MySQL Bugs: #33492: Falcon blob corruption on restart

Bug #33492	Falcon blob corruption on restart
Submitted:	23 Dec 2007 13:08	Modified:	5 May 2008 14:46
Reporter:	Philip Stoev	Email Updates:
Status:	Closed	Impact on me:	None
Category:	MySQL Server: Falcon storage engine	Severity:	S1 (Critical)
Version:	6.0.4	OS:	Any
Assigned to:	Jim Starkey	CPU Architecture:	Any

Description:
Falcon returns bad blob data after restart.

Test case will come shortly.

How to repeat:
1. Create a table with a blob field and update it.
2. Restart the server.
3. SELECT from the blob -- you will get corrupted data.

Pre-restart part of bug 33492

Attachment: bug33492-part1.test (application/octet-stream, text), 1.44 KiB.

Post-restart part of bug 33492

Attachment: bug33492-part2.test (application/octet-stream, text), 19 bytes.

Include file for test case of bug 33492

Attachment: 64K.inc (text/plain), 64.03 KiB.

Please find attached the test case for this bug. To run it, please follow the following procedure:

1. Manually start a fresh 6.0.4 server
2. Run the first part of the test case:
 perl mysql-test-run.pl --record --extern bug33492-part1
3. Shutdown and restart server manually using "mysqladmin shutdown"
2. Run the second part of the test case:
 perl mysql-test-run.pl --extern bug33492-part2

Test will fail, and corrupted data will be visible in the output.

Bug is also present in 6.0.3

I have been unable to reproduce this on a 4 processor, AMD64 SuSE 10.2
system.

Could somebody send me a zip file of data directory after the failure.  I
would also appreciate the actual output of the failure.

I am able to verify this bug with the latest BK tree.

Once you run the sequence of steps, you may see that main.bug33492-part2 may pass. Please go to mysql-test/r/bug33492-part2.result and see the garbage that is being returned.

[philips@philips mysql-test]$ uname -a
Linux philips 2.6.21-1.3194.fc7 #1 SMP Wed May 23 22:35:01 EDT 2007 i686 i686 i386 GNU/Linux

Fedora release 7 (Moonshine)
Kernel \r on an \m

vmware player with 385MB of memory.

Corrupted datadir uploaded to  ftp://ftp.mysql.com/pub/mysql/upload/bug33492.zip

Bug is also reproducible on ia64 using the 32-bit binary from

production.mysql.com:/data0/mysqldev/my/build-200712031732-6.0.4-alpha/mysql-6.0.4-alpha-build/dist/packages/mysql-6.0.4-alpha-linux-i686-glibc23.tar.gz

ftp://ftp.mysql.com/pub/mysql/upload/bug33492-before.zip

This archive contains the data directory after server shutdown, however before server restart.

Database was actually ok.  Problems was with Falcon meta-data handling
during startup.

I can confirm that this bug is no longer present when using a patched binary from

/data0/mysqldev/my/mysql-6.0.4-alpha-p1-build/dist/packages/mysql-6.0.4-alpha-p1-linux-i686-glibc23.tar.gz

Even if the blob was inserted in a previous 6.0.4 build, the latest 6.0.4 build can read it correctly, which is consistent with Mr. Sharkey's description of the cause of this bug.

Patch is in mysql-6.0-release version 6.0.4

Noted in 6.0.4 changelog.

After a server restart, Falcon mishandled metadata, resulting in
apparent corruption of BLOB data.