db-derby-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Andrew McIntyre" <mcintyr...@gmail.com>
Subject Re: [Proposal for Review] The fix for DERBY-3347 is important enough for the Development Community to issue a recommendation the User Community to upgrade.
Date Wed, 30 Apr 2008 07:52:39 GMT
Updating Stan's proposed announcement based on recent feedback...

On Mon, Apr 28, 2008 at 5:23 PM, Stanley Bradbury
<Stan.Bradbury@gmail.com> wrote:


   *** For Review and comment  ***

  NOTICE TO ALL DERBY v10.3 USERS : CRITICAL FIX NOW AVAILABLE

  The Bottom Line:

  If you are currently using Derby 10.3.1.4 or Derby 10.3.2.1, it is strongly
  recommended that you upgrade to Derby 10.4.1.3 or 10.3.3.0 to avoid
  any chance of database corruption due to an issue with multiple threads
  accessing a database that is documented in <a
href="issues.apache.org/jira/browse/DERBY-3347">DERBY-3347</a>.

  This bug can cause unrecoverable database corruption during periods of
  heavy, multi-thread I/O operations.  The error produced in the test case
  used to diagnose the problem was:

  ERROR XSDB3: Container information cannot change once written: was 0, now 80

  It is felt that other errors might also be generated when this type of
  corruption occurs.  The corruption message will most likely refer to page 0
  of the container. For example:

  ERROR XSDG1: Page Page(0 ,Container(0, 5856)) could not be
  written...

  This bug corrupts the pages on disk and can go unnoticed.  If you do not
  run database consistency checks regularly it is recommended you begin doing
  so as soon as possible after the upgrade.  To insure that corruption has not
  already occurred in existing databases, after upgrade run the database
  consistency check at least once to validate all tables in the database.  This
  process is documented at:

  http://wiki.apache.org/db-derby/DatabaseConsistencyCheck

  If corruption has already occurred the database will need to be recovered
  from the last good backup. There is no alternative solution.

  Version 10.3.3.0 can be downloaded from:
  http://db.apache.org/derby/releases/release-10.3.3.0.cgi

  Version 10.4.1.3 can be downloaded from:
  http://db.apache.org/derby/releases/release-10.4.1.3.cgi

...

Please let me know if you think there is additional information that
is required. I think addressing this to Derby v10.3 users, along with
the specific address to 10.3.1 and 10.3.2 users, should cover everyone
affected.

Are we sure this only affects Windows? If not, I say we leave platform
specific details out. Other mail from Knut made it appear that this
was definitely not a platform specific problem, but maybe platform
specific JVM details make it harder to reproduce on non-Windows
platforms.

Thanks, Knut, for writing the release note. I don't really have
anything to add, except maybe we should add exactly why we think the
problem occurs (multiple threads accessing first page in heavily
contended scenario). Maybe that's too much information for an end
user.

I propose also that we remove 10.3.1.4 and 10.3.2.1 from the download
page. Do we also remove them from the archive? That is definitely
something to check. It may be important to retain these from an Apache
infrastructure/archive perspective. I'll follow up on that later.

andrew

Mime
View raw message