nibble after dropping (although I don't see the rename/swap in the query log, which may be part of the bug). The behavior happens about 10-20% of the time for me on the t/pt-online-schema-change/alter_active_table.t test as of revision 262 in this branch: lp:/~percona-toolkit-dev/percona-toolkit/stabilize-test-suite/
Here is the test output I see:
[baron@localhost stabilize-test-suite]$ perl t/pt-online-schema-change/alter_active_table.t
1..9
# Loading sample dataset...
# Binlog position before altering table: mysql-bin.000001/3686358
not ok 1 - Altered OK
# Failed test 'Altered OK'
# in t/pt-online-schema-change/alter_active_table.t at line 184.
# 'Error copying rows from `pt_osc`.`t` to `pt_osc`.`_t_new`: DBD::mysql::st execute failed: Table 'pt_osc._t_new' doesn't exist [for Statement "INSERT LOW_PRIORITY IGNORE INTO `pt_osc`.`_t_new` (`id`, `c`, `d`) SELECT `id`, `c`, `d` FROM `pt_osc`.`t` FORCE INDEX(`PRIMARY`) WHERE ((`id` >= ?)) AND ((`id` <= ?)) /*pt-online-schema-change 19490 copy nibble*/" with ParamValues: 1='515', 0='504'] at /home/baron/stabilize-test-suite//bin/pt-online-schema-change line 6734.
#
# '
# doesn't match '(?-xism:Successfully altered `pt_osc`.`t`)'
Naturally, everything fails after this, because the tool has died.
I will attach the output of the query log and the output of the binary log. You can see that in the query log, at line 180, the DROP TABLE happens. Nibbling is apparently finished by this time, but then a final nibble appears to occur at line 204. This is the one that breaks the tool.
This is running on CentOS 5.5 with 5.5.25-log MySQL Community Server (GPL)