[undertow-dev] Resizing undertow thread pool size dynamically
Stuart Douglas
sdouglas at redhat.com
Tue Dec 8 20:33:39 EST 2015
Fixed: https://issues.jboss.org/browse/UNDERTOW-593
Stuart
----- Original Message -----
> From: "Mohammed ElGhaouat" <melghaouat at gmail.com>
> To: "Stuart Douglas" <sdouglas at redhat.com>, "Jason Greene" <jason.greene at redhat.com>, undertow-dev at lists.jboss.org
> Sent: Wednesday, 9 December, 2015 3:06:57 AM
> Subject: Re: [undertow-dev] Resizing undertow thread pool size dynamically
>
> Hello,
>
> this doesn't work for me. I dug a little bit in undertow code and it seems
> to me that the *deploymentInfo.executor *we set in the ServletExtension is
> not used by *DeploymentImpl *class, the *DeploymentImpl.getExecutor* returns
> the field *executor*
>
> @Override
> public Executor getExecutor() {
> return executor;
> }
>
> while in the constructor we have
>
> public DeploymentImpl(DeploymentManager deploymentManager, final
> DeploymentInfo deploymentInfo, ServletContainer servletContainer) {
> ............
> * this.executor = deploymentInfo.getExecutor();*
> ............
> }
>
> when i changed the getExecutor method to
>
>
> @Override
> public Executor getExecutor() {
> return deploymentInfo.getExecutor();
> }
>
> I got my executor that I set in the ServletExtension used by Wildfly.
>
> Is this explanation seems right to you ?
>
>
> Thanks
>
> Mohammed.
>
> On Mon, Oct 19, 2015 at 10:19 PM, Stuart Douglas <sdouglas at redhat.com>
> wrote:
>
> >
> >
> > ----- Original Message -----
> > > From: "Mohammed ElGhaouat" <melghaouat at gmail.com>
> > > To: "Stuart Douglas" <sdouglas at redhat.com>
> > > Cc: "Jason Greene" <jason.greene at redhat.com>,
> > undertow-dev at lists.jboss.org
> > > Sent: Tuesday, 20 October, 2015 2:22:58 AM
> > > Subject: Re: [undertow-dev] Resizing undertow thread pool size
> > dynamically
> > >
> > > Is this valid for Wildfly? (I am using 8.2.1)
> >
> > Yes, this is valid for all Servlet deployments, in both Wildfly and
> > embedded Undertow.
> >
> > Wildfly will still create its worker thread pool, but it should not get
> > used for this deployment.
> >
> > Stuart
> >
> > >
> > > From the previous replies, I understood that the ServletExtension could
> > be
> > > used for embedded Undertow not for Wildfly. Anyway, I gave it a try, but
> > > still having Wildfly using task-max-threads and io-threads parameters
> > from
> > > (subsystem=io/worker=default/).
> > >
> > > To test this solution, I created a simple wepapp with one simple Servlet
> > > and my ServletExtension.handleDeployment is executed by the Wildfly.
> > >
> > >
> > > Thanks
> > >
> > > On Sat, Oct 17, 2015 at 3:01 PM, Stuart Douglas <sdouglas at redhat.com>
> > wrote:
> > >
> > > > Like I said, you can use any thread pool you want if you modify it
> > using a
> > > > ServletExtension.
> > > >
> > > > Code looks like:
> > > >
> > > >
> > > > public class MyExtension implements ServletExtension {
> > > >
> > > > @Override
> > > > public void handleDeployment(DeploymentInfo deploymentInfo,
> > > > ServletContext servletContext) {
> > > > Executor myThreadPool = {my thread pool};
> > > > deploymentInfo.setExecutor(myThreadPool);
> > > > }
> > > > }
> > > >
> > > > Then add a META-INF/services entry for the extension.
> > > >
> > > > Stuart
> > > >
> > > > ----- Original Message -----
> > > > > From: "Mohammed ElGhaouat" <melghaouat at gmail.com>
> > > > > To: "Stuart Douglas" <sdouglas at redhat.com>
> > > > > Cc: "Jason Greene" <jason.greene at redhat.com>,
> > > > undertow-dev at lists.jboss.org
> > > > > Sent: Wednesday, 14 October, 2015 12:22:20 PM
> > > > > Subject: Re: [undertow-dev] Resizing undertow thread pool size
> > > > dynamically
> > > > >
> > > > > Hello,
> > > > >
> > > > > I would like to share with you some more details about our
> > situation. We
> > > > > are using some big machines that are shared by many software(many
> > Wildfly
> > > > > instances, Databases, ERPs ..) If i don't set pools sizes i end up
> > with
> > > > big
> > > > > pools as the default size is dependent on the number of CPU cores
> > and out
> > > > > system administrator is complaining about the OS spending time
> > checking
> > > > if
> > > > > the threads have something to do and this impact the other softwares
> > > > > installed on the same machine. If I set a small pool size which could
> > > > > sufficient in the 90% of time, i am afraid that Wildfly couldn't
> > handle
> > > > > the 10% of time when the applications are used by a large number of
> > user.
> > > > >
> > > > > Is there any workaround or are you planning to let the user to set a
> > > > > specific ThreadPoolExecutor ? so we can evict idle threads
> > > > >
> > > > >
> > > > > Thanks,
> > > > >
> > > > > Mohammed.
