Also, anything you can speed up with Jandex might also be very useful in 2.0! So let's definitely take a good look :)<br><div class="gmail_extra"><br><br><div class="gmail_quote">On Mon, Nov 5, 2012 at 11:44 AM, Lincoln Baxter, III <span dir="ltr"><<a href="mailto:lincolnbaxter@gmail.com" target="_blank">lincolnbaxter@gmail.com</a>></span> wrote:<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">Great!<br><br>Please send a pull request for the ObserverCapture issue! Thanks!<br><br>Good plan with the bootstrap API, and are you sure that Jandex would require changes in Weld-API? Look at what has been done in 2.0 with the <a href="https://github.com/forge/core/blob/2.0/plugin-container/src/main/java/org/jboss/forge/container/modules/ModularWeldDeployment.java" target="_blank">https://github.com/forge/core/blob/2.0/plugin-container/src/main/java/org/jboss/forge/container/modules/ModularWeldDeployment.java</a><br>
<br>What APIs did you have to change?<br><br>Looking forward to seeing your changes!<span class="HOEnZb"><font color="#888888"><br><br>~Lincoln</font></span><div class="HOEnZb"><div class="h5"><br><div class="gmail_extra">
<br><br><div class="gmail_quote">On Mon, Nov 5, 2012 at 2:10 AM, Thomas Frühbeck <span dir="ltr"><<a href="mailto:fruehbeck@aon.at" target="_blank">fruehbeck@aon.at</a>></span> wrote:<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
<div bgcolor="#FFFFFF" text="#000000">
<div>The first issue
(ObserverCaptureExtension.scan) is the easiest and most improving,
it only needs your approval that it's really up to the point -
it's just a few lines in fact.<br>
I will provide a proposal in my clone at github and notify you.<br>
<br>
The Bootstrap redesign idea I will provide in a seperate branch in
same repo.<br>
The Jandex-idea would need also changes in the Weld-API, so they
are somewhat theoretical :-/<br>
<br>
Regards,<br>
Thomas<br>
<br>
Am 05.11.2012 01:27, schrieb Lincoln Baxter, III:<br>
</div><div><div>
<blockquote type="cite">I think performance improvements in Forge 1.x are very
important even now!<br>
<br>
It's going to be around for a very long time and while we are
working on Forge 2.0 now, the kinds of improvements you are
talking about, Thomas, are very significant!<br>
<br>
How can we help? Do you think you could put this code into a
branch with a single commit (or just one for each change) so that
we can see just how much you had to change?<br>
<br>
The problem with the weld container is the Hibernate Tools plugin.
If that one works, and the Openshift and AS7 plugins work, then
it's safe to assume that it's working fine. It was a hack to solve
a classloading problem, but if we can fix it now, that's great!<br>
<br>
~Lincoln<br>
<div class="gmail_extra"><br>
<br>
<div class="gmail_quote">On Sat, Nov 3, 2012 at 3:27 PM, George
Gastaldi <span dir="ltr"><<a href="mailto:ggastald@redhat.com" target="_blank">ggastald@redhat.com</a>></span>
wrote:<br>
<blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">Hi Thomas,<br>
<br>
Nice to know that you've found some improvements for Forge
1.x startup time. However, we are rewriting Forge in 2.x
from the ground up and we're focusing on performance and
flexibility in first place.<br>
<br>
We encourage you to try the 2.0 branch and help us on
finding some possible performance problems if you feel so.<br>
<br>
As for now, we are avoiding any major changes and updates to
1.x.<br>
<br>
Best Regards,<br>
