Do we need a way to force xfsaild to run?

Do we need a way to force xfsaild to run?

[Carlos Maiolino]

Hi,

This is something I was thinking while working on that buffer retry stuff.

Currently, we have no way to force xfsaild to run, and I was wondering if
somebody has any opinion if it would be useful to create a new xfsctl, to start
xfsaild for debugging purposes, using xfs_io to trigger it.

My specific use case is while testing the code I've been working on for the
buffer retry stuff, where, I need to wait xfsaild to run to trigger the bug,
adding a way to force it, in this case at least, would be useful for me, I could
work on that, but only if somebody else can actually think this to be
interesting, adding a new xfsctl just for 'me' does not sound worth the time :)

Anybody else might have interest on it or maybe this is just a waste of time?

Cheers

-- 
Carlos

Re: Do we need a way to force xfsaild to run?

[Shan Hai]

On 2017年05月18日 22:21, Carlos Maiolino wrote:
> Hi,
>
> This is something I was thinking while working on that buffer retry stuff.
>
> Currently, we have no way to force xfsaild to run, and I was wondering if
> somebody has any opinion if it would be useful to create a new xfsctl, to start
> xfsaild for debugging purposes, using xfs_io to trigger it.
>
> My specific use case is while testing the code I've been working on for the
> buffer retry stuff, where, I need to wait xfsaild to run to trigger the bug,
> adding a way to force it, in this case at least, would be useful for me, I could
> work on that, but only if somebody else can actually think this to be
> interesting, adding a new xfsctl just for 'me' does not sound worth the time :)

For the debugging purpose how about mkfs with a small size of log
for instance 50MB and then use benchmark tools like fs_mark to
fill the log quickly(write lots of small files instead of a single
huge file), then the xfsaild will run frequently :)

Thanks
Shan Hai
> Anybody else might have interest on it or maybe this is just a waste of time?
>
> Cheers
>

Re: Do we need a way to force xfsaild to run?

[Darrick J. Wong]

On Thu, May 18, 2017 at 04:21:59PM +0200, Carlos Maiolino wrote:
> Hi,
> 
> This is something I was thinking while working on that buffer retry stuff.
> 
> Currently, we have no way to force xfsaild to run, and I was wondering if
> somebody has any opinion if it would be useful to create a new xfsctl, to start
> xfsaild for debugging purposes, using xfs_io to trigger it.
> 
> My specific use case is while testing the code I've been working on for the
> buffer retry stuff, where, I need to wait xfsaild to run to trigger the bug,
> adding a way to force it, in this case at least, would be useful for me, I could
> work on that, but only if somebody else can actually think this to be
> interesting, adding a new xfsctl just for 'me' does not sound worth the time :)
> 
> Anybody else might have interest on it or maybe this is just a waste of time?

"Run" the xfsaild as in 'push all the AIL items out to disk'?

I think xfs_freeze -f (followed immediately by xfs_freeze -u) will push
the AIL as part of the freezing process... though that might be more
work than you're looking for...?

--D

> 
> Cheers
> 
> -- 
> Carlos
> --
> To unsubscribe from this list: send the line "unsubscribe linux-xfs" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: Do we need a way to force xfsaild to run?

[Dave Chinner]

On Thu, May 18, 2017 at 04:21:59PM +0200, Carlos Maiolino wrote:
> Hi,
> 
> This is something I was thinking while working on that buffer retry stuff.
> 
> Currently, we have no way to force xfsaild to run, and I was wondering if
> somebody has any opinion if it would be useful to create a new xfsctl, to start
> xfsaild for debugging purposes, using xfs_io to trigger it.

THere's lots of ways to trigger it to run. You can do it on demand
via a freeze or remount-ro or by filling up the log, or get it to
run every second automatically iso there's no wait time at all by
turning /proc/sys/fs/xfs/xfssyncd_centiseconds down to the minimum.

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com