qpid-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From James Birdsall <jb...@microsoft.com>
Subject RE: Store tests
Date Tue, 10 Nov 2009 03:33:10 GMT
On Monday, November 09, 2009 12:41 PM, Alan Conway wrote:

>On 11/09/2009 11:45 AM, Kim van der Riet wrote:
>> It would be a good idea to standardize store tests if we can. If we can
>> agree that no matter what store is used, the end-result from a client
>> perspective would be the same, then tests would be fairly universal and
>> can be run against any store module.
>> All tests outlined below (which are based on some of the current async
>> store tests) assume a clean broker start with the store module loaded
>> (with no store content on disk) and stopping broker on conclusion for
>> each test (which may or may not leave messages in the store) - so there
>> needs to be some mechanism to clean up store artifacts between tests.
>> However, if tests are carefully constructed, it may be possible to avoid
>> a disk cleanup between tests by ensuring exchange and queue names do not
>> overlap between tests. No test should leave the store in an
>> unrecoverable state.
>The python broker tests framework I've been working on creates a separate 
>working directory for each test, all under a common directory so its easy to 
>clean up between test runs so there's no need to worry about conflicts between 
>> Tests marked (SERVER-SIDE) may need to be run from the broker itself or
>> on the same machine as the broker in order to gain access to store
>> artifacts.
>And they'd be easy enough to write in python.
>> Suggestions welcome.
>The set below looks good. In particular I think the idea of having the tests be 
>based on user-observable outcomes is the right way to go.

Tests for persistence providers came up a while back; I'm in the middle of writing some for
the SQL-based provider that Steve Huston is working on. My test plan is more basic, since
I'm new to AMQP and Qpid and don't have the detailed understanding of it yet, but even so
my tests have shaken out a number of bugs. The test plan is 99% not specific to the provider;
at some point I would like to try it against other providers, just to see what happens.

The mailing list strips attachments and I don't have permissions to add a page to the wiki
(I can't even LOOK at most of the testing pages!), so I will send the PDF to Kim directly.
If anyone else would like to see it, please let me know!

--James B.

Apache Qpid - AMQP Messaging Implementation
Project:      http://qpid.apache.org
Use/Interact: mailto:dev-subscribe@qpid.apache.org

View raw message