Could you test :

https://github.com/netty/netty/pull/5569

?

> Am 22.07.2016 um 11:58 schrieb Julien Viet <[email protected]>:
> 
> Yes with branch netty-4.1.3.Final
> 
>> Le 22 Jul 2016 à 09:21, 'Norman Maurer' via Netty discussions 
>> <[email protected]> a écrit :
>> 
>> Yes this sounds not right… So what are the steps to reproduce ? Fire up a 
>> vm, checkout vert.x master and run the build ?
>> 
>> 
>>> On 22 Jul 2016, at 00:51, Julien Viet <[email protected]> wrote:
>>> 
>>> Hi,
>>> 
>>> we are seing OOM "GC overhead limit exceeded"  issue in Vert.x test suite 
>>> on slow machines like Cloudbees or using virtualization on a laptop.
>>> 
>>> This issue means that the VM spend more time in GC than the rest of the VM 
>>> (98%), but I’m pretty sure everyone here knows that already.
>>> 
>>> In the case of Vert.x testsuite, it happens because we create and destroy 
>>> many Vertx instances and therefore many event loops / threads during the 
>>> tests.
>>> 
>>> I spent quite some time on the issue and I found that it happens because 
>>> the ThreadDeatchWatcher's Recycler DELAYED_RECYCLE fast thread local is a 
>>> WeakHashMap<Stack, WeakOrderQueue) used by the ThreadPoolCache grows and 
>>> retains many entries (up to 2000). This WeakHashMap contains recycled 
>>> objects with large footprint and also the Stack itself reference its Thread 
>>> that has a large footprint.
>>> 
>>> I am not saying it is a leak per se, but the maps grows and takes time to 
>>> be garbaged. This does not prevent the testsuite to run on a laptop but it 
>>> slows it down and on a slow machines it prevents running the testsuite 
>>> entirely.
>>> 
>>> I believe this behavior was introduced by this commit : 
>>> https://github.com/netty/netty/commit/afafadd3d7caf1e4b346da049baab0afeae0a4bc
>>> 
>>> The change that makes the whole difference is : 
>>> https://github.com/netty/netty/commit/afafadd3d7caf1e4b346da049baab0afeae0a4bc#diff-23eafd00fcd66829f8cce343b26c236aR226
>>> 
>>> The introduction of the field stack in the WeakOrderQueue keeps a reference 
>>> on the Stack objects, however Stack instances are also weak keys of the 
>>> fast thread local WeakHashMap and therefore defeats the purpose of the 
>>> WeakHashMap. Not entirely as it is GC’ed but much less often and increase 
>>> the memory footprint.
>>> 
>>> let me know what you think
>>> 
>>> Julien
>>> 
>>> -- 
>>> You received this message because you are subscribed to the Google Groups 
>>> "Netty discussions" group.
>>> To unsubscribe from this group and stop receiving emails from it, send an 
>>> email to [email protected].
>>> To view this discussion on the web visit 
>>> https://groups.google.com/d/msgid/netty/2A547499-D279-4286-AA95-3447EE2B621E%40julienviet.com.
>>> For more options, visit https://groups.google.com/d/optout.
>> 
>> -- 
>> You received this message because you are subscribed to the Google Groups 
>> "Netty discussions" group.
>> To unsubscribe from this group and stop receiving emails from it, send an 
>> email to [email protected].
>> To view this discussion on the web visit 
>> https://groups.google.com/d/msgid/netty/D5C167E1-D3C3-4B8F-B4B8-FD4D1793BC1F%40googlemail.com.
>> For more options, visit https://groups.google.com/d/optout.
> 
> -- 
> You received this message because you are subscribed to the Google Groups 
> "Netty discussions" group.
> To unsubscribe from this group and stop receiving emails from it, send an 
> email to [email protected].
> To view this discussion on the web visit 
> https://groups.google.com/d/msgid/netty/14FFE4AD-8D8F-4721-8A1E-C763E1F187DE%40julienviet.com.
> For more options, visit https://groups.google.com/d/optout.

-- 
You received this message because you are subscribed to the Google Groups 
"Netty discussions" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to [email protected].
To view this discussion on the web visit 
https://groups.google.com/d/msgid/netty/BC8153A8-4705-438A-BB3F-A5F099FDEAE8%40googlemail.com.
For more options, visit https://groups.google.com/d/optout.

Reply via email to