Chris, Can you try something? Create a single proxy and keep that proxy around (like in a static or something) for the duration of the test. You don't need to use it, it just needs to stay around.
We do do some caching of JAXB contexts (we should probably cache the schemas that go with that context), but those are stored in a WeakHashMap type thing so if something isn't kept around to hold onto them, they would get collected and have to be recreated. Likewise, the wsdl and and schemas from the wsdl need something holding onto them to keep them from being collected. Dan On Friday 28 March 2008, Christopher Cheng wrote: > You could try to generate java codes using " > org.apache.cxf.tools.wsdlto.WSDLToJava" with the following wsdl and > see the performance. > > http://webservices.sabre.com/wsdl/sabreXML1.0.00/tpf/OTA_HotelAvailLLS >1.5.1RQ.wsdl > > If it's jaxb that's causing the performance problem, which databinding > in your opinion has the best performance? > > > On Fri, Mar 28, 2008 at 11:26 PM, Benson Margulies > <[EMAIL PROTECTED]> > > wrote: > > Please supply some sort of test case. I've profiled extensively. I > > don't believe that service caching, as discussed, would have any > > effect. The current 2.1 code (as ported by Dan to 2.0.5) appears to > > be constrained by the performance of JAXB, and I don't think we have > > a caching opportunity there, though I would love to be wrong. > > Perhaps there's something about your > > services that use a lot more time in the service model code than the > > sample > > I've been profiling. > > > > On Fri, Mar 28, 2008 at 8:13 AM, Christopher Cheng < > > > > [EMAIL PROTECTED]> wrote: > > > I just tried to rerun it with the snapshot 2.0.5. It is running > > > about > > > > 20% > > > > > faster, but it still takes 3 seconds to build services. > > > I hope that the service caching will be available soon so I can > > > migrate the > > > current axis-1.2 application to cxf > > > > > > On Thu, Mar 20, 2008 at 7:33 PM, Daniel Kulp <[EMAIL PROTECTED]> wrote: > > > > On Thursday 20 March 2008, Christopher Cheng wrote: > > > > > Looks like 2.0.5 is going to release soon, is this service > > > > > caching > > > > in > > > > > > > 2.0.5? > > > > > > > > Yea. The stuff that Benson has done is being ported to 2.0.x. > > > > > > > > That said, it's not the full service caching. It's basically a > > > > bunch > > > > > > of > > > > > > > incremental improvements to the current algorithms, but it's > > > > definitely > > > > > > a big improvement. > > > > > > > > Dan > > > > > > > > > On Wed, Mar 5, 2008 at 11:50 AM, Christopher Cheng < > > > > > > > > > > [EMAIL PROTECTED]> wrote: > > > > > > Looking forward to have this in the next release 2.0.5 or > > > > > > 2.1 :) > > > > > > > > > > > > On Tue, Mar 4, 2008 at 11:55 PM, Daniel Kulp > > > > > > <[EMAIL PROTECTED]> > > > > > > > > wrote: > > > > > > > Willem, > > > > > > > > > > > > > > The issue is that they CAN be modified after that. One > > > > > > > popular modification is to grab the EndpointInfo out of it > > > > > > > and change > > > > the > > > > > > > > > address. Other things like adding interceptors to the > > > > > > > BindingInfo, changing properties on the service (like to > > > > > > > enable schema validation), etc... are all modifications > > > > > > > that