Lock contention in Groovy Shell interpreter under high load

Previous Topic Next Topic
 
classic Classic list List threaded Threaded
4 messages Options
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Lock contention in Groovy Shell interpreter under high load

Aman Jain
Hello,

We are evaluating GroovyShell interpreter (v2.4) in our application for dynamically executing standard Java syntax. 

Earlier, we were using Java BeanShell Interpreter for it, but it has an issue under high load which prompted us to look for an alternative such as Groovy.

We are seeing thread lock contention when multiple threads execute the below sample code simultaneously. We have multiple threads in the blocked state which is degrading application performance.

My questions are:

  1. Has anyone experienced such problem? If yes, can you please suggest any possible resolution for it? I googled around for this problem and few blogs suggested to cache groovy Script object. Is Script object thread-safe? Moreover, I need to bind variables to Scriptobject (via groovy Binding object, which is different for each execution in different threads), so I don't think caching groovy Script object is a viable option.
  2. Are there any best practices that I need to follow while using Groovy Shell interpreter with Java?
PFA, sample code, and blocked thread call stack.

Let me know if you need more information. Thanks in advance for your time.

Regards,
Aman


GS_BlockedThread.txt (4K) Download Attachment
SampleCode_GS.txt (516 bytes) Download Attachment
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Lock contention in Groovy Shell interpreter under high load

Jochen Theodorou
Just for reference...this seems to be
https://stackoverflow.com/questions/45578426/lock-contention-in-groovy-shell-interpreter-under-high-load

more inline

On 09.08.2017 18:58, Aman Jain wrote:

> Hello,
>
> We are evaluating GroovyShell
> <http://groovy-lang.org/download.html> interpreter (v2.4) in our
> application for dynamically executing standard Java syntax.
>
> Earlier, we were using Java BeanShell Interpreter for it, but it has an
> issue
> <https://stackoverflow.com/questions/45450947/lock-contention-in-java-beanshell-interpreter-under-high-load/45451768#45451768> under
> high load which prompted us to look for an alternative such as Groovy.
>
> We are seeing thread lock contention when multiple threads execute the
> below sample code simultaneously. We have multiple threads in the
> blocked state which is degrading application performance.

I am missing the counter part, the one that owns the lock. but looking at

>     java.lang.Thread.State: BLOCKED (on object monitor)
>     at java.lang.ClassLoader.loadClass(ClassLoader.java:404)
> - waiting to lock <0x00000007bc425e40> (a java.lang.Object)
> at sun.misc.Launcher$AppClassLoader.loadClass(Launcher.java:331)
> at java.lang.ClassLoader.loadClass(ClassLoader.java:411)

my strong assumption would be, that the counter part is also in
java.lang.ClassLoader. My guess would be, that we are currently trying
to load a class, probably from some other thread. If you have code like

> static String script = "int y = x * x; System.out.println(\"** value of y ** :: \" + y ); ";
>
> GroovyShell gs = new GroovyShell();
> Script evalScript = gs.parse("void evalMethod() {" + script + "}");
> // bind variables
> Binding binding = new Binding();
> binding.setVariable("x", 5);
> evalScript.setBinding(binding);
> // invoke eval method
> evalScript.invokeMethod("evalMethod", null);

you would with each thread execution do a new compilation, which is
where the majority of those class look-ups is coming from. The script
form example contains System, which will be looked-up in maybe 10
variants before the compiler is happy. That makes 9 false look-ups.
Multiply this with a hundred threads and you get 900 look-ups for
classes, that actually do not exist. So the goal would be reuse. My
suggestion would be then:

for all threads:

> static String script = "int y = x * x; System.out.println(\"** value of y ** :: \" + y ); ";
>
> GroovyShell gs = new GroovyShell();
> Script evalScript = gs.parse("void evalMethod() {" + script + "}");

in each thread:

// bind variables
Binding binding = new Binding();
binding.setVariable("x", 5);
Script threadEvalScript =
InvokerHelper.createScript(evalScript.getClas(), binding);
// invoke eval method
threadEvalScript.invokeMethod("evalMethod", null);

Even better would be of course to make a pool of script instances and
use them from there


> My questions are:
>
>  1. Has anyone experienced such problem? If yes, can you please suggest
>     any possible resolution for it? I googled around for this problem
>     and few blogs suggested to cache groovy |Script| object. Is
>     |Script| object thread-safe? Moreover, I need to bind variables to
>     |Script|object (via groovy |Binding| object, which is different for
>     each execution in different threads), so I don't think caching
>     groovy |Script| object is a viable option.

You have to understand that there is the script and the binding. Ideally
the script itself stores all the state in the binding. Then you need one
combination of script and binding per thread that is concurrently
running. How you can do that I tried to show above already a bit.

btw, Binding is not thread-safe, but nothing prevents you from
implementing your own Binding, that is. But given your statement, that
the Binding is different each time I do not think there is a problem
with this.

>  2. Are there any best practices that I need to follow while using
>     Groovy Shell interpreter with Java?

Biggest problem normally is the delayed clean-up of the generated
classes and meta classes, but this should not concern you here, once you
managed to reuse the script class.

bye Jochen
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Lock contention in Groovy Shell interpreter under high load

Aman Jain
Thank you, Jochen! It is very helpful.

So, what I understand is that I can cache either the Script (evalScript) or Script Class (evalScript.getClass()) instances in my application code and reuse them.