<br>
George Gastaldi<br>
<br>
Em 03/11/2012, às 15:41, Thomas Frühbeck <<a href="mailto:fruehbeck@aon.at" target="_blank">fruehbeck@aon.at</a>>
escreveu:<br>
<div>
<div><br>
> Hi,<br>
><br>
> following the discussions about boot-up performance
of Forge I made some<br>
> investigations using a profiler.<br>
> After testing a lot of different scenarios I would
like to share my<br>
> results and ask for your opinion on it.<br>
><br>
>
*************************************************************************************<br>
> Issue 1.: hot spot inv.<br>
>
org.jboss.forge.bus.cdi.ObserverCaptureExtension.scan<br>
> profiling shows that 37% CPU time are used in
firing<br>
> ProcessAnnotatedType events, which are AFAICS only
used to filter old<br>
> event types<br>
> "37,2% - 13.442 ms - 1.866 inv.<br>
>
org.jboss.weld.bootstrap.events.ProcessAnnotatedTypeImpl.fire"<br>
><br>
> By introducing a preparatory check in
ObserverCaptureExtension like<br>
>> if (isMethodsAnnotated(originalType,
Observes.class)) {<br>
>> ....<br>
>> }<br>
><br>
> and something like:<br>
>> private boolean
isMethodsAnnotated(AnnotatedType<?> type,
Class<?<br>
> extends Annotation> annotation) {<br>
>> if
(type.getClass().isAnnotationPresent(annotation))<br>
>> return true;<br>
>> for (Method m :
type.getClass().getMethods()) {<br>
>> if (m.isAnnotationPresent(annotation))<br>
>> return true;<br>
>> for (Annotation[] arr :
m.getParameterAnnotations()) {<br>
>> for (Annotation a : arr) {<br>
>> if (a.equals(annotation))<br>
>> return true;<br>
>> }<br>
>> }<br>
>> }<br>
>> return false;<br>
>> }<br>
><br>
> we could reduce startup by 20%<br>
> "13,8% - 2.438 ms - 1.798 inv.<br>
>
org.jboss.weld.bootstrap.events.ProcessAnnotatedTypeImpl.fire<br>
><br>
> Test: call Forge with commands "list-plugins" and
"exit", like ~> time<br>
> ((echo forge list-plugins; echo exit;)| bin/forge)<br>
><br>
> Before optimization:<br>
> ~/forge> time ((echo forge list-plugins
&& echo exit)| bin/forge )<br>
> _____<br>
> | ___|__ _ __ __ _ ___<br>
> | |_ / _ \| `__/ _` |/ _ \ \\<br>
> | _| (_) | | | (_| | __/ //<br>
> |_| \___/|_| \__, |\___|<br>
> |___/<br>
><br>
> [no project] forge-distribution-1.1.1-SNAPSHOT $
forge list-plugins<br>
>
org.richfaces.forge.richfaces-forge-plugin:1.0.5.Final:1.0.0-SNAPSHOT-b1a6ebed-462f-40e1-ac08-82f927489301<br>
>
com.ocpsoft.forge.prettyfaces-plugin:1.0.2.Final:1.0.0-SNAPSHOT-c6b071ad-a6df-4112-8fff-82ecccd003fb<br>
>
org.jboss.hibernate.forge.hibernate-tools-plugin:1.0.5.Final:1.0.0-SNAPSHOT-afb1da29-a99f-4dd7-b020-5f3ba6073cde<br>
>
org.arquillian.forge.arquillian-plugin:1.0.3-SNAPSHOT:1.0.0-SNAPSHOT-42907982-d02e-4a1d-8ea5-8e66c8013fde<br>
> [no project] forge-distribution-1.1.1-SNAPSHOT $
exit<br>
><br>
><br>
> real 0m9.721s<br>
> user 0m26.533s<br>
> sys 0m0.433s<br>
><br>
> After optimization:<br>
> ~forge1.1.1-weld1.2> time ((echo forge
list-plugins; echo exit;)| bin/forge)<br>
> Using Forge at
/raid/home/thomas/java/forge1.1.1-weld1.2<br>
> _____<br>
> | ___|__ _ __ __ _ ___<br>
> | |_ / _ \| `__/ _` |/ _ \ \\<br>
> | _| (_) | | | (_| | __/ //<br>
> |_| \___/|_| \__, |\___|<br>
> |___/<br>
><br>
> [no project] forge1.1.1-weld1.2 $ forge
list-plugins<br>
>
org.richfaces.forge.richfaces-forge-plugin:1.0.5.Final:1.0.0-SNAPSHOT-b1a6ebed-462f-40e1-ac08-82f927489301<br>
>
com.ocpsoft.forge.prettyfaces-plugin:1.0.2.Final:1.0.0-SNAPSHOT-c6b071ad-a6df-4112-8fff-82ecccd003fb<br>
>
