症状

atlassian-confluence.log に次のエラーが返される。

ORA-01555: snapshot too old: rollback segment number with name "" too small

You may also see ORA-22924 appear as a part of the error message:

ORA-01555: snapshot too old: rollback segment number with name "" too small
ORA-22924: snapshot too old

原因

The root cause is unknown at this time. This may be caused by:

• An unclean shutdown or other problems in the Oracle Database.
• Certain third party plugins that trying to update the values in some table when space that contains large/long pages are being view or edited.

ソリューション

First, we must track down where the problem is localized. By exporting the data, we can see which table (or tables) fail:

1. Shut down Confluence, and perform a full database backup
2. Using the expdp command, perform a full export of the Confluence Database Schema.
   1. An example command used could be expdp confuser/confpass DATA_PUMP_DIR=oracle_dir DUMPFILE=filename
   2. Note that the Confluence user must have full access to this directory - and that the directory is an Oracle Directory Object, rather than a file system directory
3. The export process should produce an error for one or more tables, similar to the one present in the Confluence logs. 

1. Create a new version of the affected table, that has the same columns, data types, constraints, and indexes.

   1. You can use something like this: 

      select dbms_metadata.get_ddl( 'TABLE', 'BANDANA', 'confuser' ) from dual

      confuser refers to the schema that the table is located in. Usually, that will be the same as your username that Confluence uses to access the database; although it may need to be adjusted to suit your environment.

   2. Use the CREATE TABLE query returned to create a new table, such as BANDANA_NEW

Now that we have a duplicate of the affected table, we'll insert rows into that table until we find the corrupt rows. Execute the following SQL:

SELECT CONCAT('INSERT INTO BANDANA_NEW (SELECT * FROM BANDANA WHERE BANDANAID=', CONCAT(BANDANAID, ');')) FROM BANDANA;

This produces a list of insert statements from the original table. Execute all of the statements. One or more of those statements should fail - at which point, you can identify the failing row, and delete it from the original table. Assuming that the row with the ID 12345 fails to insert into the duplicate table, you would use the following queries:

DELETE FROM BANDANA WHERE BANDANAID = 12345;

Repeat this process for all rows that failed to insert.

Cleaning Up

Delete the duplicate tables you created. In this example:

DROP TABLE BANDANA_NEW;

Then, start Confluence again.