Re: [jetty-users] Contention in ServletHolder.handle()

[Daniel Feist <dfeist@xxxxxxxxx>]

After doing some performance testing, while contention shows in
profiler clearly the performance impact is not significant I don't
thnk, so this would just be minor cleanup/optimization.

Was i right when i set I need to use "initOrder" to ensure it gets
initialized at startup?

I should really just be implementing a handler, but current
implementation uses a servlet.

thanks!

On Mon, May 12, 2014 at 5:15 PM, Jan Bartel <janb@xxxxxxxxxxx> wrote:
> Daniel,
>
> let me see if I can prevent the call to getServlet() in the case where
> you provide a pre-constructed servlet. I'll do some testing and get
> back to you - I need to confirm that the handling of that would be ok
> for cases where the servlet may have been made unavailable etc etc as
> well.
>
> Jan
>
> On 12/05/2014, Daniel Feist <dfeist@xxxxxxxxx> wrote:
>> Hi,
>>
>> Yes.  I agree this is needed is 'initOnStartup' is not configured, was
>> trying to work out why the synchonrized block was needed at all for
>> the 'initOnStartup' case though.
>>
>> That said, the reason reason I was seeing contention is probably
>> primarily due to org.eclipse.jetty.servlet.ServletHolder.getServlet()
>> because my servlet doesn't have 'initOnStartup' set.
>>
>> In my use case I embed jetty and provide Jetty with a
>> pre-instantiated/configured servlet instance wrapped in a
>> ServletHolder.  I guess i need to set the somewhat unobvious initOrder
>> to non-0 to configure this and reduce the contention?
>>
>> thanks!
>>
>>
>>
>> On Mon, May 12, 2014 at 11:15 AM, Jan Bartel <janb@xxxxxxxxxxx> wrote:
>>> Daniel,
>>>
>>> Not sure exactly the lines in ServletHolder you're referring to, but I
>>> assume its the synchronized block around line 652? If so, then this
>>> code guards the initial startup of a servlet - if it isn't
>>> load-on-startup, then any request coming in can initialize it.
>>> Therefore, if you have multiple concurrent requests, you need to
>>> protect the servlet startup so it happens only once.
>>>
>>> Jan
>>>
>>> On 8 May 2014 22:43, Daniel Feist <dfeist@xxxxxxxxx> wrote:
>>>> Hi,
>>>>
>>>> I've been doing some testing/benchmarking of Jetty 8 with Mule ESB and
>>>> seeming a decent amount of contention where it doesn't seem that
>>>> synchronization is required.
>>>>
>>>> My test scenario is just a simple echo, all i do is read the request
>>>> stream and then respond.  The contention i'm seeing can be seen here:
>>>>
>>>> https://www.dropbox.com/s/bbkdsk7ndfpzoj9/Screen%20Shot%202014-05-08%20at%209.35.34%20PM.png
>>>>
>>>> Inspecting the code, I can't understand why this needs to even be
>>>> synchonrized.
>>>>
>>>> Any thoughts? Is there a config option I should be using that i'm not?
>>>>  Is my test scenario unrealistic?
>>>>
>>>> thanks!
>>>> Dan
>>>> _______________________________________________
>>>> jetty-users mailing list
>>>> jetty-users@xxxxxxxxxxx
>>>> https://dev.eclipse.org/mailman/listinfo/jetty-users
>>>
>>>
>>>
>>> --
>>> Jan Bartel <janb@xxxxxxxxxxx>
>>> www.webtide.com
>>> 'Expert Jetty/CometD developer,production,operations advice'
>>> _______________________________________________
>>> jetty-users mailing list
>>> jetty-users@xxxxxxxxxxx
>>> https://dev.eclipse.org/mailman/listinfo/jetty-users
>> _______________________________________________
>> jetty-users mailing list
>> jetty-users@xxxxxxxxxxx
>> https://dev.eclipse.org/mailman/listinfo/jetty-users
>>
>
>
> --
> Jan Bartel <janb@xxxxxxxxxxx>
> www.webtide.com
> 'Expert Jetty/CometD developer,production,operations advice'
> _______________________________________________
> jetty-users mailing list
> jetty-users@xxxxxxxxxxx
> https://dev.eclipse.org/mailman/listinfo/jetty-users