This sunday I upgraded our production FAS940 from 6.5.3 to 7.0 and noticed NFS clients are reporting the user or group quota as the volume size and free space when running df, when 6.5.3 and earlier used to report the qtree quota. I was suprised by this change and so far have been told by NetApp that it is a bug fix, not a bug. Just wanted to send a heads up to the list to let people know, and wondering if anyone else was affected negatively by this. I like part of the idea of the change but was not expecting it at all. Didn't see it in docs. Its interesting but inappropriate for some of our volumes the way we have quotas setup, and it prevents non-admins from seeing if the volume or qtree is nearing full.
Adam McDougall mcdouga9@egr.msu.edu writes:
This sunday I upgraded our production FAS940 from 6.5.3 to 7.0 and noticed NFS clients are reporting the user or group quota as the volume size and free space when running df, when 6.5.3 and earlier used to report the qtree quota. I was suprised by this change and so far have been told by NetApp that it is a bug fix, not a bug. Just wanted to send a heads up to the list to let people know, and wondering if anyone else was affected negatively by this. I like part of the idea of the change but was not expecting it at all. Didn't see it in docs. Its interesting but inappropriate for some of our volumes the way we have quotas setup, and it prevents non-admins from seeing if the volume or qtree is nearing full.
This doesn't sound like a "bug fix" to me, but a change of spec! I'm more than a little doubtful whether it's a good one.
A number of questions occur to me:
1. Does df show user quota when there is no qtree involved? (i.e. the user quota applies to the whole volume)
2. If there is a qtree quota but no user quota, does df still show the qtree quota as before?
3. _Which_ user? The one in the authorisation info in the NFS call? (that doesn't seem consistent with client-side caching) The one owning the NFS-mounted directory?
4. Whatever the answer to (3), how do group quotas come into it? If there is both a user and a group quota, which is used?
(Well, some others too, but that's a start!) Hopefully someone actually running ONTAP 7.0 [*] and NFS will be able to answer some of them!
[*] I'm pencilling in Easter for ONTAP 7.0 here, although Adam's posting doesn't make me any happier about jumping in...
Sorry for the long delay before replying, I've been busy lately.
On Tue, Feb 22, 2005 at 04:31:02PM +0000, Chris Thompson wrote:
Adam McDougall mcdouga9@egr.msu.edu writes:
This sunday I upgraded our production FAS940 from 6.5.3 to 7.0 and noticed NFS clients are reporting the user or group quota as the volume size and free space when running df, when 6.5.3 and earlier used to report the qtree quota. I was suprised by this change and so far have been told by NetApp that it is a bug fix, not a bug. Just wanted to send a heads up to the list to let people know, and wondering if anyone else was affected negatively by this. I like part of the idea of the change but was not expecting it at all. Didn't see it in docs. Its interesting but inappropriate for some of our volumes the way we have quotas setup, and it prevents non-admins from seeing if the volume or qtree is nearing full.
This doesn't sound like a "bug fix" to me, but a change of spec! I'm more than a little doubtful whether it's a good one.
A number of questions occur to me:
1. Does df show user quota when there is no qtree involved? (i.e. the user quota applies to the whole volume)
On my old existing traditional volumes, the data was in a qtree with a qtree quota set to prevent the qtree from consuming all the space in the volume. After 7.0, these qtree mounts showed the user quota in df instead of the qtree quota as in 6.5 and previous. When I created some flexvols, I did not use a qtree for the whole set of data, and df consistantly shows the volume size and free space. (I just now realized I forgot to test creating a new volume and mounting a qtree. With flexvols, I don't intend to use a qtree to put a cap on the data, so I've been mounting the volume directly.)
2. If there is a qtree quota but no user quota, does df still show the qtree quota as before?
I have not tried this. As above I forgot to try testing with mounting a new qtree.
3. _Which_ user? The one in the authorisation info in the NFS call? (that doesn't seem consistent with client-side caching) The one owning the NFS-mounted directory?
