Hi,all

I set dfs.name.dir to a comma-delimited list of directories, dir1 is in /dev/sdb1 dir2 is in /dev/sdb2 and dir3 is nfs derectory.
What happens if /dev/sdb1 disk error, so dir1 cannot be read and write?

What happens if nfs server down, so dir3 cannot be read and write?
Will hadoop ignore the bad directory and use the good directory and continue server?

Thanks.

Yes. But depending on the version you're using, you may have to
manually restart the NN after fixing the mount points, to get the
directories in action again.

2011/5/25 ccxixicc <[EMAIL PROTECTED]>:
>
> Hi,all
> I set dfs.name.dir to a comma-delimited list of directories, dir1 is in
> /dev/sdb1 dir2 is in /dev/sdb2 and dir3 is nfs derectory.
> What happens if /dev/sdb1 disk error, so dir1 cannot be read and write?
> What happens if nfs server down, so dir3 cannot be read and write?
> Will hadoop ignore the bad directory and use the good directory and continue
> server?
> Thanks.
>

--
Harsh J

I'm using 0.20.2.

I had some test. I dont know how to simulate a disk failure, just chmod 000 dir1, the namenode shutdown immediately. And NN will hang if the nfs server down.
 
 
------------------ Original ------------------
From:  "Harsh J"<[EMAIL PROTECTED]>;
Date:  Wed, May 25, 2011 03:49 PM
To:  "hdfs-user"<[EMAIL PROTECTED]>;
Subject:  Re: What if one of the directory(dfs.name.dir) rw error ?

 
 Yes. But depending on the version you're using, you may have to
manually restart the NN after fixing the mount points, to get the
directories in action again.

2011/5/25 ccxixicc <[EMAIL PROTECTED]>:
>
> Hi,all
> I set dfs.name.dir to a comma-delimited list of directories, dir1 is in
> /dev/sdb1 dir2 is in /dev/sdb2 and dir3 is nfs derectory.
> What happens if /dev/sdb1 disk error, so dir1 cannot be read and write?
> What happens if nfs server down, so dir3 cannot be read and write?
> Will hadoop ignore the bad directory and use the good directory and continue
> server?
> Thanks.
>

--
Harsh J

In my experience I had to edit hdfs and mapred setup on a server with
a disk missing.

Tom

2011/5/25 ccxixicc <[EMAIL PROTECTED]>:
>
> Hi,all
> I set dfs.name.dir to a comma-delimited list of directories, dir1 is in
> /dev/sdb1 dir2 is in /dev/sdb2 and dir3 is nfs derectory.
> What happens if /dev/sdb1 disk error, so dir1 cannot be read and write?
> What happens if nfs server down, so dir3 cannot be read and write?
> Will hadoop ignore the bad directory and use the good directory and continue
> server?
> Thanks.
>

>>I dont know how to simulate a disk failure..

Couple of things you could do. chmod 000 is one.
1. umount -l
2. mount ro only
3. If machine has hot swappable disks, pull out a disk.

-Bharath

________________________________
From: ccxixicc <[EMAIL PROTECTED]>
To: hdfs-user <[EMAIL PROTECTED]>
Sent: Wednesday, May 25, 2011 1:07 AM
Subject: Re: What if one of the directory(dfs.name.dir) rw error ?
I'm using 0.20.2. 

I had some test. I dont know how to simulate a disk failure, just chmod 000 dir1, the namenode shutdown immediately. And NN will hang if the nfs server down.

 
 
------------------ Original ------------------
From:  "Harsh J"<[EMAIL PROTECTED]>;
Date:  Wed, May 25, 2011 03:49 PM
To:  "hdfs-user"<[EMAIL PROTECTED]>;
Subject:  Re: What if one of the directory(dfs.name.dir) rw error ?
 
Yes. But depending on the version you're using, you may have to
manually restart the NN after fixing the mount points, to get the
directories in action again.

2011/5/25 ccxixicc <[EMAIL PROTECTED]>:
>
> Hi,all
> I set dfs.name.dir to a comma-delimited list of directories, dir1 is in
> /dev/sdb1 dir2 is in /dev/sdb2 and dir3 is nfs derectory.
> What happens if /dev/sdb1 disk error, so dir1 cannot be read and write?
> What happens if nfs server down, so dir3 cannot be read and write?
> Will hadoop ignore the bad directory and use the good directory and continue
> server?
> Thanks.
>

--
Harsh J

You can simulate disk failure by some fault injection techniques.
Applying AspectJ is one of them.

On Wed, May 25, 2011 at 3:07 AM, ccxixicc <[EMAIL PROTECTED]> wrote:

>
> I'm using 0.20.2.
> I had some test. I dont know how to simulate a disk failure, just chmod 000
> dir1, the namenode shutdown immediately. And NN will hang if the nfs server
> down.
>
>
>
> ------------------ Original ------------------
> *From: * "Harsh J"<[EMAIL PROTECTED]>;
> *Date: * Wed, May 25, 2011 03:49 PM
> *To: * "hdfs-user"<[EMAIL PROTECTED]>;
> *Subject: * Re: What if one of the directory(dfs.name.dir) rw error ?
>
> Yes. But depending on the version you're using, you may have to
> manually restart the NN after fixing the mount points, to get the
> directories in action again.
>
> 2011/5/25 ccxixicc <[EMAIL PROTECTED]>:
> >
> > Hi,all
> > I set dfs.name.dir to a comma-delimited list of directories, dir1 is in
> > /dev/sdb1 dir2 is in /dev/sdb2 and dir3 is nfs derectory.
> > What happens if /dev/sdb1 disk error, so dir1 cannot be read and write?
> > What happens if nfs server down, so dir3 cannot be read and write?
> > Will hadoop ignore the bad directory and use the good directory and
> continue
> > server?
> > Thanks.
> >
>
>
>
> --
> Harsh J
>
>

On Wed, May 25, 2011 at 10:48PM, Thanh Do wrote:
>    You can simulate disk failure by some fault injection techniques.
>    Applying AspectJ is one of them.

Fault injection is there, so you can just check src/test/aop and
src/test/system for references, etc.

>
>    On Wed, May 25, 2011 at 3:07 AM, ccxixicc <[EMAIL PROTECTED]> wrote:
>
>      I'm using 0.20.2.A
>      I had some test. I dont know how to simulate a disk failure, just chmod
>      000 dir1, the namenode shutdown immediately. And NN will hang if the nfs
>      server down.
>      A
>      A
>      ------------------A OriginalA ------------------
>      From: A "Harsh J"<[EMAIL PROTECTED]>;
>      Date: A Wed, May 25, 2011 03:49 PM
>      To: A "hdfs-user"<[EMAIL PROTECTED]>;
>      Subject: A Re: What if one of the directory(dfs.name.dir) rw error ?
>      A
>      Yes. But depending on the version you're using, you may have to
>      manually restart the NN after fixing the mount points, to get the
>      directories in action again.
>
>      2011/5/25 ccxixicc <[EMAIL PROTECTED]>:
>      >
>      > Hi,all
>      > I set dfs.name.dir to a comma-delimited list of directories, dir1 is
>      in
>      > /dev/sdb1 dir2 is in /dev/sdb2 and dir3 is nfs derectory.
>      > What happens if /dev/sdb1 disk error, so dir1 cannot be read and
>      write?
>      > What happens if nfs server down, so dir3 cannot be read and write?
>      > Will hadoop ignore the bad directory and use the good directory and
>      continue
>      > server?
>      > Thanks.
>      >
>
>      --
>      Harsh J