[erlang-questions] apparent memory leak in OTP 17

Sverker Eriksson sverker.eriksson@REDACTED
Fri Sep 5 18:33:15 CEST 2014


The statistics says there are a *lot* of poorly utilized carriers 
(mmap'ed regions)
waiting (in mbcs_pool) to die or be revived.

Do you have any notion about the previous peak memory load of the node?
Has it been loaded with hundreds of Gb with binaries?

/Sverker, Erlang/OTP

On 09/05/2014 01:31 PM, mats cronqvist wrote:
>    I have an application (a web server using cowboy and jiffy) that keeps
> getting killed by the Linux out-of-memory assassin.
>    This is rather unsurprising given that the emulator seems to allocate a
> factor of 1000 times more memory than it uses.
>    E.g., in the snapshot below, erlang:memory/0 reports memory usage of ~500
> Mb, while /proc/self/stat reports ~600 Gb.
>    Digging into the allocator info reveals that the binary_alloc / mbcs_pool
> allocator has booked 606086.75 Mb, and uses 357.35 Mb.
>    For the brave, I have attached a file full of system data.
>
>




More information about the erlang-questions mailing list