Hammer Mirror Questions

Matthew Dillon dillon at apollo.backplane.com
Thu Apr 1 09:45:16 PDT 2010

:Should the mtime/ctime on the slave pfs match the master.
: From the Master
:ls -l /pfs/data/yy
:-rw-r--r--  1 root  wheel  78 Mar 31 22:22 /pfs/data/yy
: From the Slave
:ls -l /pfs/data/yy
:-rw-r--r--  1 root  wheel  78 Mar 30 22:03 /pfs/data/yy

    ctime should.  mtime and atime not necessarily.  And mtime/atime
    certainly will not match on snapshots.

:Also on the remote slave system I have the /pfs/data mounted as /data 
:but the data in the mount does not seem to change unless I umount and re 
:mount it.
: From the remote machine
:ls -l /data/yy /pfs/data/yy
:-rw-r--r--  1 root  wheel  39 Mar 30 22:03 /data/yy
:-rw-r--r--  1 root  wheel  78 Mar 30 22:03 /pfs/data/yy

    A slave PFS softlink uses the transaction id of the most recent
    mirror synchronization.  When you null-mount the softlink you
    basically freeze the transaction id for that mount to whatever the
    null mount code saw in the softlink as-of the time of the mount.
    So you have to umount/mount if you want to see it updated.  Most
    people don't bother mounting slave PFSs, they just CD into the
    softlink and re-CD if they want to see it updated.

    This isn't a bug.  The filesystem can't safely rip-out (make changes)
    to the topology without telling the kernel about it and that is
    basically impossible to track when mirroring, so any mount or cd into
    a slave PFS freezes the transaction id for that mount/cd.

					Matthew Dillon 
					<dillon at backplane.com>

More information about the Kernel mailing list