hsperfdata causing long GC/safepoint times: Don't use mmap?

[Florian Weimer]

On 04/01/2015 04:32 PM, Alex Bagehot wrote:

> This provided the [kind of] stack we are looking for showing java provoking
> mmap activty and then being delayed by it:

Thanks.  It's pretty clear that this is the mtime update:

> 7fff8126a874 ext4_mark_inode_dirty ([kernel.kallsyms])
> 7fff8126db60 ext4_dirty_inode ([kernel.kallsyms])
> 7fff81210472 __mark_inode_dirty ([kernel.kallsyms])
> 7fff811ff301 update_time ([kernel.kallsyms])
> 7fff811ff500 file_update_time ([kernel.kallsyms])
> 7fff8126dd75 ext4_page_mkwrite ([kernel.kallsyms])
> 7fff81199add do_page_mkwrite ([kernel.kallsyms])

I wonder if that's what you see in production, too.

-- 
Florian Weimer / Red Hat Product Security