The user executing df has his/her own quota reported as the df results.
4. Whatever the answer to (3), how do group quotas come into it? If there is both a user and a group quota, which is used?
I have not mixed these on our volumes, as my old home directory volume had a qtree with a qtree quota and per-user quotas, but my old research data volume had a qtree with a qtree quota and per-group quotas. If I can get the "new" behavior to poke its head out again, I'll try this.
(Well, some others too, but that's a start!) Hopefully someone actually running ONTAP 7.0 [*] and NFS will be able to answer some of them!
[*] I'm pencilling in Easter for ONTAP 7.0 here, although Adam's posting doesn't make me any happier about jumping in...
Chris Thompson Email: cet1@cam.ac.uk
On a whole, I am pretty happy with 7.0 and this was an opportune time to convert to flexvols since we recently obtained a new 2T disk shelf. I did not want to pass up the chance at using our new disk shelf for conversion, and old volumes were filling up so I had to work quick.
Flexvols are working awesome, my quotas appear to be working fine under the hood, and other than df reporting, the only other gotcha for me has been that I need to perform a quota off/on to refresh quotas until they fix "quota resize" for me. I put my trust in NetApp to provide a release that would serve data without crashing, and it is filling that function, and I accept the small issues I've observed as a burden for choosing to run a .0 release of software.
I think the df display is a bad problem. Our users don't have privs to rsh to the filers and do quota reports.
Joe User goes to his home directory, fills it up, gets an error that he can't write, does a df and says - hey I have 1T of space here. Well, no you don't, you have 2G, the home directory qtree has 1T.
Or worse, Joe User does a df and sees that he has 1T of space and assumes that he can write all he wants so he writes a procedure with this assumption. User directory is bad... now let's discuss project space. What am I supposed to tell my users... sorry but you all have to come to one of the sysadmins to know what you have used and how much project space is available. Yeah...that'll be fun.
We have just upgraded 6 of our 11 filers to 6.5.3P4. We are upgrading 3 more tonight. I was not aware of this issue.
Does anyone have a bug open with NA on this so I can track it and put in my $.02 on it? If you have already mentioned it and I missed it in a prior email, sorry about that. C-
On Tue, Mar 15, 2005 at 04:46:16PM -0500, Adam McDougall wrote:
Sorry for the long delay before replying, I've been busy lately.
On Tue, Feb 22, 2005 at 04:31:02PM +0000, Chris Thompson wrote:
Adam McDougall mcdouga9@egr.msu.edu writes:
This sunday I upgraded our production FAS940 from 6.5.3 to 7.0 and noticed NFS clients are reporting the user or group quota as the volume size and free space when running df, when 6.5.3 and earlier used to report the qtree quota. I was suprised by this change and so far have been told by NetApp that it is a bug fix, not a bug. Just wanted to send a heads up to the list to let people know, and wondering if anyone else was affected negatively by this. I like part of the idea of the change but was not expecting it at all. Didn't see it in docs. Its interesting but inappropriate for some of our volumes the way we have quotas setup, and it prevents non-admins from seeing if the volume or qtree is nearing full.
This doesn't sound like a "bug fix" to me, but a change of spec! I'm more than a little doubtful whether it's a good one.
A number of questions occur to me:
- Does df show user quota when there is no qtree involved? (i.e. the user quota applies to the whole volume)
On my old existing traditional volumes, the data was in a qtree with a qtree quota set to prevent the qtree from consuming all the space in the volume. After 7.0, these qtree mounts showed the user quota in df instead of the qtree quota as in 6.5 and previous. When I created some flexvols, I did not use a qtree for the whole set of data, and df consistantly shows the volume size and free space. (I just now realized I forgot to test creating a new volume and mounting a qtree. With flexvols, I don't intend to use a qtree to put a cap on the data, so I've been mounting the volume directly.)
- If there is a qtree quota but no user quota, does df still show the qtree quota as before?