affect the servicemodel. > > > > > > > > > > > > > > Databinding is "less" of an issue, but still one. For > > > > > > > example, the databinding has an "mtomThreshold" and a > > > > > > > namespaceMap that > > > > can > > > > > > > > > be set programatically. > > > > > > > > > > > > > > In both of those cases, a modification should only affect > > > > > > > the proxy on which the modification is made. Thus, the > > > > servicemodel > > > > > > > > > would need a full "clone", which may not be easy to do. > > > > > > > It's certainly a lot more code to add. > > > > > > > > > > > > > > That said, I don't think it's a bad idea at all. On the > > > > contrary, > > > > > > > > > it's a > > > > > > > good idea. It's just a bit of work. > > > > > > > > > > > > > > Dan > > > > > > > > > > > > > > On Tuesday 04 March 2008, Willem Jiang wrote: > > > > > > > > Hi Dan, > > > > > > > > > > > > > > > > I think we could cache the ServiceInfo object and > > > > > > > > DataBinding object which will take lots of time to be > > > > > > > > created and can't be modified after that. Any thought? > > > > > > > > > > > > > > > > Willem > > > > > > > > > > > > > > > > Daniel Kulp wrote: > > > > > > > > > On Tuesday 04 March 2008, Christopher Cheng wrote: > > > > > > > > >> I haven't read the codes of Axis, but I have been > > > > > > > > >> using it for 2 years and it doesn't have this > > > > > > > > >> problem. Having an overhead for 5-8 seconds is not an > > > > > > > > >> option for a production system in term of performance > > > > > > > > >> > > > > > > > > >> What you are saying is that all the services must > > > > > > > > >> implement Clonable interface? > > > > > > > > > > > > > > > > > > No, not the services themselves. Just our interal > > > > > > > > > service model stuff that holds all the metadata. > > > > > > > > > Basically, when creating a new service, we need to > > > > > > > > > calculate the metadata. We should cache that if > > > > > > > > > possible. > > > > > > > > > > > > > > > > > > Dan > > > > > > > > > > > > > > > > > >> On Tue, Mar 4, 2008 at 4:48 AM, Daniel Kulp > > > > > > > > >> <[EMAIL PROTECTED]> > > > > > > > > > > > > > > wrote: > > > > > > > > >>> Christopher, > > > > > > > > >>> > > > > > > > > >>> We cannot just "reuse" the Service model objects as > > > > > > > > >>> they > > > > are > > > > > > > > > > >>> completely modifiable at runtime. For example, one > > > > > > > > >>> proxy could be reconfigured with new address or have > > > > > > > > >>> additional interceptors added or similar. Those > > > > > > > > >>> shouldn't affect others. > > > > > > > > >>> > > > > > > > > >>> On option that propably would make sense is to add a > > > > > > > > >>> fast "clone" functionality to the entire service > > > > > > > > >>> model so we could cache one, and then clone it > > > > > > > > >>> whenever one is really needed. All the individual > > > > > > > > >>> versions could be modified > > > > and > > > > > > > > > > >>> not affect the original. However, doing that would > > > > > > > > >>> be > > > > quite > > > > > > > > > > >>> a bit of work as there are a LOT of classes that > > > > > > > > >>> would > > > > need > > > > > > > > > > >>> to be updated to support that. > > > > > > > > >>> > > > > > > > > >>> Patches towards that end would