db-derby-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From John Embretsen <John.Embret...@Sun.COM>
Subject Re: Testing for 10.1.3
Date Fri, 02 Jun 2006 22:24:34 GMT
Ole Solberg wrote:

> We now have extended our tests for the 10.1.2.4 Snapshot:

[SNIP]

> 2. New tests:
>    **********

[SNIP]

> "Long Running test":
> --------------------
> Using the DOTS ATCJ2(auction simulation) we have had a database running 
> for 14 days:
> no issues seen so far.

Here is an update on the long running test:

This test was started approximately 22 days ago. Yesterday, the Derby 
Network Server JVM threw an "OutofMemoryError: Java heap space", i.e. 
after approximately 21 days and 10 hours.

My question to the Derby community is:

Has anyone ever tested any version of Derby for such a long period of
time?

I am asking because I would like to find out if this is a test
"weakness" or a Derby memory leak (or both).


More details on the test:

The test was run on Sun JDK 1.5.0 (max heap sizes 128 MB), remote
client/server, using the DOTS ATCJ2 test case (modified). The original
ATCJ2 test caused OutOfMemoryError after only a few hours with a similar
heap size, so I went through the source and removed much of the really
"bad" stuff discussed here:
http://www.nabble.com/OutOfMemoryErrors-when-testing-Derby-with-DOTS-t1010027.html

In the modified version of the test, PreparedStatements are reused
throughout the "simulation", and "?" placeholders are used instead of
simple String concatenation.

The same test was run against the trunk not so long ago, and was able
to run for at least 4 weeks without producing the same error (although
memory usage graphs seem rather similar at first glance). I am assuming
that this increased robustness (10.2 vs. 10.1) has something to do with
the fact that DERBY-210 patches are in trunk, but not in the 10.1
branch, as far as I know.


> We intend to repeat these tests on the final 10.1.3 snapshot/release 
> candidate.

Yet, given that it takes more than 21 days to run into this particular
issue, it will not be discovered/verified in time for the actual
release, if the next test run is based on the release candidate.


-- 
John


Mime
View raw message