I have not tried this. As above I forgot to try testing with mounting a new qtree.
- _Which_ user? The one in the authorisation info in the NFS call? (that doesn't seem consistent with client-side caching) The one owning the NFS-mounted directory?
The user executing df has his/her own quota reported as the df results.
- Whatever the answer to (3), how do group quotas come into it? If there is both a user and a group quota, which is used?
I have not mixed these on our volumes, as my old home directory volume had a qtree with a qtree quota and per-user quotas, but my old research data volume had a qtree with a qtree quota and per-group quotas. If I can get the "new" behavior to poke its head out again, I'll try this.
(Well, some others too, but that's a start!) Hopefully someone actually running ONTAP 7.0 [*] and NFS will be able to answer some of them!
[*] I'm pencilling in Easter for ONTAP 7.0 here, although Adam's posting doesn't make me any happier about jumping in...
Chris Thompson Email: cet1@cam.ac.uk
On a whole, I am pretty happy with 7.0 and this was an opportune time to convert to flexvols since we recently obtained a new 2T disk shelf. I did not want to pass up the chance at using our new disk shelf for conversion, and old volumes were filling up so I had to work quick.
Flexvols are working awesome, my quotas appear to be working fine under the hood, and other than df reporting, the only other gotcha for me has been that I need to perform a quota off/on to refresh quotas until they fix "quota resize" for me. I put my trust in NetApp to provide a release that would serve data without crashing, and it is filling that function, and I accept the small issues I've observed as a burden for choosing to run a .0 release of software.
I may be confused but it sounds like you are describing the old behavior, which 6.5.3 exhibits. I was annoyed that df was showing space other than the qtree quota like it used to. We are using a feature in samba to run a 'dfree' script that we wrote to query the remaining quota for a particular directory and return the size and used to samba as the disk size. This made windows users more aware of the space ranges they should be concerned with for home directories.
Our other volumes had directories of shared data, so any way you cut them, its fairly difficult for a user to determine exact consumption. If they have access to run a script from a host or web page then you could customize a check for them, which may do something as simple as an rquota request (if supported for that type of quota).
All in all, I was alerted today that a Bug ID 154530 was created and will be "fixed" in a future release. All I really ask is for a good documentation of the intended behavior, and optimally, a admin-configurable behavior for statfs which df uses.
On Wed, Mar 16, 2005 at 11:36:04AM -0600, Chris Blackmor wrote:
I think the df display is a bad problem. Our users don't have privs to rsh to the filers and do quota reports.
Joe User goes to his home directory, fills it up, gets an error that he can't write, does a df and says - hey I have 1T of space here. Well, no you don't, you have 2G, the home directory qtree has 1T.
Or worse, Joe User does a df and sees that he has 1T of space and assumes that he can write all he wants so he writes a procedure with this assumption. User directory is bad... now let's discuss project space. What am I supposed to tell my users... sorry but you all have to come to one of the sysadmins to know what you have used and how much project space is available. Yeah...that'll be fun.
We have just upgraded 6 of our 11 filers to 6.5.3P4. We are upgrading 3 more tonight. I was not aware of this issue.
Does anyone have a bug open with NA on this so I can track it and put in my $.02 on it? If you have already mentioned it and I missed it in a prior email, sorry about that. C-
On Tue, Mar 15, 2005 at 04:46:16PM -0500, Adam McDougall wrote:
Sorry for the long delay before replying, I've been busy lately.
On Tue, Feb 22, 2005 at 04:31:02PM +0000, Chris Thompson wrote:
Adam McDougall mcdouga9@egr.msu.edu writes:
This sunday I upgraded our production FAS940 from 6.5.3 to 7.0 and noticed NFS clients are reporting the user or group quota as the volume size and free space when running df, when 6.5.3 and earlier used to report the qtree quota. I was suprised by this change and so far have been told by NetApp that it is a bug fix, not a bug. Just wanted to send a heads up to the list to let people know, and wondering if anyone else was affected negatively by this. I like part of the idea of the change but was not expecting it at all. Didn't see it in docs. Its interesting but inappropriate for some of our volumes the way we have quotas setup, and it prevents non-admins from seeing if the volume or qtree is nearing full.
