Right. But the point was that after decommissioning a node, it shows up as a dead node. By taking it out of slaves or whatever you set dfs.hosts to point to, the node is effectively out of the cluster. So when you restart the cluster, you won't touch the machine and the namenode won't see it or care about it.
Of course you'll need to rebalance the cluster and run fsck to make sure you have the correct amount of replications on the rest of the cluster. -----Original Message----- From: Suresh Srinivas [mailto:[email protected]] Sent: Thursday, January 06, 2011 2:59 PM To: [email protected] Subject: Re: decommission a rack Removing from slaves file only prevents launching datanode process on that node. If you you use dfs.hosts configuration to set include file with list of datanodes, remove it from that as well. This will ensure: * datanode cannot register with the namenode. * datanode will not be shown in the dead list. Regards, Suresh On 1/6/11 12:53 PM, "Michael Segel" <[email protected]> wrote: Its ok if its in the dead column. Take it out of your slaves file and when you reboot, it will not show up at all. Until then its just a foot note in history. :-) > Date: Wed, 5 Jan 2011 21:24:57 -0500 > Subject: Re: decommission a rack > From: [email protected] > To: [email protected] > > Thanks for the response. > > I did decommission the server and it took couple of minutes. However, > I see the server in my "Dead Datanotes" page. How can I completely > forget about this box? > > > On Wed, Jan 5, 2011 at 10:01 AM, Segel, Mike <[email protected]> wrote: > > Change your topology script? > > > > One thing about rack awareness. You never want to have an unbalanced > > configuration. The caveat is if your uplinks between the ToR Switch are > > 'fat' enough. (4 10GBe should do it.) > > > > If you watched your system with the one machine on a separate rack > > (assuming it was also on a separate switch) you'd see a major network > > bottle neck. (We use Ganglia) > > > > If you're seeing the rack count wrong, you could just bounce the cloud and > > it should show only two. > > > > Note: Deprecating a machine can take a long time. You may want to consider > > just killing the node, waiting till the cluster recognizes that its down > > and I believe just kicking off an FSCK or maybe running the balancer > > program will clean that up faster than deprecating. (YMMV) > > > > HTH > > > > -Mike > > > > > > -----Original Message----- > > From: Mag Gam [mailto:[email protected]] > > Sent: Wednesday, January 05, 2011 5:56 AM > > To: [email protected] > > Subject: decommission a rack > > > > Using 0.21 > > > > I have a server which was in rack3. It was the only server in this > > rack so I decommissioned the server. fsck shows only 2 racks however > > topology shows 3 racks. How can I fix this? > > > > > > The information contained in this communication may be CONFIDENTIAL and is > > intended only for the use of the recipient(s) named above. If you are not > > the intended recipient, you are hereby notified that any dissemination, > > distribution, or copying of this communication, or any of its contents, is > > strictly prohibited. If you have received this communication in error, > > please notify the sender and delete/destroy the original message and any > > copy of it from your computer or paper files. > > The information contained in this communication may be CONFIDENTIAL and is intended only for the use of the recipient(s) named above. If you are not the intended recipient, you are hereby notified that any dissemination, distribution, or copying of this communication, or any of its contents, is strictly prohibited. If you have received this communication in error, please notify the sender and delete/destroy the original message and any copy of it from your computer or paper files.
