https://reviews.apache.org/r/10661/


On Thu, Apr 18, 2013 at 1:55 AM, 王国栋 <[email protected]> wrote:

> Thanks for your explanation.
>
> I will try to send a patch later this week.
>
> Guodong
>
>
> On Thu, Apr 18, 2013 at 3:14 PM, Vinod Kone <[email protected]> wrote:
>
> > Aha. That makes sense now.
> >
> > When mesos calculates disk usage, it just does a 'statvfs' on the the
> root
> > ("/") path. Since the root path is mounted on /dev/sda1 in your case, it
> > picks up that file system.
> >
> > The fix is for the slave to do disk calculation on the file system on
> which
> > the "work_dir" is mounted instead of "/".
> >
> > This is as simple as doing 'fs::usage(flags.work_dir)' in
> > Slave::checkDiskUsage() and 'fs::available(flags.work_dir) in
> > Slave::Slave()' (see src/slave/slave.cpp).
> >
> > I can get to it some time this week. But, feel free to send a patch!
> >
> >
> >
> > On Wed, Apr 17, 2013 at 8:37 PM, Vinod Kone <[email protected]> wrote:
> >
> > >
> > > Hi Guodong,
> > >
> > > Could you explain a bit about what's not working with "--work_dir"
> flag?
> > > If you mount a directory (say /mesos) in /data1 and give
> > > "--workdir=/mesos", I would presume it would work?
> > >
> > > If it doesn't can you send out some output?
> > >
> > > Thanks,
> > > Vinod
> > >
> > >
> > >
> > >
> > > On Wed, Apr 17, 2013 at 8:25 PM, 王国栋 <[email protected]> wrote:
> > >
> > >> Hi,
> > >>
> > >> It seems that I can not set the disk device which provide disk
> resource
> > in
> > >> slave.
> > >> Here is my situation.
> > >>
> > >> I install mesos in /user/mesos/ on sda, but sda is the disk for OS, so
> > the
> > >> disk size is small. When I start the slave, mesos find the disk
> resource
> > >> from sda.
> > >> But our machine have several large size disk which is mounted on
> /data1
> > >> etc.
> > >>
> > >> So I am wondering how to set mesos slave to use /data1 as the disk
> > >> resource
> > >> provider. I tried --work_dir flag, but it doesn't work on our node.
> > >>
> > >> Thanks.
> > >>
> > >>
> > >> Guodong
> > >>
> > >
> > >
> >
>

Reply via email to