Fixing issue performance of group and owner file attribute cache which was introduced in 2.6.1

[Classe]

A fix for #16

[Classe]

Is this fix okay from your point of view? Asking if I should attend fixing the rest of the builds

[Classe]

@plamentotev @krosenvold what do you think about this?

[plamentotev]

Hi,
Thank you for the contribution. Really appreciate it that you took your time to investigate and share back with the community. Those kind of problems are quite hard to spot and find especially when they manifest only on certain configurations.

I can take a look latter this week, but in general the changes look ok.

[michael-o]

The question is: why does unix:* incur so much overhead? We are likely fixing the symptom, but not the real cause.

[Classe]

I would say that fetching group name and owner name can be lookup heavy on some systems with some security related to mapping group and owner name to gid/uid. This causes the issue together with the 2.6.1 that unintentionally disabled this naming lookup caching.

[olamy]

LGTM maybe you can write some jmh tests to compare. I'm happy to run them on a beefy machine.

[plamentotev]

I've added some small comments. Also if you can squash all commit in single one and add comment to the commit message. I saw that you've added a comment in the issue describing the problem and the solution but would be great if you add that to the commit message as well.

[Classe]

Just need to figure out how to squash the commits and push it to same pull request

[Classe]

Now I think I got it right :) thanks for the patience.

[plamentotev]

Looks great. Thanks again for your contribution.

[Classe]

No problem, did you miss merging the pull request?
"Closed with unmerged commits
This pull request is closed, but the Classe:fix/attribute-cache branch has unmerged commits"

[Classe]

Ah, saw now that it was merged, strange message in github however

[plamentotev]

That it because I've merged it locally and I've changed the commit message (amended your commit) to include the closes keywords. That makes the merge looks strange in GitHub (because of the amended commit technically it is not merged). I guess I'll have to look about another way to do the merges as this is really a bit confusing.