From: John Embretsen
Date: January 16 2009 1:25pm
Subject: Please review: Test case for Bug#41870 (unbounded table space growth)
List-Archive: http://lists.mysql.com/falcon/386
Message-Id: <49708AB7.7070006@Sun.COM>
MIME-Version: 1.0
Content-Type: text/plain; charset=ISO-8859-1
Content-Transfer-Encoding: 7BIT

Hi all (especially Hakan and Kevin),

I have spent quite some time this week creating a regression test for
this bug, regarding unbounded table space growth during BLOB updates.

Bug: http://bugs.mysql.com/bug.php?id=41870
Test patch: http://lists.mysql.com/commits/63446

The bug was fixed this week when Kevin introduced new scavenger code.
The new test is called 'falcon_blob_space-big'. It is a --big-test
because it

  a) needs (currently) at least 13 MB of tablespace space (20 MB if
there is a regression)
  b) needs roughly 80-90 seconds running time if disk-based (yet only 3
if --mem is used) (on my test host), so it might time out in Pushbuild.

This means it will not run regularly in Pushbuild, but it will run as
part of our Weekly Falcon Test Overview, at least.
Let me know if you think the test can be made non-big.

I hope you/someone can take a look and see if this test makes sense.
Included comments should explain most of it, but don't hesitate to ask
if you have questions.

I have tested the new test on Linux and Solaris x86, both with and
without --mem. I also verified that it fails without the bugfix, in
which case it fails like this:

----<8--------<8--------<8--------<8--------<8

falcon.falcon_blob_space-big   [ fail ]

--- suite/falcon/r/falcon_blob_space-big.result
+++ suite/falcon/r/falcon_blob_space-big.reject
@@ -41,7 +41,12 @@

 ** Checking file size of default tablespace file (falcon_user.fts)...
 ** Tablespace growth threshold: 20971520 bytes
-** File size growth below test threshold. Test OK.
+FAILURE: Tablespace filesize growth exceeded our threshold!
+         Size of tablespace file was: 23695360
+         Growth during BLOB updates was: 22536192
+
+         Please investigate if this is a regression or if the
+         test case needs adjustments.

 ** Final checks and cleanup...


mysqltest: Result content mismatch

----<8--------<8--------<8--------<8--------<8

Also tested with the new test runner ("MTR 2" in rpl branch).


thanks,

-- 
John