org.jboss.hibernate.forge.hibernate-tools-plugin:1.0.5.Final:1.0.0-SNAPSHOT-afb1da29-a99f-4dd7-b020-5f3ba6073cde<br>
>
org.arquillian.forge.arquillian-plugin:1.0.3-SNAPSHOT:1.0.0-SNAPSHOT-42907982-d02e-4a1d-8ea5-8e66c8013fde<br>
> [no project] forge1.1.1-weld1.2 $ exit<br>
><br>
><br>
> real 0m6.016s<br>
> user 0m14.482s<br>
> sys 0m0.449s<br>
><br>
><br>
>
*************************************************************************************<br>
> Issue 2.: double starting of Weld container<br>
> I admit that I do not fully understand, what I did
and if this really<br>
> provides the full solution, but it worked and
produced a fully<br>
> functional Forge instance.<br>
><br>
> I tried to restructure the Bootstrap logic a bit:<br>
><br>
> initializeClassloader();<br>
> WeldContainer container =
weld.initialize();<br>
> manager =
container.getBeanManager();<br>
><br>
> try<br>
> {<br>
> loadPlugins();<br>
> weld.reInitialize();<br>
><br>
> where initializeClassloader already uses the
CompositeClassloader:<br>
><br>
> private static void initializeClassloader() {<br>
> ModuleLoader moduleLoader =
Module.getBootModuleLoader();<br>
><br>
> CompositeClassLoader composite = new
CompositeClassLoader();<br>
>
composite.add(Module.forClassLoader(Bootstrap.class.getClassLoader(),<br>
> true).getClassLoader());<br>
><br>
>
Thread.currentThread().setContextClassLoader(composite);<br>
> }<br>
><br>
> and loadPlugins simply adds more classloaders:<br>
><br>
> CompositeClassLoader composite =<br>
>
(CompositeClassLoader)Thread.currentThread().getContextClassLoader();<br>
> for (PluginEntry plugin : toLoad)<br>
> {<br>
> Module module =<br>
>
moduleLoader.loadModule(ModuleIdentifier.fromString(plugin.toModuleId()));<br>
>
composite.add(module.getClassLoader());<br>
><br>
><br>
> This way the classes are already available in the
correct classloader<br>
> for the later ModularWeld.reInitialize():<br>
><br>
> public void reInitialize() {<br>
> bootstrap.revisit();<br>
> bootstrap.validateBeans();<br>
> bootstrap.endInitialization();<br>
> }<br>
><br>
> Where Bootstrap.revisit simply revisits the
BeanDeployments which have<br>
> been newly found:<br>
><br>
> private static class DeploymentVisitor {<br>
> .....<br>
> public Map<BeanDeploymentArchive,
BeanDeployment> revisit() {<br>
> for (BeanDeploymentArchive archvive :<br>
> deployment.getBeanDeploymentArchives()) {<br>
> if<br>
>
(!managerAwareBeanDeploymentArchives.containsKey(archvive))
{<br>
> visit(archvive,
managerAwareBeanDeploymentArchives,<br>
> new HashSet<BeanDeploymentArchive>(), true);<br>
> BeanDeployment bd =<br>
> managerAwareBeanDeploymentArchives.get(archvive);<br>
> bd.createBeans(environment);<br>
> bd.deployBeans(environment);<br>
> }<br>
> }<br>
> return
managerAwareBeanDeploymentArchives;<br>
> }<br>
><br>
><br>
>
*************************************************************************************<br>
> Issue 3.: invocation of potentially expensive
operations on<br>
> non-annotated or annotation-relevant classes<br>
> The current BeanDeployer simply invokes
ProcessAnnotatedType on any<br>
> class, w/o checking, if this class is used by any
annotation anywhere.<br>
> By jandexing all jars and assembling the dispersed
Jandex indices during<br>
> URLScanning we could reduce the expensive WeldClass
conversion.<br>
><br>
> Original:<br>
> 11.108 ms - 2.024 inv.
org.jboss.weld.bootstrap.BeanDeployer.addClass<br>
> Jandex-aware:<br>
> 4.721 ms - 990 inv.