be great. :-) > > > > > > > > >>> > > > > > > > > >>> Dan > > > > > > > > >>> > > > > > > > > >>> On Monday 03 March 2008, Christopher Cheng wrote: > > > > > > > > >>>> wsdl is indeed cached in > > > > > > > > >>>> WSDLManagerImpl.definitionsMap > > > > > > > > >>>> > > > > > > > > >>>> After reading the codes, I have some questions. My > > > > > > > > >>>> workstation is a PentiumD 3GHz with 1.5GB RAM > > > > > > > > >>>> > > > > > > > > >>>> In the class > > > > > > > > >>>> " > > > > org.apache.cxf.service.factory.ReflectionServiceFactoryBea > > > > > > > > > > >>>>n.buil d Serv iceFromWSDL(String url)", > > > > > > > > >>>> "setService(factory.create());" takes 3 seconds > > > > > > > > >>>> "getDataBinding().initialize(getService());" takes > > > > > > > > >>>> 2 seconds > > > > > > > > >>>> > > > > > > > > >>>> These 2 methods are called even in 2nd calls. Can > > > > > > > > >>>> you > > > > also > > > > > > > > > > >>>> cache the services perhaps using jakarta > > > > > > > > >>>> commons-pool? or eh-cache? > > > > > > > > >>>> > > > > > > > > >>>> By the way, I found that the wsdl is put into the > > > > > > > > >>>> definitionMap twice if " > > > > > > > > >>>> WSDLManagerImpl.getDefinition > > > > (URL > > > > > > > > > > >>>> url)" is called. First in "getDefintion(URL url)" > > > > > > > > >>>> at line 147 and second in "loadDefinition(String > > > > > > > > >>>> url)" at line > > > > 201. > > > > > > > > > > >>>> On Mon, Mar 3, 2008 at 4:22 PM, Christopher Cheng < > > > > > > > > >>>> > > > > > > > > >>>> [EMAIL PROTECTED]> wrote: > > > > > > > > >>>>> wsdl is indeed cached in > > > > > > > > >>>>> WSDLManagerImpl.definitionsMap > > > > > > > > >>>>> > > > > > > > > >>>>> After reading the codes, I have some questions. My > > > > > > > > >>>>> workstation is a PentiumD 3GHz with 1.5GB RAM > > > > > > > > >>>>> > > > > > > > > >>>>> In the class " > > > > org.apache.cxf.service.factory.ReflectionServiceFactoryBea > > > > > > > > > > >>>>>n.buil dSer viceFromWSDL(String url)", > > > > > > > > >>>>> "setService(factory.create());" takes 3 seconds > > > > > > > > >>>>> "getDataBinding().initialize(getService());" takes > > > > > > > > >>>>> 2 seconds > > > > > > > > >>>>> > > > > > > > > >>>>> These 2 methods are called even in 2nd calls. Can > > > > > > > > >>>>> you > > > > also > > > > > > > > > > >>>>> cache the services perhaps using jakarta > > > > > > > > >>>>> commons-pool? > > > > or > > > > > > > > > > >>>>> eh-cache? > > > > > > > > >>>>> > > > > > > > > >>>>> By the way, I found that the wsdl is put into the > > > > > > > > >>>>> definitionMap twice if " > > > > > > > > >>>>> WSDLManagerImpl.getDefinition > > > > (URL > > > > > > > > > > >>>>> url)" is called. First in "getDefintion(URL url)" > > > > > > > > >>>>> at > > > > line > > > > > > > > > > >>>>> 147 and second in "loadDefinition(String url)" at > > > > > > > > >>>>> line 201. > > > > > > > > >>>>> > > > > > > > > >>>>> > > > > > > > > >>>>> > > > > > > > > >>>>> On Mon, Mar 3, 2008 at 9:43 AM, Willem Jiang > > > > > > > > >>>>> <[EMAIL PROTECTED]> > > > > > > > > >>>>> > > > > > > > > >>>>> wrote: > > > > > > > > >>>>>> Hi, > > > > > > > > >>>>>> > > > > > > > > >>>>>> I just went through the code, we really cache the > > > > > > > > >>>>>> WSDL definition in