> > > > >
> > > > >
> > > > > On Wed, Aug 12, 2015 at 11:03 AM, Mohammed ElGhaouat <
> > > > melghaouat at gmail.com>
> > > > > wrote:
> > > > >
> > > > > > I am using Wildfly.
> > > > > >
> > > > > > Thanks,
> > > > > >
> > > > > > Mohammed.
> > > > > >
> > > > > > On Wed, Aug 12, 2015 at 10:55 AM, Stuart Douglas <
> > sdouglas at redhat.com>
> > > > > > wrote:
> > > > > >
> > > > > >> Are you using Wildfly or embedded Undertow?
> > > > > >>
> > > > > >> If it is the later you can just use
> > > > > >> io.undertow.servlet.api.DeploymentInfo#setExecutor to use whatever
> > > > > >> executor
> > > > > >> implementation you want.
> > > > > >>
> > > > > >> The reason why most executors don't reduce the number is because
> > > > there is
> > > > > >> generally very little point, a parked thread is generally very
> > cheap,
> > > > > >> while
> > > > > >> creating new threads is relatively expensive.
> > > > > >>
> > > > > >> Stuart
> > > > > >>
> > > > > >> ----- Original Message -----
> > > > > >> > From: "Mohammed ElGhaouat" <melghaouat at gmail.com>
> > > > > >> > To: "Jason Greene" <jason.greene at redhat.com>
> > > > > >> > Cc: undertow-dev at lists.jboss.org
> > > > > >> > Sent: Wednesday, 12 August, 2015 6:19:11 PM
> > > > > >> > Subject: Re: [undertow-dev] Resizing undertow thread pool size
> > > > > >> dynamically
> > > > > >> >
> > > > > >> > We are using the servlet API and I am referring to worker pool.
> > > > Simply
> > > > > >> we
> > > > > >> > don't want keeping bunch of idle threads in the JVM consuming
> > some
> > > > > >> resources
> > > > > >> > without doing any thing useful.
> > > > > >> >
> > > > > >> > So with the bounded queue executor, when the value of the
> > > > > >> task-max-threads
> > > > > >> > parameter is reached, the number of threads in the worker pool
> > > > couldn't
> > > > > >> be
> > > > > >> > decreased ?
> > > > > >> >
> > > > > >> > Thank you.
> > > > > >> >
> > > > > >> >
> > > > > >> > Mohammed.
> > > > > >> >
> > > > > >> > On Tue, Aug 11, 2015 at 9:50 PM, Jason Greene <
> > > > jason.greene at redhat.com
> > > > > >> >
> > > > > >> > wrote:
> > > > > >> >
> > > > > >> >
> > > > > >> >
> > > > > >> > > On Aug 11, 2015, at 4:42 AM, Mohammed ElGhaouat <
> > > > > >> melghaouat at gmail.com >
> > > > > >> > > wrote:
> > > > > >> > >
> > > > > >> > >
> > > > > >> > > Hi,
> > > > > >> > >
> > > > > >> > > I would like to know if there is a way to make undertow
> > reducing
> > > > the
> > > > > >> size
> > > > > >> > > of the thread pool when the server is less loaded. Is there
> > any
> > > > > >> > > parameter(or other way) that make an idle thread die after
> > some
> > > > > >> inactivity
> > > > > >> > > time ?
> > > > > >> >
> > > > > >> >
> > > > > >> > Are you referring to the worker pool or the I/O pool? The I/O
> > pool
> > > > is
> > > > > >> special
> > > > > >> > and is fixed. The worker pool currently uses a JDK
> > > > ThreadPoolExecutor
> > > > > >> with
> > > > > >> > an unbounded queue which is a behavior pattern typically desired
> > > > for web
> > > > > >> > servers. That’s not pluggable at the moment, but it could be
> > > > possible.
> > > > > >> >
> > > > > >> > If you are using the HttpHandler APIs, there is a method on
> > > > > >> > HttpServerDispatch that allows you to use your own custom
> > executor
> > > > for
> > > > > >> > blocking tasks (which would allow you to tune the default worker
> > > > task
> > > > > >> pool
> > > > > >> > very small). If you are using servlet APIs though that uses the
> > > > standard
> > > > > >> > pools we provide.
> > > > > >> >
> > > > > >> > Is there a particular reason you want to kill idle threads?
> > Threads
> > > > are
> > > > > >> cheap
> > > > > >> > unless you are storing massive amounts of thread local data.
> > > > > >> >
> > > > > >> > --
> > > > > >> > Jason T. Greene
> > > > > >> > WildFly Lead / JBoss EAP Platform Architect
> > > > > >> > JBoss, a division of Red Hat
> > > > > >> >
> > > > > >> >
> > > > > >> >
> > > > > >> > _______________________________________________
> > > > > >> > undertow-dev mailing list
> > > > > >> > undertow-dev at lists.jboss.org
> > > > > >> > https://lists.jboss.org/mailman/listinfo/undertow-dev
> > > > > >>
> > > > > >
> > > > > >
> > > > >
> > > >
> > >
> >
>
More information about the undertow-dev
mailing list