I tried by caching Script Class (evalScript.getClass()) instance and reuse it in my code and the results are much better. For each thread, I am retrieving the Script Class from my cache and generating its new instance by calling scriptClass.newInstance()

Does above caching of Script Class (evalScript.getClass()) instance in my application code can cause any memory leak or delayed clean-up of the generated classes and meta classes (as you mentioned in your last point)?

Thanks,
Aman

On Thu, Aug 10, 2017 at 5:53 AM, Jochen Theodorou <[hidden email]> wrote:
Just for reference...this seems to be https://stackoverflow.com/questions/45578426/lock-contention-in-groovy-shell-interpreter-under-high-load

more inline

On 09.08.2017 18:58, Aman Jain wrote:
Hello,

We are evaluating GroovyShell <http://groovy-lang.org/download.html> interpreter (v2.4) in our application for dynamically executing standard Java syntax.

Earlier, we were using Java BeanShell Interpreter for it, but it has an issue <https://stackoverflow.com/questions/45450947/lock-contention-in-java-beanshell-interpreter-under-high-load/45451768#45451768> under high load which prompted us to look for an alternative such as Groovy.

We are seeing thread lock contention when multiple threads execute the below sample code simultaneously. We have multiple threads in the blocked state which is degrading application performance.

I am missing the counter part, the one that owns the lock. but looking at

    java.lang.Thread.State: BLOCKED (on object monitor)
    at java.lang.ClassLoader.loadClass(ClassLoader.java:404)
        - waiting to lock <0x00000007bc425e40> (a java.lang.Object)
        at sun.misc.Launcher$AppClassLoader.loadClass(Launcher.java:331)
        at java.lang.ClassLoader.loadClass(ClassLoader.java:411)

my strong assumption would be, that the counter part is also in java.lang.ClassLoader. My guess would be, that we are currently trying to load a class, probably from some other thread. If you have code like

static String script = "int y = x * x; System.out.println(\"** value of y ** :: \" + y ); ";

GroovyShell gs = new GroovyShell();
Script evalScript = gs.parse("void evalMethod() {" + script + "}");
// bind variables
Binding binding = new Binding();
binding.setVariable("x", 5);
evalScript.setBinding(binding);
// invoke eval method
evalScript.invokeMethod("evalMethod", null);

you would with each thread execution do a new compilation, which is where the majority of those class look-ups is coming from. The script form example contains System, which will be looked-up in maybe 10 variants before the compiler is happy. That makes 9 false look-ups. Multiply this with a hundred threads and you get 900 look-ups for classes, that actually do not exist. So the goal would be reuse. My suggestion would be then:

for all threads:

static String script = "int y = x * x; System.out.println(\"** value of y ** :: \" + y ); ";

GroovyShell gs = new GroovyShell();
Script evalScript = gs.parse("void evalMethod() {" + script + "}");

in each thread:

// bind variables
Binding binding = new Binding();
binding.setVariable("x", 5);
Script threadEvalScript = InvokerHelper.createScript(evalScript.getClas(), binding);
// invoke eval method
threadEvalScript.invokeMethod("evalMethod", null);

Even better would be of course to make a pool of script instances and use them from there


My questions are:

 1. Has anyone experienced such problem? If yes, can you please suggest
    any possible resolution for it? I googled around for this problem
    and few blogs suggested to cache groovy |Script| object. Is
    |Script| object thread-safe? Moreover, I need to bind variables to
    |Script|object (via groovy |Binding| object, which is different for
    each execution in different threads), so I don't think caching
    groovy |Script| object is a viable option.

You have to understand that there is the script and the binding. Ideally the script itself stores all the state in the binding. Then you need one combination of script and binding per thread that is concurrently running. How you can do that I tried to show above already a bit.

btw, Binding is not thread-safe, but nothing prevents you from implementing your own Binding, that is. But given your statement, that the Binding is different each time I do not think there is a problem with this.

 2. Are there any best practices that I need to follow while using
    Groovy Shell interpreter with Java?

Biggest problem normally is the delayed clean-up of the generated classes and meta classes, but this should not concern you here, once you managed to reuse the script class.

bye Jochen

Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: Lock contention in Groovy Shell interpreter under high load

Jochen Theodorou
On 10.08.2017 20:40, Aman Jain wrote:
> Thank you, Jochen! It is very helpful.
>
> So, what I understand is that I can cache either the Script (evalScript)
> *or* Script Class (evalScript.getClass()) instances in my application
> code and reuse them.

well, since you put the "or" in bold I tend to think you think this is
an exclusive choice. You can use both if you want. Depends really on if
it makes sense for you or not. You will always have one class, you will
always need x instances for x parallel operations. If you cache the
instance, you indirectly cache the class. The only factor then is if I
make the instances each time fresh, or produce only x. Producing only x
instances will have a little bit better results, but then you have to
have a clean up strategy for the pool as well. Instance and class go
together... so not really either-or... but of course nothing is really
for free

> I tried by caching Script Class (evalScript.getClass()) instance and
> reuse it in my code and the results are much better. For each thread, I
> am retrieving the Script Class from my cache and generating its new
> instance by calling scriptClass.newInstance()

ok

> Does above caching of Script Class (evalScript.getClass()) instance in
> my application code can cause any memory leak or delayed clean-up of the
> generated classes and meta classes (as you mentioned in your last point)?

nothing to worry about. The problem starts when you have potentially 100
thousands of such classes (*not* instances)

bye Jochen
Loading...