This doesn't sound like a "bug fix" to me, but a change of spec! I'm more than a little doubtful whether it's a good one.
A number of questions occur to me:
- Does df show user quota when there is no qtree involved? (i.e. the user quota applies to the whole volume)
On my old existing traditional volumes, the data was in a qtree with a qtree quota set to prevent the qtree from consuming all the space in the volume. After 7.0, these qtree mounts showed the user quota in df instead of the qtree quota as in 6.5 and previous. When I created some flexvols, I did not use a qtree for the whole set of data, and df consistantly shows the volume size and free space. (I just now realized I forgot to test creating a new volume and mounting a qtree. With flexvols, I don't intend to use a qtree to put a cap on the data, so I've been mounting the volume directly.)
- If there is a qtree quota but no user quota, does df still show the qtree quota as before?
I have not tried this. As above I forgot to try testing with mounting a new qtree.
- _Which_ user? The one in the authorisation info in the NFS call? (that doesn't seem consistent with client-side caching) The one owning the NFS-mounted directory?
The user executing df has his/her own quota reported as the df results.
- Whatever the answer to (3), how do group quotas come into it? If there is both a user and a group quota, which is used?
I have not mixed these on our volumes, as my old home directory volume had a qtree with a qtree quota and per-user quotas, but my old research data volume had a qtree with a qtree quota and per-group quotas. If I can get the "new" behavior to poke its head out again, I'll try this.
(Well, some others too, but that's a start!) Hopefully someone actually running ONTAP 7.0 [*] and NFS will be able to answer some of them!
[*] I'm pencilling in Easter for ONTAP 7.0 here, although Adam's posting doesn't make me any happier about jumping in...
Chris Thompson Email: cet1@cam.ac.uk
On a whole, I am pretty happy with 7.0 and this was an opportune time to convert to flexvols since we recently obtained a new 2T disk shelf. I did not want to pass up the chance at using our new disk shelf for conversion, and old volumes were filling up so I had to work quick.
Flexvols are working awesome, my quotas appear to be working fine under the hood, and other than df reporting, the only other gotcha for me has been that I need to perform a quota off/on to refresh quotas until they fix "quota resize" for me. I put my trust in NetApp to provide a release that would serve data without crashing, and it is filling that function, and I accept the small issues I've observed as a burden for choosing to run a .0 release of software.
-- ----------------------------------------------------------------------- * Chris Blackmor _______ | * * Advanced Micro Devices ____ | | My comments are * * Phone: (512) 602-1608 /| | | | mine, and mine * * Fax: (512) 602-5155 | |___| | | alone. * * Email: chris.blackmor@amd.com |____/ | | * ----------------------------------------------------------------------- * Matter transference beams, he decided, were not as much fun as a * * good solid kick in the head. * * Hitchhikers Guide to the Universe * -----------------------------------------------------------------------
mcdouga9@egr.msu.edu (Adam McDougall) wrote:
[...]
All in all, I was alerted today that a Bug ID 154530 was created and will be "fixed" in a future release. All I really ask is for a good documentation of the intended behavior, and optimally, a admin-configurable behavior for statfs which df uses.
154530 has appeared publicly now, and its description makes interesting reading:
| In releases prior to 7.0, running the 'df' command on an NFS mount | of an exported qtree shows the export size as the smaller of the | volume size or the qtree quota, if one exists. In 7.0, this behavior | was inadvertently changed to return the size as the minimum of the | volume size or any applicable quota, including user and group quotas | for the user running 'df' in addition to any qtree quota.
The implication is that no change to the statvfs() results was ever actually intended in 7.0. That makes me a lot happier (except that I have to wait for the fix, of course).
-
Adam McDougall -
Chris Blackmor -
Chris Thompson