posix_memalign error

[Maxim Dounin]

Hello!

On Tue, Jul 31, 2018 at 09:52:29AM +0530, Anoop Alias wrote:

> I am repeatedly seeing errors like
> 
> ######################
> 2018/07/31 03:46:33 [emerg] 2854560#2854560: posix_memalign(16, 16384)
> failed (12: Cannot allocate memory)
> 2018/07/31 03:54:09 [emerg] 2890190#2890190: posix_memalign(16, 16384)
> failed (12: Cannot allocate memory)
> 2018/07/31 04:08:36 [emerg] 2939230#2939230: posix_memalign(16, 16384)
> failed (12: Cannot allocate memory)
> 2018/07/31 04:24:48 [emerg] 2992650#2992650: posix_memalign(16, 16384)
> failed (12: Cannot allocate memory)
> 2018/07/31 04:42:09 [emerg] 3053092#3053092: posix_memalign(16, 16384)
> failed (12: Cannot allocate memory)
> 2018/07/31 04:42:17 [emerg] 3053335#3053335: posix_memalign(16, 16384)
> failed (12: Cannot allocate memory)
> 2018/07/31 04:42:28 [emerg] 3053937#3053937: posix_memalign(16, 16384)
> failed (12: Cannot allocate memory)
> 2018/07/31 04:47:54 [emerg] 3070638#3070638: posix_memalign(16, 16384)
> failed (12: Cannot allocate memory)
> ####################
> 
> on a few servers
> 
> The servers have enough memory free and the swap usage is 0, yet somehow
> the kernel denies the posix_memalign with ENOMEM ( this is what I think is
> happening!)
> 
> The numbers requested are always 16, 16k . This makes me suspicious
> 
> I have no setting in nginx.conf that reference a 16k
> 
> Is there any chance of finding out what requests this and why this is not
> fulfilled

There are at least some buffers which default to 16k - for 
example, ssl_buffer_size (http://nginx.org/r/ssl_buffer_size).

You may try debugging log to futher find out where the particular 
allocation happens, see here for details:

http://nginx.org/en/docs/debugging_log.html

But I don't really think it worth the effort.  The error is pretty 
clear, and it's better to focus on why these allocations are 
denied.  Likely you are hitting some limit.

-- 
Maxim Dounin
http://mdounin.ru/