On Wed, Sep 19, 2018 at 4:15 AM Jerry Lee <leisurelysw24@xxxxxxxxx> wrote:
>
> Hi,
>
> I got the following BUG_ON() trace when there is high memory pressure
> on a system. I'm curious whether it's kind of aggressive to trigger a
> BUG() when a memory allocation failure is encountered? Is it suitable
> to change the behaviour to propagate a -ENOMEM back to the caller or
> to retry allocation several times? Any ideas and feedback are
> appreciated, thanks!
Hi Jerry,
We can't really handle an error from ceph_msg_data_create(), so
returning it to the caller won't work. It ends up being called from
send_request(), which is a void function for a very good reason.
I don't think we should be attempting to allocate ceph_msg_data in
ceph_msg_data_create() at all. There is usually only a couple of data
items in any given message, so it should be possible to embed them in
ceph_msg.
Let me see if I can come up with a patch in the next few days.
Thanks,
Ilya
