[jira] Updated: (DERBY-1786) a crash during re-encryption of an existing database with lot of tables can make database unrecoverable on a next boot.

["Suresh Thalamati (JIRA)" <de...@db.apache.org>]

     [ http://issues.apache.org/jira/browse/DERBY-1786?page=all ]

Suresh Thalamati updated DERBY-1786:
------------------------------------

    Attachment: derby-1786.diff

The problem was when transaction log spans more than one log file during (re)
encryption of the database and if there is a crash just before switching the
database to use the new encryption properties; On recovery checkpoint in the
first log file is used as reference and the next log file is assumed to have
the commit log record for (re) encryption and deleted incorrectly to force the
roll-back , which lead to the incomplete rollback of re-encryption. And that
caused recovery failures on next (re) encryption crashed. 

This patch fixes the problem by ensuring there a checkpoint record in the
last log file before creating a new log file with new encryption properties
and writing the commit log record. Log is also flushed before making the
transaction log use the new encryption key to avoid any part of old log 
records in the buffers getting encrypted with the new encryption key. 
While working on this problem , I noticed error message thrown incase of 
re-encryption failures are confusing, added a new error message to indicate 
failures specific to (re) encryption. 


/suresh

> a crash during re-encryption of an existing  database with lot of tables can make database unrecoverable on a  next boot.
> -------------------------------------------------------------------------------------------------------------------------
>
>                 Key: DERBY-1786
>                 URL: http://issues.apache.org/jira/browse/DERBY-1786
>             Project: Derby
>          Issue Type: Bug
>    Affects Versions: 10.2.1.0, 10.3.0.0
>            Reporter: Suresh Thalamati
>         Assigned To: Suresh Thalamati
>         Attachments: derby-1786.diff, stk.txt
>
>
> While running reencryt  crash/recovery functional test  with  1000 tables (>2000 containers) , I hit a following boot failire. 
> ERROR XJ040: Failed to start database 'wombat_pwd_en', see the next exception fo
> r details.
> ERROR XSTB0: An exception was thrown during transaction abort.
> java.sql.SQLException: Failed to start database 'wombat_pwd_en', see the next ex
> ============= begin nested exception, level (1) ===========
> ERROR XSRS4: Error renaming file (during backup) from E:\suresht\adhoctests\reencryption\dbs\wombat_pwd_en\seg0\c340.dat to E:\suresht\adhoctests\reencryption\dbs\wombat_pwd_en\seg0\n340.dat.
> ---
> After bit of  debugging ; my suspicion  is this problem is happening becuase of more than one log file switch during re-encryption and current 
> re-encryption recovery code does not seem to handle it correctly. 
>   

-- 
This message is automatically generated by JIRA.
-
If you think it was sent incorrectly contact one of the administrators: http://issues.apache.org/jira/secure/Administrators.jspa
-
For more information on JIRA, see: http://www.atlassian.com/software/jira