Hi,

It seems that jaxb context init is still the culprit. We use the
com.sun.xml.bind.v2.ContextFactory , switching for example to the
Eclipselink implementation
(org.eclipse.persistence.jaxb.JAXBContextFactory) was even slower. Spring's
lazy init is no good here, makes me wonder if cxf could do a 'true' lazy
init and defer this expensive init of the service until actual methods are
called on it. Feasible ?

Jorg

On Mon, Oct 5, 2015 at 10:42 AM Jorg Heymans <[email protected]> wrote:

>
> The wsdlLocation attribute is not specified, and to be honest i have no
> idea if there is any remote downloading involved during startup. We just
> specify id and serviceClass attribute and let cxf do its thing. I will
> try to profile and report back.
>
> Jorg
>
> On Sun, Oct 4, 2015 at 5:23 PM Andrei Shakirin <[email protected]>
> wrote:
>
>> Hi,
>>
>> 2.5 to 3 minutes is quite long even for 15 clients initialization.
>> Did you specify wsdlLocation attribute in jaxws:client? Can performance
>> be related to remote WSDL downloading?
>> Could you bind any profiling tool and discover which operation caused
>> performance problem?
>>
>> Regards,
>> Andrei.
>>
>> > -----Original Message-----
>> > From: Jorg Heymans [mailto:[email protected]]
>> > Sent: Donnerstag, 1. Oktober 2015 08:45
>> > To: [email protected]
>> > Subject: ReflectionServiceFactoryBean performance
>> >
>> > Hi,
>> >
>> > We have about 15 jaxws client definitions in our application context
>> defined
>> > like this :
>> >
>> > <jaxws:client id="myService" serviceClass="my.service.Service"
>> >  address="http://....."/>
>> >
>> > Initializing all of these during startup takes on average about 2.5 to
>> 3 minutes.
>> > This is already after adding -
>> > Dcom.sun.xml.bind.v2.runtime.JAXBContextImpl.fastBoot=true , before
>> that it
>> > was more like 5-6 minutes.
>> >
>> > Is there a way to improve this ? We are going to add more services as
>> the
>> > application grows, and already now this cxf init takes up more than
>> half of our
>> > deployment time.
>> >
>> > Thanks,
>> > Jorg Heymans
>>
>

Reply via email to