Hello,

On 4/16/12 12:49 PM, Alex Hermann wrote:

Hello,

i'm seeing a lot of failed memory allocations for shared memory after the shared memory has been fully used once (see shmem:max_used_size below).

Used_size keeps increasing fast on a heavy loaded server, until it reaches the total_size. From that moment memory allocations start failing and eventually used_size starts dropping. After while, even when there's a lot of free memory, new allocations keep failing.

ERROR:tm:build_uac_req: no more share memory (19852) ERROR:tm:t_uac: failed to build message ERROR:presence:send_notify_request: in function tmb.t_request_within ERROR:presence:notify: sending Notify not successful ERROR:presence:publ_notify: Could not send notify for dialog

When i request the shmem statistics at such time, there's lots of free memory:

shmem:total_size = 268435456 shmem:used_size = 3749696 shmem:real_used_size = 4822288 shmem:max_used_size = 268048584 shmem:free_size = 263613168 shmem:fragments = 62156

This is on 1.5.x, any suggestion on why it keeps failing and how to prevent this error?

could be a fragmentation issue, although the number of fragments is not very high.

You can recompile with QM_DBG_MALLOC and QM_JOIN_FREE (remove F_MALLOC define) and try again and see if it is better.

In upcoming 3.3, the join option is config parameter, for both Q_ and F_ mallocs.

Cheers, Daniel