org.jboss.weld.bootstrap.BeanDeployer.addClass<br>
><br>
> Possible solution:<br>
> - 1. URLHandler scans Jandex file locations<br>
><br>
> protected void addToDiscovered(String name, URL
url) {<br>
> ...<br>
> } else if (name.endsWith(JANDEX_IDX)) {<br>
> discoveredJandexIndexUrls.add(url);<br>
> }<br>
> }<br>
><br>
> - 2. URLScanner assembles all found Jandex
indices:<br>
><br>
> List<Index> jandexIndexes = new
ArrayList<Index>();<br>
> for (URL jandexUrl :
handler.getDiscoveredJandexIndexUrls()) {<br>
> jandexIndexes.add(new<br>
> IndexReader(jandexUrl.openStream()).read());<br>
> }<br>
><br>
> - 3. BeanDeployer does expensive WeldClass
loading only if class is<br>
> mentioned in the Jandex indices<br>
> - 3.a. first collapse the Jandex-indices to a
simple list of class<br>
> names:<br>
> for (Index jandexIndex : jandexIndexes) {<br>
> for (List<AnnotationInstance>
aiList :<br>
> jandexIndex.getAnnotations().values()) {<br>
> for (AnnotationInstance ai :
aiList) {<br>
> AnnotationTarget at =
ai.target();<br>
> if (at instanceof ClassInfo) {<br>
>
jandexAnnotated.add(((ClassInfo)at).name().toString());<br>
> } else if (at instanceof
MethodInfo) {<br>
>
jandexAnnotated.add(((MethodInfo)at).declaringClass().name().toString());<br>
> } else if (at instanceof
MethodParameterInfo) {<br>
>
jandexAnnotated.add(((MethodParameterInfo)at).method().declaringClass().name().toString());<br>
> } else if (at instanceof
FieldInfo) {<br>
>
jandexAnnotated.add(((FieldInfo)at).declaringClass().name().toString());<br>
> }<br>
> }<br>
> }<br>
> }<br>
><br>
> - 3.b. query the aggregated list of
Jandex-aware classes<br>
><br>
> public BeanDeployer addClass(String className)
{<br>
> Class<?> clazz =
loadClass(className);<br>
> if (clazz != null &&
isBeanCandidate(clazz) &&<br>
> jandexContains(className)) {<br>
> WeldClass<?> weldClass =
loadWeldClass(clazz);<br>
><br>
><br>
> Yet unsolved issue is the question, wheter an index
is found in the<br>
> corresponding Archive or not, so not to enforce
Jandexing _all_ jars for<br>
> Forge.<br>
> Jandexing is very easy though: jandex -m
<jar><br>
><br>
> The abovementioned solutions have been implemented
and produced a<br>
> functional and correctly working Forge instance.<br>
> AFAICS the solution for issue 2 could be a means to
solve the reloading<br>
> of plugins too, but I didn't implement and verify
this.<br>
><br>
> The sources are not available online, because the
current implementation<br>
> is highly fragile and not safe for simple usage -
too many changes in<br>
> APIs, dependencies, etc.<br>
><br>
> Comments welcome!<br>
><br>
> Regards,<br>
> Thomas<br>
><br>
> _______________________________________________<br>
> forge-dev mailing list<br>
> <a href="mailto:forge-dev@lists.jboss.org" target="_blank">forge-dev@lists.jboss.org</a><br>
> <a href="https://lists.jboss.org/mailman/listinfo/forge-dev" target="_blank">https://lists.jboss.org/mailman/listinfo/forge-dev</a><br>
<br>
_______________________________________________<br>
forge-dev mailing list<br>
<a href="mailto:forge-dev@lists.jboss.org" target="_blank">forge-dev@lists.jboss.org</a><br>
<a href="https://lists.jboss.org/mailman/listinfo/forge-dev" target="_blank">https://lists.jboss.org/mailman/listinfo/forge-dev</a><br>
</div>
</div>
</blockquote>
</div>
<br>
<br clear="all">
<br>
-- <br>
Lincoln Baxter, III<br>
<a href="http://ocpsoft.org" target="_blank">http://ocpsoft.org</a><br>
"Simpler is better."<br>
</div>
<br>
<fieldset></fieldset>
<br>
<pre>_______________________________________________
forge-dev mailing list
<a href="mailto:forge-dev@lists.jboss.org" target="_blank">forge-dev@lists.jboss.org</a>
<a href="https://lists.jboss.org/mailman/listinfo/forge-dev" target="_blank">https://lists.jboss.org/mailman/listinfo/forge-dev</a>
</pre>
</blockquote>
<br>
</div></div></div>
<br>_______________________________________________<br>
forge-dev mailing list<br>
<a href="mailto:forge-dev@lists.jboss.org" target="_blank">forge-dev@lists.jboss.org</a><br>
<a href="https://lists.jboss.org/mailman/listinfo/forge-dev" target="_blank">https://lists.jboss.org/mailman/listinfo/forge-dev</a><br>
<br></blockquote></div><br><br clear="all"><br>-- <br>Lincoln Baxter, III<br><a href="http://ocpsoft.org" target="_blank">http://ocpsoft.org</a><br>"Simpler is better."<br>
</div>
</div></div></blockquote></div><br><br clear="all"><br>-- <br>Lincoln Baxter, III<br><a href="http://ocpsoft.org" target="_blank">http://ocpsoft.org</a><br>"Simpler is better."<br>
</div>