[erlang-questions] [ANN] Erlang/OTP 18.0-rc1 is available for testing.

[Max Lapshin]

Here is our wrapper around ffmpeg libavcodec:

https://gist.github.com/maxlapshin/b0a1cb075daf1f2110b9

When I use dirty nifs (uncomment // #define USE_DIRTY_NIFS),
after some time of working, whole erlang VM goes down in CPU usage, slowdown
response in console and starts timeouting on HTTP requests.

When I start thread per instance of encoder/decoder, everything works.

Usual time, spent in one such long call is around 10-20 ms.




On Fri, Mar 27, 2015 at 12:00 AM, Steve Vinoski <vinoski@REDACTED> wrote:

>
>
> On Thu, Mar 26, 2015 at 2:59 PM, Max Lapshin <max.lapshin@REDACTED>
> wrote:
>
>> About dirty schedulers:  seems that they still can block node. I've tried
>> again to enable them on 17.4, but had to return back to self-made thread
>> pool.
>>
>> What should I do to help to debug it?
>>
>
> Supply me with the details of what you're doing, showing me code if
> possible. The smaller the test case that shows the problem, the better.
> Also please explain what you mean by "block node".
>
> --steve
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://erlang.org/pipermail/erlang-questions/attachments/20150327/55106067/attachment.htm>