Re: [nft PATCH 0/5] Enhance cache filter for list commands

[Phil Sutter <phil@nwl.cc>]

On Wed, Mar 11, 2026 at 11:06:52AM -0400, Eric Garver wrote:
> On Wed, Mar 11, 2026 at 12:11:10AM +0100, Phil Sutter wrote:
> > Reducing the amount of data fetched from kernel improves performance
> > with large rule sets but also reduces adverse side-effects if multiple
> > versions of nftables access the same kernel rule set. Being able to
> > ignore parts of the rule set one is not interested in allows for (more or
> > less) safe coexistence if each tool is operating on the data it created
> > itself only.
> > 
> > This series reduces caching for list commands which specify a family
> > and/or table. To help testing this, patch 1 extends netlink debug output
> > to include chains, flowtables and objects so a test case may check if
> > they are fetched or not.
> > 
> > The remaining patches actually increase filter use.
> > 
> > Phil Sutter (5):
> >   cache: Include chains, flowtables and objects in netlink debug output
> >   cache: Respect family in all list commands
> >   cache: Relax chain_cache_dump filter application
> >   cache: Filter for table when listing sets or maps
> >   cache: Filter for table when listing flowtables
> > 
> >  src/cache.c                                 | 11 ++--
> >  src/mnl.c                                   | 60 ++++++++++++++++++---
> >  tests/shell/testcases/listing/cache_filters | 53 ++++++++++++++++++
> >  3 files changed, 113 insertions(+), 11 deletions(-)
> >  create mode 100755 tests/shell/testcases/listing/cache_filters
> 
> I ran this series against the firewalld testsuite. All green.
> Thanks Phil!

Thanks for testing, Eric! Shame on me for not putting you in Cc as you
asked for. Next time I'll probably best add a Cc: tag to one of the
commits immediately. ;)

Cheers, Phil