Ocfs2 devel - Nov 2015 - [PATCH v2 2/4] ocfs2: sysfile interfaces for online file check

On Tue, Nov 03, 2015 at 04:20:27PM +0800, Junxiao Bi
wrote:
> Hi Gang,
> 
> On 11/03/2015 03:54 PM, Gang He wrote:
> > Hi Junxiao,
> > 
> > Thank for your reviewing.
> > Current design, we use a sysfile as a interface to check/fix a file
(via pass a ino number).
> > But, this operation is manually triggered by user, instead of
automatically  fix in the kernel.
> > Why?
> > 1) we should let users make this decision, since some users do not
want to fix when encountering a file system corruption, maybe they want to keep
the file system unchanged for a further investigation.
> If user don't want this, they should not use error=continue option, let
> fs go after a corruption is very dangerous.
Maybe we need another errors=XXX flag (maybe errors=fix)?

You both make good points, here's what I gather from the conversation:

 - Some customers would be sad if they have to manually fix corruptions.
   This takes effort on their part, and if the FS can handle it
   automatically, it should.

 - There are valid concerns that automatically fixing things is a change in
   behavior that might not be welcome, or worse might lead to unforseeable
   circumstances.

 - I will add that fixing things automatically implies checking them
   automatically which could introduce some performance impact depending on
   how much checking we're doing.

So if the user wants errors to be fixed automatically, they could mount with
errros=fix, and everyone else would have no change in behavior unless they
wanted to make use of the new feature.

> > 2) frankly speaking, this feature will probably bring a second
corruption if there is some error in the code, I do not suggest to use
automatically fix by default in the first version.
> I think if this feature could bring more corruption, then this should be
> fixed first.
Btw, I am pretty sure that Gang is referring to the feature being new and
thus more likely to have problems. There is nothing I see in here that is
file system corrupting.
	--Mark


--
Mark Fasheh

On 11/24/2015 01:46 PM, Mark Fasheh wrote:
> On Tue, Nov 03, 2015 at 04:20:27PM +0800, Junxiao Bi wrote:
>> Hi Gang,
>>
>> On 11/03/2015 03:54 PM, Gang He wrote:
>>> Hi Junxiao,
>>>
>>> Thank for your reviewing.
>>> Current design, we use a sysfile as a interface to check/fix a file
(via pass a ino number).
>>> But, this operation is manually triggered by user, instead of
automatically  fix in the kernel.
>>> Why?
>>> 1) we should let users make this decision, since some users do not
want to fix when encountering a file system corruption, maybe they want to keep
the file system unchanged for a further investigation.
>> If user don't want this, they should not use error=continue option,
let
>> fs go after a corruption is very dangerous.
> Maybe we need another errors=XXX flag (maybe errors=fix)?
Great idea Mark! I think adding errors=fix would be a good way to 
address both concerns :) It gives some control if anyone is 
uncomfortable of things getting checked/fixed automatically.
>
> You both make good points, here's what I gather from the conversation:
>
>   - Some customers would be sad if they have to manually fix corruptions.
>     This takes effort on their part, and if the FS can handle it
>     automatically, it should.
>
>   - There are valid concerns that automatically fixing things is a change
in
>     behavior that might not be welcome, or worse might lead to unforseeable
>     circumstances.
>
>   - I will add that fixing things automatically implies checking them
>     automatically which could introduce some performance impact depending
on
>     how much checking we're doing.
>
> So if the user wants errors to be fixed automatically, they could mount
with
> errros=fix, and everyone else would have no change in behavior unless they
> wanted to make use of the new feature.
>
>
>>> 2) frankly speaking, this feature will probably bring a second
corruption if there is some error in the code, I do not suggest to use
automatically fix by default in the first version.
>> I think if this feature could bring more corruption, then this should
be
>> fixed first.
> Btw, I am pretty sure that Gang is referring to the feature being new and
> thus more likely to have problems. There is nothing I see in here that is
> file system corrupting.
> 	--Mark
>
>
> --
> Mark Fasheh
>
> _______________________________________________
> Ocfs2-devel mailing list
> Ocfs2-devel at oss.oracle.com
> https://oss.oracle.com/mailman/listinfo/ocfs2-devel

