[LTP] [PATCH 3/3] fanotify10: Make evictable marks tests more reliable

Petr Vorel pvorel@suse.cz
Thu Nov 17 16:50:13 CET 2022


Hi Jan, Amir,

> On Wed, Nov 16, 2022 at 12:58 PM Jan Kara <jack@suse.cz> wrote:

> > On Wed 16-11-22 10:17:14, Pengfei Xu wrote:
> > > Hi Jan Kara,

> > > On 2022-11-15 at 13:47:38 +0100, Jan Kara wrote:
> > > > Tests verifying that evictable inode marks don't pin inodes in memory
> > > > are unreliable because slab shrinking (triggered by drop_caches) does
> > > > not reliably evict inodes - dentries have to take round through the LRU
> > > > list and only then get reclaimed and inodes get unpinned and then inodes
> > > > have to be rotated through their LRU list to get reclaimed. If there are
> > > > not enough freed entries while shrinking other slab caches, drop_caches
> > > > will abort attempts to reclaim slab before inodes get evicted.

> > > > Tweak evictable marks tests to use more files and marks in parallel and
> > > > just verify that some (at least half) of the marks got evicted. This
> > > > should be more tolerant to random fluctuation in slab reclaim
> > > > efficiency.

> > > If possible, could you add the Tested-by tag:
> > > Tested-by: Pengfei Xu <pengfei.xu@intel.com>

> > Sure, will do. I'll just wait whether there will be some other review
> > comments.

> I would want to say Reviewed-by, but I could only say Eyeballed-by.
> I like the change and thanks for figuring this out, but the review
> was very hard, so I did not have time to do it thoroughly.

> Good luck, Petr ;-)
Thanks :). I'm ill, hoping to be back working next week, I'll have look soon.

Kind regards,
Petr

> Thanks,
> Amir.


More information about the ltp mailing list