RE: [PATCH v5 0/2] Add config file support for l3fwd

["Morten Brørup" <mb@smartsharesystems.com>]

> From: Bruce Richardson [mailto:bruce.richardson@intel.com]
> Sent: Friday, 25 February 2022 11.40
> 
> On Fri, Feb 25, 2022 at 10:36:29AM +0000, Ananyev, Konstantin wrote:
> >
> > > On Thu, Feb 24, 2022 at 02:46:24PM +0100, Thomas Monjalon wrote:
> > > > 24/02/2022 12:06, Ananyev, Konstantin:
> > > > > > > > > > > > > >> Or have a generic library for reading LPM
> entries.  L3fwd is supposed
> > > > > > > > > > > > > >> to be as small as possible (it no longer
> is), and the real work should
> > > > > > > > > > > > > >> be done by libraries to make it easier to
> build other applications.
> > > > > > > > > > > > > >
> > > > > > > > > > > > > > I never heard users ask about such thing,
> > > > > > > > > > > > > > but if there is a demand for that, then I
> suppose it could be considered.
> > > > > > > > > > > > > > CC-ing LPM/FIB maintainers to comment.
> > > > > > > > > > > > > > Though I believe it should be a subject of
> separate patch and discussion
> > > > > > > > > > > > > > (I think many questions will arise - what
> format should be, how to support
> > > > > > > > > > > > > > different types of user-data, to make it
> generic enough, etc.).
> > > > > > > > > > > > >
> > > > > > > > > > > > > Agree, it is very application specific, so it
> could be really difficult
> > > > > > > > > > > > > to make it generic.
> > > > > > > > > > > >
> > > > > > > > > > > > But several other also have LPM tables, so why
> not have common code for other applications.
> > > > > > > > > > > >
> > > > > > > > > > > > examples/l3fwd-power/main.c
> > > > > > > > > > > > examples/ipsec-secgw/rt.c
> > > > > > > > > > > > examples/ip_fragmentation/main.c
> > > > > > > > > > > > examples/l3fwd/l3fwd_lpm.c
> > > > > > > > > > > > examples/ip_reassembly/main.c
> > > > > > > > > > >
> > > > > > > > > > > Ah yes, that's good point.
> > > > > > > > > > > All these examples (except ipsec-secgw) started as
> l3fwd clones,
> > > > > > > > > > > so all of them have hard-coded LPM (and EM) tables
> too.
> > > > > > > > > > > Yes it would be good thing to address that problem
> too,
> > > > > > > > > > > and have some common code (and common routes file
> format) for all of them.
> > > > > > > > > > > I don't know is that a good idea to introduce parse
> file function in LPM/FIB library
> > > > > > > > > > > itself, might be better to  have something like
> examples/common/lpm_parse*.
> > > > > > > > > > > Anyway, this is an extra effort, and I think no-one
> has time for it in 22.03 timeframe.
> > > > > > > > > > > My suggestion would be for 22.03 go ahead with
> current l3fwd patches,
> > > > > > > > > > > then later we can consider to make it common and
> update other examples.
> > > > > > > > > >
> > > > > > > > > > I don't think this patch is urgent.
> > > > > > > > > > I suggest taking time to have common code for all
> examples
> > > > > > > > > > and target a merge in DPDK 22.07.
> > > > > > > > >
> > > > > > > > > Well, yes, from one perspective it not really a
> critical one,
> > > > > > > > > we do live with hard-coded routes inside l3fwd for
> nearly 10 year by now.
> > > > > > > > > Though l3fwd is one of mostly used examples inside DPDK
> and
> > > > > > > > > it is quite a pain to patch/rebuild it each time
> someone needs to run
> > > > > > > > > l3fwd with a different routing table.
> > > > > > > > > Merging this patch will allow people to use l3fwd for
> more realistic test
> > > > > > > > > scenarios in a painless manner.
> > > > > > > > > So I believe this patch is really helpful and should be
> beneficial for the whole community.
> > > > > > > > > Looking from that perspective, I don't see why it has
> to be "all or nothing" attitude here.
> > > > > > > > > Why we can't move one step at a time instead?
> > > > > > > > > That would allow to split and effort in terms of
> development/testing/upstreaming/etc.
> > > > > > > >
> > > > > > > > When a feature is merged, there is less incentives to
> rework.
> > > > > > > > That's why, when a feature is not urgent,
> > > > > > > > it is better to wait for the complete work.
> > > > > > >
> > > > > > > That's true till some extent, though from other side
> > > > > > > even without further rework that patch improves situation
> > > > > > > from what we have right now.
> > > > > > > So I don't see any harm here.
> > > > > >
> > > > > > It is adding a lot of code to an example which is already too
> big.
> > > > > > There are a lot of complain about the size of l3fwd.
> > > > > > That's why I think it makes sense to require this extra code
> > > > > > (not demonstrating anything, but just for testing
> convenience)
> > > > > > outside of the example.
> > > > >
> > > > > Ok, so your main concern is l3fwd code size increase, right?
> > > > > Then would it help if for we'll move file parsing code into a
> separate file(s)
> > > > > (under examples/l3fwd) for now?
> > > > > Something like examples/l3fwd/(lpm_em)_route_parse.c.
> > > >
> > > > Yes it would help to isolate the config file parsing code.
> > > > What others think?
> > > >
> > > I still would like config code for loading an LPM table or FIB
> table to be
> > > put inside the relevant libraries themselves, rather than having it
> in the
> > > examples themselves (even if shared between them).
> >
> > Honestly, I don't see any good reasons for that:
> > I presume users of these libraries already have their own routing
> > config files with their own format requirements and I suppose
> > these formats differ a lot (depending on use-case).

I can confirm this assumption.

Our appliances use our own configuration file handling, which supports full, partial and diff configurations. Furthermore, it uses a schema with type definitions for simple syntax checking, as well as a programmable validator to check system-wide configuration validity and integrity.

> 
> Yes, I agree that all existing users of the libraries will have this in
> place. However, for new users, this may be useful for bootstrapping
> things,
> and it also makes it generally useable for both example apps, and any
> apps
> in the "app" folder, i.e. if l3fwd gets moved there. (Something I agree
> with, as it's now more than just example code).
> 

Don't put any config file parsers or similar in the fast path libraries. It is unwanted bloat!

If you want to standardize on a DPDK specific config file format for the DPDK provided examples and applications, which I do agree with the benefits of having, then provide a separate library to interact with the underlying fast path libraries.