CXF. > > > > > > > > >>>>>> Could you send your test case and wsdl file to me > > > > > > > > >>>>>> ? I > > > > may > > > > > > > > > > >>>>>> need to trace it for more information. > > > > > > > > >>>>>> > > > > > > > > >>>>>> Regards, > > > > > > > > >>>>>> > > > > > > > > >>>>>> Willem > > > > > > > > >>>>>> > > > > > > > > >>>>>> Christopher Cheng wrote: > > > > > > > > >>>>>>> I think the issue is still there. I used a for > > > > > > > > >>>>>>> loop to execute the > > > > > > > > >>>>>> > > > > > > > > >>>>>> same > > > > > > > > >>>>>> > > > > > > > > >>>>>>> call, all takes approximately the same time to > > > > > > > > >>>>>>> create service. I guess > > > > > > > > >>>>>> > > > > > > > > >>>>>> that > > > > > > > > >>>>>> > > > > > > > > >>>>>>> the service itself is not cached. > > > > > > > > >>>>>>> > > > > > > > > >>>>>>> On Sun, Mar 2, 2008 at 11:43 AM, Willem Jiang > > > > > > > > >>>>>>> <[EMAIL PROTECTED]> > > > > > > > > >>>>>>> > > > > > > > > >>>>>>> wrote: > > > > > > > > >>>>>>>> Here is the JIRA[1] which describe the same > > > > > > > > >>>>>>>> thing > > > > that > > > > > > > > > > >>>>>>>> you want. I don't know if it was really > > > > > > > > >>>>>>>> resolved, > > > > could > > > > > > > > > > >>>>>>>> you try the latest released CXF 2.0.4 for it. > > > > > > > > >>>>>>>> > > > > > > > > >>>>>>>> If the issue is still there, please let me know > > > > > > > > >>>>>>>> , I will put it to my next week todo list. > > > > > > > > >>>>>>>> > > > > > > > > >>>>>>>> [1]https://issues.apache.org/jira/browse/CXF-69 > > > > > > > > >>>>>>>>9 [2]http://cwiki.apache.org/CXF/download.html > > > > > > > > >>>>>>>> > > > > > > > > >>>>>>>> Willem. > > > > > > > > >>>>>>>> > > > > > > > > >>>>>>>> Christopher Cheng wrote: > > > > > > > > >>>>>>>>> I am migrating from Axis1.2 to CXF 2.0.3 > > > > > > > > >>>>>>>>> > > > > > > > > >>>>>>>>> I understand that it will take a long time to > > > > > > > > >>>>>>>>> build for the first > > > > > > > > >>>>>> > > > > > > > > >>>>>> time. > > > > > > > > >>>>>> > > > > > > > > >>>>>>>> What > > > > > > > > >>>>>>>> > > > > > > > > >>>>>>>>> I am wondering is that why it takes so long > > > > > > > > >>>>>>>>> for the second and third > > > > > > > > >>>>>>>> > > > > > > > > >>>>>>>> time? > > > > > > > > >>>>>>>> > > > > > > > > >>>>>>>>> Is there any caching of services? Axis does > > > > > > > > >>>>>>>>> not seem to have this > > > > > > > > >>>>>> > > > > > > > > >>>>>> issue. > > > > > > > > >>>>>> > > > > > > > > >>>>>>>>> Christopher Cheng wrote: > > > > > > > > >>>>>>>>>> Attached is the log > > > > > > > > >>>>>>>>>> > > > > > > > > >>>>>>>>>> ----- Original Message ----- > > > > > > > > >>>>>>>>>> From: "Willem Jiang" <[EMAIL PROTECTED]> > > > > > > > > >>>>>>>>>> To: <[email protected]> > > > > > > > > >>>>>>>>>> Sent: Saturday, March 01, 2008 8:53 PM > > > > > > > > >>>>>>>>>> Subject: Re: service caching? > > > > > > > > >>>>>>>>>> > > > > > > > > >>>>>>>>>>> Hi > > > > > > > > >>>>>>>>>>> > > > > > > > > >>>>>>>>>>> Could you set the logger level to FINE ? > > > > > > > > >>>>>>>>>>> So we can get more information about the > > > > > > > > >>>>>>>>>>> service publishing. > > > > > > > > >>>>>>>>>>> > > > > > > > > >>>>>>>>>>> Willem. > > > > > > > > >>>>>>>>>>> > > > > > > > > >>>>>>>>>>> Christopher