Topic
4 replies Latest Post - ‏2013-12-06T03:03:51Z by yuri
Aakash_Agarwal
Aakash_Agarwal
2 Posts
ACCEPTED ANSWER

Pinned topic Replication of Installed files on Data nodes

‏2013-12-04T05:38:45Z |

Hi,

how GPFS handles the situation in which some/all of files installed with gpfs on Data nodes get corrupted? Please note I am not talking about the Data blocks here but the files which are placed on the data node as installation process, if those files get corrupt GPFS is capable of detecting that and recovering from it?(even with manual intervention) Or do we need to have a separate Replication mechanism for that like RAID?

 

Thanks

 

  • yuri
    yuri
    92 Posts
    ACCEPTED ANSWER

    Re: Replication of Installed files on Data nodes

    ‏2013-12-04T23:51:50Z  in response to Aakash_Agarwal

    One can optionally enable replication of data in GPFS.  This is logically similar to using RAID1/mirroring on the disk controller level.  This has the obvious drawback of lower storage efficiency (50%).  Normally GPFS is configured to use RAID volumes provided by a disk controller/RAID adapter, using RAID5 or RAID6 erasure code.

    yuri

    • Aakash_Agarwal
      Aakash_Agarwal
      2 Posts
      ACCEPTED ANSWER

      Re: Replication of Installed files on Data nodes

      ‏2013-12-05T10:02:17Z  in response to yuri

      By optionally enabling replication of data you mean the files which are placed on data node as part of installation or the data placed on GPFS? because we want it just for the binary files which are placed while installing BI and should affect the storage capacity of the disk on which these files are placed and not the overall capacity of the node. 

      Aakash

      • yuri
        yuri
        92 Posts
        ACCEPTED ANSWER

        Re: Replication of Installed files on Data nodes

        ‏2013-12-06T03:03:51Z  in response to Aakash_Agarwal

        Perhaps I misunderstood what you're after.  If you concern is the integrity of GPFS code binaries (and other GPFS product files), you can easily verify this using the OS package management tool.  For example, on Linux you can use 'rpm -V gpfs.base'.  This task has nothing to do with GPFS per se, it's something common to all packages.  GPFS has no special code to verify the install integrity.

        yuri

  • ufa
    ufa
    86 Posts
    ACCEPTED ANSWER

    Re: Replication of Installed files on Data nodes

    ‏2013-12-05T10:51:30Z  in response to Aakash_Agarwal

    I understand that your installation (i.e. files below /usr/lpp/mmfs and/or /var/mmfs) is corrupted.

    I am not sure whether you mean they are really broken, or hacked by something/someone. I am also not sure whether that has affected all your nodes or just a subset.

    If you have accidentally broken the installation on one or a few GPFS nodes, I'd reinstall GPFS on those and restore the cluster config to them by running

    mmsdrrestore -p <IP_of_a_proper_cluster_member>

    on each of them.

    In each case, if the /var/mmfs/gen/mmsdrfs file is still in the right shape on at least one of your nodes, you should be able to recover.

    You should consider getting system backups which would be able to restore your systems from bare metal.

    ufa