From: Thomas S H. <tha...@gm...> - 2011-03-04 17:43:02
|
I am currently migrating my moosefs environment into a new setup, new OS, new partitioning, changing out disks for larger disks etc. So I need to be able to take a chunkserver down, delete the chunks on it, and then have it rebalance as quickly as possible back into the overall moosefs cluster, what is the fastest way to do this? Should I make the chunks for removal before taking them offline? Will that speed up the balancing? -Thomas S Hatch |
From: Anh K. H. <ky...@vi...> - 2011-03-08 09:22:06
|
On Fri, 4 Mar 2011 10:42:56 -0700 Thomas S Hatch <tha...@gm...> wrote: > I am currently migrating my moosefs environment into a new setup, > new OS, new partitioning, changing out disks for larger disks etc. > > So I need to be able to take a chunkserver down, delete the chunks > on it, and then have it rebalance as quickly as possible back into > the overall moosefs cluster, what is the fastest way to do this? > Should I make the chunks for removal before taking them offline? > Will that speed up the balancing? I think the safest way is to mark all disks on the chunk as *obsolete*, and increase the values of "CHUNKS_WRITE_REP_LIMIT" and "CHUNKS_READ_REP_LIMIT" in the file "mfsmaster.cfg" (the master and the chunk are required to restart; the number depends on your network and your file system performance.) My practice: I have two chunk servers and goal 2 for all files. I replaced a chunk by shutting it down. It's unsafe, but it works for me as my files are local :) If number of chunks is greater than your maximum goal, it's safe to shut down the chunk server that you want to replace (IMHO). Regards, -- Anh Ky Huynh @ UTC+7 Registered Linux User #392115 |
From: Anh K. H. <ky...@vi...> - 2011-03-08 09:33:49
|
On Tue, 8 Mar 2011 16:21:43 +0700 "Anh K. Huynh" <ky...@vi...> wrote: > ... If number of chunks is greater than > your maximum goal, it's safe to shut down the chunk server that you > want to replace (IMHO). My bad. Even if the number of chunks is greater than the maximum goal, we have to ensure that there isn't any file that's under goal before shutting down any chunk server. This note comes from FAQ AFAIK -- Anh Ky Huynh @ UTC+7 Registered Linux User #392115 |
From: Michal B. <mic...@ge...> - 2011-03-08 10:08:12
|
Hi! Absolutely the fastest way would be (for each chunkserver): 1. Stop the chunkserver X 2. Copy (just using 'cp') all files from the old disks to the new ones 3. Run the chunkserver X using the new disks (of course the chunkserver may now run on a new mainboard, etc.) The standard method would be connecting new CSs along with the old ones (perfectly all in the same moment) and marking all the old disks for removal. Then you have to wait until system migrates chunks to the new servers. But it would take several days. Kind regards Michal From: Thomas S Hatch [mailto:tha...@gm...] Sent: Friday, March 04, 2011 6:43 PM To: moosefs-users Subject: [Moosefs-users] speed up balancing I am currently migrating my moosefs environment into a new setup, new OS, new partitioning, changing out disks for larger disks etc. So I need to be able to take a chunkserver down, delete the chunks on it, and then have it rebalance as quickly as possible back into the overall moosefs cluster, what is the fastest way to do this? Should I make the chunks for removal before taking them offline? Will that speed up the balancing? -Thomas S Hatch |
From: Thomas S H. <tha...@gm...> - 2011-03-08 15:03:41
|
Thanks Anh, I should have specified that we had already made those changes to the configs But I had no idea that you could just copy the chunks over, but now that I think about the architecture of MooseFS, that makes perfect sense! -Thanks 2011/3/8 Michal Borychowski <mic...@ge...> > Hi! > > > > Absolutely the fastest way would be (for each chunkserver): > > 1. Stop the chunkserver X > > 2. Copy (just using ‘cp’) all files from the old disks to the new > ones > > 3. Run the chunkserver X using the new disks (of course the > chunkserver may now run on a new mainboard, etc.) > > > > The standard method would be connecting new CSs along with the old ones > (perfectly all in the same moment) and marking all the old disks for > removal. Then you have to wait until system migrates chunks to the new > servers. But it would take several days. > > > > > > Kind regards > > Michal > > > > > > *From:* Thomas S Hatch [mailto:tha...@gm...] > *Sent:* Friday, March 04, 2011 6:43 PM > *To:* moosefs-users > *Subject:* [Moosefs-users] speed up balancing > > > > I am currently migrating my moosefs environment into a new setup, new OS, > new partitioning, changing out disks for larger disks etc. > > > > So I need to be able to take a chunkserver down, delete the chunks on it, > and then have it rebalance as quickly as possible back into the overall > moosefs cluster, what is the fastest way to do this? Should I make the > chunks for removal before taking them offline? Will that speed up the > balancing? > > > > -Thomas S Hatch > |