[Linux-cluster] GNBD server out-of-memory panics?

[Tom Lanyon]

Hi all,

Has anyone seen a server exporting disks via GNBD spontaneously start  
throwing "out of memory" kernel errors to the console?

The call trace looks like this:

out_of_memory
autoremove_wake_function
__alloc_pages
__alloc_pages
cache_alloc_refil
kmem_cache_alloc
sk_alloc
inet_create
__sock_create
:dlm:process_send_sockets
:dlm:tcp_connect_to_sock
thread_return
thread_return
error_exit
:dlm:process_send_sockets
error_exit
:dlm:process_send_sockets

All of our monitoring/graphing systems show that the server has over 3  
GB of free memory (it does nothing other than exporting GNBD devices)  
and even part of the error messages show that its used 0 bytes swap  
and has a lot of available memory.

Any ideas?

Thanks,
Tom