[Bug 210007] Review Request: libtune - standard API to access the kernel tunables

bugzilla at redhat.com bugzilla at redhat.com
Thu Feb 14 16:01:27 UTC 2008


Please do not reply directly to this email. All additional
comments should be made in the comments box of this bug report.

Summary: Review Request: libtune - standard API to access the kernel tunables


https://bugzilla.redhat.com/show_bug.cgi?id=210007





------- Additional Comments From nadia.derbey at bull.net  2008-02-14 11:01 EST -------
(In reply to comment #73)
> So.
> 
> This code is a *really bad idea*.
> 
> Apps should not be twiddling with tunables in /proc or /sys - they are system
> wide attributes. 

Agreed, but many applications need some tunables to be set higher than the
default values. So, today, there must be administrators doing an "echo xxx >
/proc/kernel/sys/yyy", either by hand or via a script.
This is only what the libtune is intended to do.

> You can have multiple apps - what if they want different things?

This is a related problem which libtune was not made to solve. Applications made
to solve this problem in different ways could use libtune. System administrators
could then choose between a variety of potentially competing applications or
manual tuning.

Then, if a single such application emerges libtune would be a good place to add
that without worrying about changing every app that uses the tunable. Though
this does assume changing most or all existing apps to use the library initially.

> This library doesn't actually prevent you from situations where the kernel
> changes without patching and rebuild, so you don't gain anything from an app
> standpoint.

Yes, in that case the libtune package must be patched and rebuilt. If the
tunable change is simple enough the advantage is the apps that depend on that
tunable shouldn't need to be patched and rebuilt themselves -- libtune insulates
them against most tunable interface changes.

> It's never the sort of thing which will be in Fedora base, or RHEL,
> and therefore any app that would want to use it would need ot have code to
> handle it not being there anyway. So, what's the point?
> 

 Yes, if it got into only one then there'd be little point. However isn't
it safer to have a migration path? There'd be a period where an app
would use both libtune and its own code. Then, once libtune is in both
Fedora and RHEL, the application-specific code could be dropped.
Alternatively, if libtune were thrown out then use of libtune could be
reverted without breaking things. Without this kind of migration you
almost have a chicken-and-the-egg paradox.

-- 
Configure bugmail: https://bugzilla.redhat.com/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the QA contact for the bug, or are watching the QA contact.




More information about the Fedora-package-review mailing list