[sheepdog] [PATCH 1/2] sheep: remove fd leak in vdi deletion process

[Hitoshi Mitake]

At Tue, 28 Jan 2014 11:44:11 +0800,
Liu Yuan wrote:
> 
> [1  <text/plain; UTF-8 (quoted-printable)>]
> 2014-1-28 AM11:42于 "Liu Yuan" <namei.unix at gmail.com>写道：
> >
> > we have 100 nodes, no object cache enabled. after several days stress
> testing, we sheep daemon on some nodes having several megabyte RSS even
> when IO is very low, normally, sheep should take less than 100m for idle
> >
> 
> several hundred mega

OK, please share more detailed information (usage of hypervolume, HTTP
interface, etc) when you have a PC.

Thanks,
Hitoshi

> 
> > 2014-1-28 AM11:37于 "Hitoshi Mitake" <mitake.hitoshi at gmail.com>写道：
> >
> >> At Tue, 28 Jan 2014 11:32:16 +0800,
> >> Liu Yuan wrote:
> >> >
> >> > On Tue, Jan 28, 2014 at 11:17:36AM +0800, Liu Yuan wrote:
> >> > > On Tue, Jan 28, 2014 at 11:38:21AM +0900, Hitoshi Mitake wrote:
> >> > > > The commit 5e460b4d6444 introduced fd leak in the vdi deletion
> >> > > > process. This patch removes it.
> >> > > >
> >> > >
> >> > > You should release di and dw too, I guess this is why I found severe
> memory
> >> > > leak in our test cluster.
> >> >
> >> > After a close look, seems that di and dw is releaed. Need double check.
> >>
> >> Yes delete_one_vdi_done() frees di and dw.
> >>
> >> BTW, how do you detect memory leak in your cluster? valgrind? It must
> >> be a serious problem. If you know a way to reproduce, I'd like you to
> >> share it.
> >>
> >> Thanks,
> >> Hitoshi
> >>
> [2  <text/html; UTF-8 (quoted-printable)>]
>