[prev in list] [next in list] [prev in thread] [next in thread]
List: evms-devel
Subject: Re: [Evms-devel] Delete of volume encounters I/O error
From: "Mike Dunn [mike.dunn () algx ! com]" <mike ! dunn () algx ! com>
Date: 2004-03-23 15:34:25
Message-ID: 40605901.3010006 () algx ! com
[Download RAW message or body]
Actually, I did some more work last night, and was able to reproduce
the set of circumstances which create this error.
I began by removing my configuration ... since the evms tools were
enforcing certain consistency rules, I was unable to delete objects
using them. Instead, I rebooted, preventing evms from activating on
boot, and used fdisk to remove the partitions (DosSegMgr).
I then created a new segment manager for this disk, created a
container, several regions, a volume, and a snap of that volume. This
time I made the snap writable, and proceeded to copy some test data to
the snapped volume. Folling that, I mounted the snap volume, and
verified that it was a point in time copy by adding data to the snapped vol.
At this point everything seemed to work well. Read and write
operations to both volumes were robust, so I decided to try and reset
the snap volume. This is, as I recall now, where my problems began to
happen. I apologize, as now I think that I had you debugging a symptom
of my real problem.
I umounted the snap volume, and used evmsn to reset the snap volume,
and the iterface froze. IOwait on the system went immediately to 100%
and stayed there (with no associated reads/writes to the physical disk).
Finally, other I/O tests like fdisk, sync, or du would freeze as soon
as they were executed, and I had to crash the system again.
I've recovered in nearly the same way now, but evms was more willing
to delete objects this time.
Thanks!
Mike
Mike Dunn wrote:
> Hi Steve,
>
> I really appreciate your help with this. It has me really confused. I
> tried the patch, applying in on top of the debian patched evms-2.3.0
> tree. It compiled cleanly, I added the package, and confirmed that the
> evms binary was updated. However, the delete behaved in the same way.
> Also, you are correct about the read-only option being set on these
> volumes.
>
> I've also noticed very strange behaivor from the snapped volume, before
> I tried to remove the snap volumes. It seems that write operations
> work, but read operations, not so much :). For instance, I would write
> files/directories to the volume, but using 'du' to check how large they
> were caused the O/S to go to 100% I/O wait according to iostat.
>
> There was no physical disk activity, so it looked like an
> instrumentation problem, except the 'du' process was then unresponsive
> to anything (ctrl-c, TERM, KILL, ctrl-z, strace). In my experience a
> processes that behaves like this is usually blocked on I/O.
>
> Assured now that this is completely my fault :), I then decided to see
> if I could do anything to affect the snapped volume from evms. Both the
> CLI and the gui refuse to allow me to delete the snapped volume (see
> below), but this may be a normal check-condition as the volume is in the
> snapped state.
>
> If I were to try and rebuild the configuration from scratch, is the best
> way to just put a new label on the disk? It seems like evms probably
> has a configuration copy somewhere else though.
>
> Thanks!
> Mike Dunn
>
>
> fenny:~# evms -d everything
>
>
> EVMS Command Line Interpreter Version 2.3.0
>
> EVMS: query:volumes,volume=/dev/evms/Storage
>
> Volume Name: /dev/evms/Storage
> Major: 254
> Minor: 4
> Active: TRUE
> Volume Size: 74.54 GB
> Minor Number: 4
> This volume is not mounted.
> The volume is either unformatted or contains a filesystem for which EVMS
> does not currently have an FSIM.
>
> EVMS: delete:/dev/evms/Storage
>
>
> A command line parameter was out of range!
>
> EVMS: delete recursive:/dev/evms/Storage
>
>
> A command line parameter was out of range!
>
> EVMS:
>
> Steve Dobbelstein wrote:
>
>>
>>
>>
>> Mike Dun wrote:
>>
>>> Hi Steve,
>>>
>>> Thanks for you response. I apologize, but I meant to mention that this
>>> error occurs when I either execute a "delete" or a "delete recursive" on
>>> the volume object:
>>>
>>> fenny:/tmp# evms
>>>
>>>
>>> EVMS Command Line Interpreter Version 2.3.0
>>>
>>> EVMS: delete:/dev/evms/Storage-snap4
>>>
>>>
>>> I/O error!
>>>
>>> EVMS: commit
>>>
>>>
>>> I/O error!
>>>
>>> EVMS: quit
>>> fenny:/tmp# evms
>>>
>>>
>>> EVMS Command Line Interpreter Version 2.3.0
>>>
>>> EVMS: delete recursive:/dev/evms/Storage-snap4
>>>
>>>
>>> I/O error!
>>>
>>> EVMS: commit
>>>
>>>
>>> I/O error!
>>>
>>> EVMS: quit
>>> fenny:/tmp#
>>>
>>> Is this, specifically, the method that you were referring to?
>>>
>>> Thanks in Advance
>>> Mike Dunn
>>
>>
>>
>> Hello again, Mike.
>>
>> I think I now know what is going on. My guess is that your snapshot is
>> read-only. Correct?
>>
>> I was able to reproduce the problem if I created a read-only snapshot.
>> What happens is that the code is smart enough to not warn you that the
>> file
>> system will be removed if the volume is read_only. However, it's not
>> smart
>> enough to not schedule the unmkfs on a read-only volume. In other words,
>> it's a bug. Thanks for finding it.
>>
>> Try the attached patch and see if you are able to delete your snapshot
>> volume.
>>
>> Steve D.
>>
>> (See attached file: volume_delete.patch)
>>
>
>
> -------------------------------------------------------
> This SF.Net email is sponsored by: IBM Linux Tutorials
> Free Linux tutorial presented by Daniel Robbins, President and CEO of
> GenToo technologies. Learn everything from fundamentals to system
> administration.http://ads.osdn.com/?ad_id=1470&alloc_id=3638&op=click
> _______________________________________________
> Evms-devel mailing list
> Evms-devel@lists.sourceforge.net
> To subscribe/unsubscribe, please visit:
> https://lists.sourceforge.net/lists/listinfo/evms-devel
>
-------------------------------------------------------
This SF.Net email is sponsored by: IBM Linux Tutorials
Free Linux tutorial presented by Daniel Robbins, President and CEO of
GenToo technologies. Learn everything from fundamentals to system
administration.http://ads.osdn.com/?ad_id=1470&alloc_id=3638&op=click
_______________________________________________
Evms-devel mailing list
Evms-devel@lists.sourceforge.net
To subscribe/unsubscribe, please visit:
https://lists.sourceforge.net/lists/listinfo/evms-devel
[prev in list] [next in list] [prev in thread] [next in thread]
Configure |
About |
News |
Add a list |
Sponsored by KoreLogic