Maintenance of mirrored filesystem

[Tomohiro Kusumi]

Note that using live dedup on H1 is not recommended (by Matt).
I don't think anyone is really using it too.

2017-12-15 21:24 GMT+02:00 Zachary Crownover <zachary.crownover at gmail.com>:
> Also turning on dedup can save space
>
> On Dec 15, 2017 10:50 AM, "John R. Shannon" <john at johnrshannon.com> wrote:
>>
>> Thank you. That got it.
>>
>> On 12/15/17 10:49, Tim Darby wrote:
>> > On Fri, Dec 15, 2017 at 8:24 AM, John R. Shannon <john at johnrshannon.com>
>> > wrote:
>> >
>> >> I setup hammer filesystems on identical disk drives. I created a number
>> >> of PFS that are mirrored. Over time, the storage allocation on the
>> >> MASTER has remained constant while it continues to grow on the SLAVE as
>> >> shown in the following:
>> >>
>> >> MASTER                   5588G  1456G  4132G    26%    /master
>> >> SLAVE                    5588G  2775G  2813G    50%    /slave
>> >> /master/pfs/@@-1:00002   5588G  1456G  4132G    26%
>> >> /master/backups/dns
>> >> /master/pfs/@@-1:00001   5588G  1456G  4132G    26%
>> >> /master/backups/colleen
>> >> /master/pfs/@@-1:00003   5588G  1456G  4132G    26%
>> >> /master/backups/domain0
>> >> /master/pfs/@@-1:00004   5588G  1456G  4132G    26%
>> >> /master/backups/imap
>> >> /master/pfs/@@-1:00005   5588G  1456G  4132G    26%    /master
>> >>
>> >> ...
>> >>
>> >> Am I missing a maintenance step that needs to be performed on the
>> >> slave?
>> >
>> >
>> > By default, the daily periodic script for hammer cleanup only works on
>> > mounted volumes, so that's why your unmounted slaves are not being
>> > cleaned
>> > up. You can tell it to clean up unmounted volumes by adding this line to
>> > /etc/periodic.conf:
>> >
>> > daily_clean_hammer-pfslist="path_to_first_unmounted_pfs
>> > path_to_second_unmounted_pfs etc"
>> >
>> > The unfortunate side effect of this is that will cause it to stop
>> > cleaning
>> > up mounted PFSes, so I created a separate hammer cleanup script in
>> > /etc/periodic/daily that cleans up the mounted ones.
>> >
>> > Tim
>> >
>>
>> --
>>
>> John R. Shannon
>> john at johnrshannon.com