commons-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Mark Thomas <>
Subject Re: [pool][performance] Source management question
Date Mon, 03 Jan 2011 15:17:20 GMT
On 25/12/2010 16:18, Phil Steitz wrote:
> Thanks, Gary!
> On Sat, Dec 25, 2010 at 9:11 AM, Gary Gregory
> <>wrote:
> Hm. Perfo already depends on pool. It would be less controversial to add the
>> test to perfo but that would not demonstrate the bug in pool itself.
>  I thought about that.  That would amount to attaching a [performance]
> config file to a [pool] JIRA ticket.  That would demonstrate the bug, but
> without a unit test in [pool] itself, we would have no way to ensure that it
> did not resurface.  Also, since [performance] is not released, the config is
> likely to continue to change.
> think I would still depend on perfo.
>> But [performance] is a sandbox component that has not been - and may never
> be - released, so this ends up as us a snapshot dependency, which I would
> like to avoid.

Sorry for coming to this late - I have been away for New Year.

The test - in what ever form - really needs to be in pool to ensure we
don't introduce a regression.

Here is a thought, what about an svn external? It can be pinned to HEAD
or a specific version. However, I'd be hapy with any of the other
solutions you suggested as well.


> Phil
>> Gary
>> On Dec 25, 2010, at 9:03, "Gary Gregory" <>
>> wrote:
>>> I would just have this new test depend on [performance] and be done with
>> it.
>>> Gary
>>> On Dec 25, 2010, at 0:53, "Phil Steitz" <> wrote:
>>>> I have found what I think is a bug in GKOP[1] using [performance].  I
>> need
>>>> the functionality in the Waiter and WaiterFactory classes in
>>>> o.a.c.performance.pool to build a test case showing the bug.  Having
>> these
>>>> classes available to pool's unit tests would be good.  I am not sure
>> what
>>>> the best approach is to make these classes available to the [pool] tests
>> and
>>>> would appreciate some advice.  I could just copy them, but I don't like
>> the
>>>> idea of maintaining both versions.  Even if [performance] was a proper
>>>> component and had a release, I don't much like the idea of adding a
>>>> dependency.  I could move them to [pool]'s test package, but then
>>>> [performance] could not access them unless [pool] were to export a test
>>>> jar.  Any ideas?  Thanks in advance.
>>>> Phil
>>>> [1] For those waiting with baited breath to learn what the bug is, it
>>>> manifests as maxActivePerKey exceeded by one.  This can happen when idle
>>>> instances retrieved from the pool fail validation and destroy has
>>>> non-trivial latency.  The problem is (I think) the result of clearOldest
>> not
>>>> updating the per-key internal processing counts.
>>> ---------------------------------------------------------------------
>>> To unsubscribe, e-mail:
>>> For additional commands, e-mail:
>> ---------------------------------------------------------------------
>> To unsubscribe, e-mail:
>> For additional commands, e-mail:

To unsubscribe, e-mail:
For additional commands, e-mail:

View raw message