Zero GC Couldn't Recycle In Time

[Per Liden]

Hi,

On 12/4/19 3:42 AM, 李木木 wrote:
> Hi,
>        we recently tried zero gc and jdk11 for our online risk-control

ZGC is actually not short for "zero gc" ;)

> service to replace of G1 and jdk8.It works well during the first week,but
> we find the heap memory increase slowly next week,which seems zgc can't
> recycle garbage normally because if we call system.gc manually ,the heap
> memory will fall back to normal.Besides,when we use G1 with the same code
> ,we never met this sitation.Have you met this condition before and give me
> some advice?

Have you run long enough to see that the heap usage isn't stabilizing at 
some point?

Calling System.gc() doesn't do anything special, it's just doing a 
normal GC like any other. However, you might be calling it more often 
than what ZGC thinks is needed, which could mean that you just happen to 
see less fragmentation building up. If this is the case, you could lower 
the tolerated fragmentation using for example -XX:ZFragmentationLimit=10.

(Btw, it seems the mailing list ate your images)

cheers,
/Per

> 
>            Following is configuration:
>            Linux  64  centos   7.4.1708
>            CPU：40Threads, @2.2GHz
>            MEM：128G
>            OS Disk：240G(SSD) * 2
>            JVM config： -Xmx105g -XX:+UnlockExperimentalVMOptions -XX:+UseZGC
>            Service Type:Web Service
>            Request QPS:2000
>           JDK Version：11.0.3
> 
>            Pic1：Available Heap Memory Decrease Gradually During The Second
> Week
> 
> [image: 11.png]
> 
>             Pic2:GC Become Very Frequent  At Last Which Leads To High Usage
> Of CPU
> 
> [image: 33.png]
> 
>             Pic3：First Part: System.gc Is Called Regularly
>                        Second Part:Stop Call System.gc Manually,Heap Memory
> Increase
> 
> [image: 22.png]
> 
>               Look forward to your kind reply,Thanks!
>