[Lustre-discuss] the usage rate of inode on mds

[lixiangna]

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

 

 

 

 

 

---

网 易 Yeah.net 免 费 邮 箱 全 新 改 版，珍 藏 帐 号 开 放，快 来 抢 注 >>

[Andreas Dilger]

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.

_______________________________________________
Lustre-discuss mailing list
Lustre-...@clusterfs.com
https://mail.clusterfs.com/mailman/listinfo/lustre-discuss

[lixiangna]

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" <xpri...@126.com>
Subject: [Lustre-discuss] the usage rate of inode on mds
To: lustre-...@clusterfs.com
Message-ID: <46EDFA30.0...@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
 
 
 
 

 

 

 

---

网 易 Yeah.net 邮 箱 夏 日 大 变 脸，带 你 感 受 新 一 代 邮 箱 的 全 新 魅 力 >>