0

My setup is is like this: 1 master server 4 chunk server

Now one of the chunk server is break down, and the data can not recovered. Can I repair all the data from master server? How to do it?

sarnold
  • 102,305
  • 22
  • 181
  • 238
truease.com
  • 1,261
  • 2
  • 17
  • 30
  • 1
    What `goal` did you have set on your data? Was it greater than 1 for all files and directories? – sarnold May 02 '12 at 02:25
  • If your `goal` was simply set to 1 -- which seems likely, given as how you can't recover the data -- chances are that your data is lost. The master server does not contain file data. –  May 02 '12 at 02:37
  • yes. I have set the goal to 1 and the data lost forever. Now I set the goal to 3. Lucky the data is collect from internet, we just need to download it again. – truease.com May 10 '12 at 11:32
  • But I really don't like the way moosefs acts. Do I need to setup a full backup server? I means do mfsmount at a linux box, and rsync all the data to local storage. – truease.com May 10 '12 at 11:36
  • 1
    mooseFS is not the problem. You are taking risk with goal set to 1. It is equivalent to raid 0 spanning, also know as scary raid, because if one server goes you can't recover it. I know it is old but wanted to clear up the situation. Always plan for failure. – Atherion Apr 11 '13 at 17:19
  • MooseFS is designed to keep data safe. Without need of any external backup systems. But you need to set the goal equal to 2 or higher. It means that data will be kept in 2 (or more) copies on your data servers. You should also consider duplicating metadata kept by Master Server - setting up another machine either as the Metalogger (free) or the online Master Server backup (HA available in Pro version). – Jakub Jul 14 '17 at 08:05

1 Answers1

1

I consider that the data on the broken chunk server can be divided into 2 different types.

  • First, the goal of the lost data is 1.
  • Second, the goal of the lost data is larger than 1.

If it is the first circumstance, the data can not be repaired.

If it is the second circumstnce, the lost can be found in other chunk server. you can add a new chunk server in to the mfs cluster, the master can decide the replication between the chunk servers.

Dreen
  • 6,976
  • 11
  • 47
  • 69
songwei.beijing
  • 9
  • 2