Cheng wrote: > > > > > > > > >>>>>>>>>>>> When I call the service as a client, it > > > > > > > > >>>>>>>>>>>> takes 5 seconds to load. > > > > > > > > >>>>>> > > > > > > > > >>>>>> I am > > > > > > > > >>>>>> > > > > > > > > >>>>>>>>>>>> not > > > > > > > > >>>>>>>>>>>> sure whether it takes 5 seconds to create > > > > > > > > >>>>>>>>>>>> the service or it takes > > > > > > > > >>>>>> > > > > > > > > >>>>>> 5 > > > > > > > > >>>>>> > > > > > > > > >>>>>>>>>>>> seconds > > > > > > > > >>>>>>>>>>>> to look up for ciper filters. I am > > > > > > > > >>>>>>>>>>>> wondering if > > > > the > > > > > > > > > > >>>>>>>>>>>> services are cached... > > > > > > > > >>>>>>>>>>>> Feb 29, 2008 2:14:42 PM > > > > org.apache.cxf.service.factory.ReflectionServiceFac > > > > > > > > > > >>>>>>>>>>>>tory Bean buildServiceFromWSDL > > > > > > > > >>>>>>>>>>>> INFO: Creating Service > > > > > > > > >>>>>>>>>>>> { > > > > https://webservices.sabre.com/websvc}OTA_HotelAvai<https://webservic > >es.sabre.com/websvc%7DOTA_HotelAvai> > > <https://webservices.sabre.com/websvc%7DOTA_HotelAvai> > > > > > <https://webservices.sabre.com/websvc%7DOTA_HotelAvai> > > > > > > > > > > > >>>>>>>>>>>>lSer< > > > > https://webservices.sabre.com/websvc%7DOTA_Hote > > > > > > > > > > >>>>>>>>>>>>lAvailSer> > > > > > > > > >>>>>>>>>>>> vice< > > > > https://webservices.sabre.com/websvc%7DOTA_Hot > > > > > > > > > > >>>>>>>>>>>>elAva ilService> > > > > > > > > >>>>>>>>>>>> < > > > > https://webservices.sabre.com/websvc%7DOTA_HotelAv > > > > > > > > > > >>>>>>>>>>>>ailS ervic e> > > > > > > > > >>>>>> > > > > > > > > >>>>>> < > > > > https://webservices.sabre.com/websvc%7DOTA_HotelAvailSer > > > > > > > > > > >>>>>>vice> from > > > > > > > > >>> > > > > > > > > >>> WSDL: > > > > > > > > >>>>>>>>>>>> file:/C:/Java/abacus-webconnect-1.14.0.rc1 > > > > > > > > >>>>>>>> > > > > > > > > >>>>>>>> /wsdl/OTA_HotelAvailLLS1.4.1RQ.wsdl > > > > > > > > >>>>>>>> > > > > > > > > >>>>>>>>>>>> Feb 29, 2008 2:14:47 PM > > > > > > > > >>>>>>>>>>>> org.apache.cxf.transport.https.SSLUtils > > > > > > > > >>>>>>>>>>>> getCiphersuites INFO: The cipher suites > > > > > > > > >>>>>>>>>>>> have not been configured, falling back to > > > > > > > > >>>>>>>> > > > > > > > > >>>>>>>> cipher > > > > > > > > >>>>>>>> > > > > > > > > >>>>>>>>>>>> suite filters. > > > > > > > > >>>>>>>>>>>> Feb 29, 2008 2:14:47 PM > > > > > > > > >>>>>>>>>>>> org.apache.cxf.transport.https.SSLUtils > > > > > > > > >>>>>>>>>>>> getCiphersFromList INFO: The cipher suites > > > > > > > > >>>>>>>>>>>> have been set to > > > > > > > > >>>>>> > > > > > > > > >>>>>> SSL_RSA_WITH_RC4_128_MD5, > > > > > > > > >>>>>> > > > > > > > > >>>>>>>>>>>> SSL_RSA_WITH_RC4_128_SHA, > > > > SSL_RSA_WITH_DES_CBC_SHA, > > > > > > > > > > >>>>>>>>>>>> SSL_DHE_RSA_WITH_DES_CBC_SHA, > > > > > > > > >>>>>>>>>>>> SSL_DHE_DSS_WITH_DES_CBC_SHA, > > > > > > > > >>>>>>>>>>>> SSL_RSA_EXPORT_WITH_RC4_40_MD5, > > > > > > > > >>>>>> > > > > > > > > >>>>>> SSL_RSA_EXPORT_WITH_DES40_CBC_SHA, > > > > > > > > >>>>>> > > > > > > > > >>>>>>>>>>>> SSL_DHE_RSA_EXPORT_WITH_DES40_CBC_SHA, > > > > > > > > >>>>>>>>>>>> SSL_DHE_DSS_EXPORT_WITH_DES40_CBC_SHA, > > > > > > > > >>>>>>>>>>>> SSL_RSA_WITH_NULL_MD5, > > > > > > > > >>>>>>>>>>>> SSL_RSA_WITH_NULL_SHA, > > > > > > > > >>>>>>>>>>>> SSL_DH_anon_WITH_RC4_128_MD5, > > > > > > > > >>>>>>>>>>>> TLS_DH_anon_WITH_AES_128_CBC_SHA, > > > > > > > > >>>>>> > > > > > > > > >>>>>> SSL_DH_anon_WITH_3DES_EDE_CBC_SHA, > > > > > > > > >>>>>> > > > > > > > > >>>>>>>>>>>> SSL_DH_anon_WITH_DES_CBC_SHA, > > > > > > > > >>>>>>>>>>>> SSL_DH_anon_EXPORT_WITH_RC4_40_MD5, > > > > > > > > >>>>>>>>>>>> SSL_DH_anon_EXPORT_WITH_DES40_CBC_SHA, > > > > > > > > >>>>>>>>>>>> TLS_KRB5_WITH_DES_CBC_SHA, > > > > > > > > >>>>>>>>>>>> TLS_KRB5_WITH_DES_CBC_MD5, > > > > > > > > >>>>>>>>>>>> TLS_KRB5_EXPORT_WITH_RC4_40_SHA, > > > > > > > > >>>>>>>>>>>> TLS_KRB5_EXPORT_WITH_RC4_40_MD5, > > > > > > > > >>>>>> > > > > > > > > >>>>>> TLS_KRB5_EXPORT_WITH_DES_CBC_40_SHA, > > > > > > > > >>>>>> > > > > > > > > >>>>>>>>>>>> TLS_KRB5_EXPORT_WITH_DES_CBC_40_MD5. Feb > > > > > > > > >>>>>>>>>>>> 29, > > > > 2008 > > > > > > > > > > >>>>>>>>>>>> 2:14:48 PM > > > > org.apache.cxf.interceptor.LoggingOutInterceptor$LoggingC > > > > > > > > > > >>>>>>allba ckon Close > > > > > > > > >>>>>> > > > > > > > > >>>>>>>>>>>> I have put this in my cxf.xml as in > > > > http://cwiki.apache.org/CXF20DOC/client-http-transp > > > > > > > > > > >>>>>>>>>>>>ort. html , but > > > > > > > > >>>>>> > > > > > > > > >>>>>> it > > > > > > > > >>>>>> > > > > > > > > >>>>>>>>>>>> doesn't > > > > > > > > >>>>>>>>>>>> help... <http:conduit > > > > > > > > >>>>>>>>>>>> name="*.http-conduit"> > > > > > > > > >>>>>>>>>>>> <http:tlsClientParameters > > > > > > > > >>>>>>>>>>>> secureSocketProtocol="SSL"> > > > > > > > > >>>>>>>>>>>> <sec:cipherSuitesFilter> <!-- these filters > > > > ensure > > > > > > > > > > >>>>>>>>>>>> that a ciphersuite with export-suitable or > > > > > > > > >>>>>>>>>>>> null encryption is used, but exclude > > > > > > > > >>>>>>>>>>>> anonymous Diffie-Hellman key change > > > > > > > > >>>>>> > > > > > > > > >>>>>> as > > > > > > > > >>>>>> > > > > > > > > >>>>>>>>>>>> this is vulnerable to > > > > > > > > >>>>>>>>>>>> man-in-the-middle attacks > > > > > > > > >>>>>> > > > > > > > > >>>>>> --> > > > > > > > > >>>>>> > > > > > > > > >>>>>>>>>>>> <sec:include>.*_EXPORT_.*</sec:include> > > > > > > > > >>>>>>>>>>>> > > > > > > > > >>>>>>>>>>>> <sec:include>.*_EXPORT1024_.*</sec:include> > > > > > > > > >>>>>>>>>>>> <sec:include>.*_WITH_DES_.*</sec:include> > > > > > > > > >>>>>>>>>>>> <sec:include>.*_WITH_NULL_.*</sec:include> > > > > > > > > >>>>>>>>>>>> <sec:exclude>.*_DH_anon_.*</sec:exclude> > > > > > > > > >>>>>>>>>>>> </sec:cipherSuitesFilter> > > > > > > > > >>>>>>>>>>>> </http:tlsClientParameters> > > > > > > > > >>>>>>>>>>>> </http:conduit> > > > > > > > > >>>>>>>>>> > > > > > > > > >>>>>>>>>> http://www.nabble.com/file/p15773252/cxf.log > > > > cxf.log > > > > > > > > > > >>> -- > > > > > > > > >>> J. Daniel Kulp > > > > > > > > >>> Principal Engineer, IONA > > > > > > > > >>> [EMAIL PROTECTED] > > > > > > > > >>> http://www.dankulp.com/blog > > > > > > > > > > > > > > -- > > > > > > > J. Daniel Kulp > > > > > > > Principal Engineer, IONA > > > > > > > [EMAIL PROTECTED] > > > > > > > http://www.dankulp.com/blog > > > > > > > > -- > > > > J. Daniel Kulp > > > > Principal Engineer, IONA > > > > [EMAIL PROTECTED] > > > > http://www.dankulp.com/blog -- J. Daniel Kulp Principal Engineer, IONA [EMAIL PROTECTED] http://www.dankulp.com/blog
