At Thu, 09 May 2013 13:22:20 +0800,
Liu Yuan wrote:
>
> On 05/09/2013 10:38 AM, MORITA Kazutaka wrote:
> > buf = valloc(rlen);
> > if (!buf) {
> > + ret = SD_RES_NO_MEM;
> > sd_eprintf("%m");
> > goto out;
>
> It seems that we just rollback when SD_RES_NO_MEM happens, is this the
> right behavior? This will lead possible stale object.
Hmm, breaking block storage semantic in OOM situations looks bad. It's more
likely to happen than rapid node changes.
Then, the only way we can take here looks like aborting sheep (use xvalloc
instead) to me. What do you think?
Thanks,
Kazutaka
--
sheepdog mailing list
sheepdog@lists.wpkg.org
http://lists.wpkg.org/mailman/listinfo/sheepdog
