Hi everyone, I did a test to see the usage rate of inode on mds: test1 Beginning my test, the fs stats of MDS are: Fs size: 4279928 (1K-blocks) Inode count: 1222992 Block count: 1222948 Reserved block count: 61147 Free blocks: 1012426 Free Inode? 1222965 I touched files ceaselessly. When touch 1190792 files, I got the error?No space left on device.? And there were still left 32173 inodes free. I guess any pre-allocate policy in lustre, and then I did a test again. test2 Beginning my test, the fs stats of MDS are: Inode count: 490080 Block count: 489982 Reserved block count: 24499 Free blocks: 408479 Free inodes: 490067 At the end the maximum files number is 490060, and 7 free inodes left. It seems ruleless. Why the usage rate of free inode cannot be 100%? Any rule under? Thanks and best wishes, Xiang ? ? Yeah.net ? ? ? ? ? ? ? ??? ? ? ? ? ??? ? ? ? >> -------------- next part -------------- An HTML attachment was scrubbed... URL: http://mail.clusterfs.com/pipermail/lustre-discuss/attachments/20070917/47ce5e2a/attachment.html
On Sep 17, 2007 11:53 +0800, lixiangna wrote:> I did a test to see the usage rate of inode on mds: > > test1 > Beginning my test, the fs stats of MDS are: > Fs size: 4279928 (1K-blocks) > Inode count: 1222992 > Block count: 1222948 > Reserved block count: 61147 > Free blocks: 1012426 > Free Inode? 1222965 > I touched files ceaselessly. When touch 1190792 files, I got the error?No space left on device.? And there were still left 32173 inodes free. I guess any pre-allocate policy in lustre, and then I did a test again. > > test2 > Beginning my test, the fs stats of MDS are: > Inode count: 490080 > Block count: 489982 > Reserved block count: 24499 > Free blocks: 408479 > Free inodes: 490067 > At the end the maximum files number is 490060, and 7 free inodes left. It seems ruleless. Why the usage rate of free inode cannot be 100%? Any rule under?Check "lfs df -i" output. It may be that some OSTs have run out of inodes? Cheers, Andreas -- Andreas Dilger Principal Software Engineer Cluster File Systems, Inc.
Hi? The ost is still have lots of free inode and blocks. And the most strange thing is: If I umount the mds and then mount it with ldiskfs type, it still cannot touch any files on the device. But if mount it with ext3 type, touch new file become available. So it seems the problem is ldiskfs ? -------------------------------------------------------- Message: 1 Date: Mon, 17 Sep 2007 11:53:20 +0800 (CST) From: "lixiangna" <xprincess at 126.com> Subject: [Lustre-discuss] the usage rate of inode on mds To: lustre-discuss at clusterfs.com Message-ID: <46EDFA30.00000D.01003 at bj126app2.126.com> Content-Type: text/plain; charset="gb2312" Hi everyone, I did a test to see the usage rate of inode on mds: test1 Beginning my test, the fs stats of MDS are: Fs size: 4279928 (1K-blocks) Inode count: 1222992 Block count: 1222948 Reserved block count: 61147 Free blocks: 1012426 Free Inode? 1222965 I touched files ceaselessly. When touch 1190792 files, I got the error?No space left on device.? And there were still left 32173 inodes free. I guess any pre-allocate policy in lustre, and then I did a test again. test2 Beginning my test, the fs stats of MDS are: Inode count: 490080 Block count: 489982 Reserved block count: 24499 Free blocks: 408479 Free inodes: 490067 At the end the maximum files number is 490060, and 7 free inodes left. It seems ruleless. Why the usage rate of free inode cannot be 100%? Any rule under? Thanks and best wishes, Xiang ? ? Yeah.net ? ? ? ? ? ? ??? ? ? ? ? ? ? ? ? ? ? ? ? ? >> -------------- next part -------------- An HTML attachment was scrubbed... URL: http://mail.clusterfs.com/pipermail/lustre-discuss/attachments/20070918/3279c830/attachment.html