Hi Mark and Junxiao,

>>> 
> On Tue, Nov 03, 2015 at 04:20:27PM +0800, Junxiao Bi wrote:
>> Hi Gang,
>> 
>> On 11/03/2015 03:54 PM, Gang He wrote:
>> > Hi Junxiao,
>> > 
>> > Thank for your reviewing.
>> > Current design, we use a sysfile as a interface to check/fix a
file (via
> pass a ino number).
>> > But, this operation is manually triggered by user, instead of
automatically
>  fix in the kernel.
>> > Why?
>> > 1) we should let users make this decision, since some users do not
want to
> fix when encountering a file system corruption, maybe they want to keep the
> file system unchanged for a further investigation.
>> If user don't want this, they should not use error=continue option,
let
>> fs go after a corruption is very dangerous.
> 
> Maybe we need another errors=XXX flag (maybe errors=fix)?
> 
> You both make good points, here's what I gather from the conversation:
> 
>  - Some customers would be sad if they have to manually fix corruptions.
>    This takes effort on their part, and if the FS can handle it
>    automatically, it should.
> 
>  - There are valid concerns that automatically fixing things is a change in
>    behavior that might not be welcome, or worse might lead to unforseeable
>    circumstances.
> 
>  - I will add that fixing things automatically implies checking them
>    automatically which could introduce some performance impact depending on
>    how much checking we're doing.
> 
> So if the user wants errors to be fixed automatically, they could mount
with
> errros=fix, and everyone else would have no change in behavior unless they
> wanted to make use of the new feature.
That is what I want to say, add a mount option to let users to decide. Here, I
want to split "error=fix"
mount option  task out from online file check feature, I think this part should
be a independent feature.
We can implement this feature after online file check is done, I want to split
the feature into some more
detailed features, implement them one by one. Do you agree this point?
> 
> 
>> > 2) frankly speaking, this feature will probably bring a second
corruption
> if there is some error in the code, I do not suggest to use automatically
fix
> by default in the first version.
>> I think if this feature could bring more corruption, then this should
be
>> fixed first.
> 
> Btw, I am pretty sure that Gang is referring to the feature being new and
> thus more likely to have problems. There is nothing I see in here that is
> file system corrupting.
> 	--Mark
> 
> 
> --
> Mark Fasheh

On 11/25/2015 05:46 AM, Mark Fasheh wrote:
> On Tue, Nov 03, 2015 at 04:20:27PM +0800, Junxiao Bi wrote:
>> Hi Gang,
>>
>> On 11/03/2015 03:54 PM, Gang He wrote:
>>> Hi Junxiao,
>>>
>>> Thank for your reviewing.
>>> Current design, we use a sysfile as a interface to check/fix a file
(via pass a ino number).
>>> But, this operation is manually triggered by user, instead of
automatically  fix in the kernel.
>>> Why?
>>> 1) we should let users make this decision, since some users do not
want to fix when encountering a file system corruption, maybe they want to keep
the file system unchanged for a further investigation.
>> If user don't want this, they should not use error=continue option,
let
>> fs go after a corruption is very dangerous.
> 
> Maybe we need another errors=XXX flag (maybe errors=fix)?
Sound great. This is a good option since user may have not enough
knowledge whether to fix the found issue.

Thanks,
Junxiao.
> 
> You both make good points, here's what I gather from the conversation:
> 
>  - Some customers would be sad if they have to manually fix corruptions.
>    This takes effort on their part, and if the FS can handle it
>    automatically, it should.
> 
>  - There are valid concerns that automatically fixing things is a change in
>    behavior that might not be welcome, or worse might lead to unforseeable
>    circumstances.
> 
>  - I will add that fixing things automatically implies checking them
>    automatically which could introduce some performance impact depending on
>    how much checking we're doing.
> 
> So if the user wants errors to be fixed automatically, they could mount
with
> errros=fix, and everyone else would have no change in behavior unless they
> wanted to make use of the new feature.
> 
> 
>>> 2) frankly speaking, this feature will probably bring a second
corruption if there is some error in the code, I do not suggest to use
automatically fix by default in the first version.
>> I think if this feature could bring more corruption, then this should
be
>> fixed first.
> 
> Btw, I am pretty sure that Gang is referring to the feature being new and
> thus more likely to have problems. There is nothing I see in here that is
> file system corrupting.
> 	--Mark
> 
> 
> --
> Mark Fasheh
>