Table containing cyrillic letters are destroyed

[Dmitriy Razumovskiy]

Hi,

I tried to sync differences between tables on master and slave with a command:
pt-table-sync --execute --replicate percona.checksums master1

As the result, the data both on master and slave were destroyed because all cyrillic characters were replaced with question marks.

The table collation is ut8_general_ci. In my.cnd under [mysqld] I have character-set-server=utf8

Versions:
Percona Server 5.6.35-80.0.1.jessie
Percona Toolkit 2.2.20-1

[Jaime Sicam (jssicam)]

Hi,
  Can you try using the --charset option to specify the default character set of pt-table-sync?

Eg. Compare the differences between these two runs:
pt-table-sync --print --charset=utf8 --replicate percona.checksums master1
pt-table-sync --print --charset=utf8 --replicate percona.checksums master1

Thanks.

[Jaime Sicam (jssicam)]

Oops. I meant:

pt-table-sync --print --replicate percona.checksums master1
pt-table-sync --print --charset=utf8 --replicate percona.checksums master1

Thanks!

[Jaime Sicam (jssicam)]

I'll set the status of this case to invalid because I think this can be resolved by setting --charset

[Dmitriy Razumovskiy (darland)]

Hi Jaime,

I appreciate your workaround, however I still believe that this behaviour is a very serious issue that has to be properly resolved for two reasons:

1. UTF-8 is default charset for the vast majority of applications and databases. Keeping data in UTF-8 means that you're secured from data corruption due to incompatible charset
2. In the manual of pt-table-sync there is no single mention, that running the tool with default setting most probably destroy your data. People like me carefully read the manual and with a false sense of safety run pt-table-sync agains a master on production environment. As the result - master and both slaves are destroyed

If pt-table-sync is really meant for production use I would suggest either:
- set default charset to UTF-8
- set charset as a required parameter
- (less safe) update the manual and write in bold, that the tool most probably destroy your data if you do not specify a charset of the DB

Leaving this bug without action in invalid status means that more and more people continue destroying the there databases and as a consequence - trust to Percona Community.

[Shahriyar Rzayev (rzayev-sehriyar)]

Percona now uses JIRA for bug reports so this bug report is migrated to: https://jira.percona.com/browse/PT-1412