Topic
2 replies Latest Post - ‏2013-12-13T15:58:52Z by Zena1893
Zena1893
Zena1893
4 Posts
ACCEPTED ANSWER

Pinned topic GPFS with PR policy

‏2013-12-12T13:09:43Z |

Hi,

I've PR enabled (usePersistentReserve yes - PR_SHARED) and during DR test it appears the reserve is not released after mmumount/mmshutdown.

I've noticed this behaviour cause the flash copy fails due to "lun reserved" error. So I need to clear the reservation manually.

Is it expected the registration for PR will not be deleted after mmumount/mmshutdown?

Thanks, Marco

Updated on 2013-12-13T15:58:10Z at 2013-12-13T15:58:10Z by Zena1893
  • Jiaqiang
    Jiaqiang
    1 Post
    ACCEPTED ANSWER

    Re: GPFS with PR policy

    ‏2013-12-13T11:01:12Z  in response to Zena1893

    It depends on several scenarios:

    (1) if you already mount the file system to multiple nodes in the cluster, then when you mmumount/mmshutdown the daemon on a single node, the umount phase would trigger the node fence and clear keys and reservations. But if you mmshutdown -a to shutdown all daemons of nodes, it is unnecessary to do node fence and therefore reservation would not be cleared. PR fence(key and reservation clear) would only be done in scenarios when there would possibly a race between fsmgr log recovery IO and fenced node IO with unexpiried disk lease. If the scenario has no possibility to trigger this race, we would not clear the reservation.

    (2) if you do not mount the file system or only mount the file system on this single node, and directly do mmshutdown on this node, key/reservation would not be cleared.

    jiaqiang

    • Zena1893
      Zena1893
      4 Posts
      ACCEPTED ANSWER

      Re: GPFS with PR policy

      ‏2013-12-13T15:58:52Z  in response to Jiaqiang

      Hi Jiaqiang

      thanks for explaining that.

      Marco