[SR-Users] RFC: rating - ranking system for community

[Henning Westerholt]

Am Donnerstag, 14. März 2019, 14:19:22 CET schrieb Alex Balashov:
> Perhaps a component-orientated view is not the correct one here. It
> almost sounds like what is being sought is a kind of "cookbook of
> Kamailio patterns"[1], if you like. This answers a lot of questions
> that also capture preferences about third-party FOSS componentry, such
> as:
> 
> "Should I do HA failover with VRRP or Heartbeat?"
> 
> "Is Kamailio + PostgreSQL a stable combination for a high-volume
> registrar?"
> 
> "Are there pitfalls to DMQ dialog replication?"

Hi Alex,

to continue on this idea - the "big complex enterprise software vendors" (e.g. 
SAP) provides so called "value maps". They provide a way how to plan, deploy 
and operation a certain technology stack. It starts with a general knowledge 
base and then continue with deployment best-practices and guidelines on 
certain tasks or events.

Exactly as you described:

> "What is the best way to build a load balancer with failover and gateway
> monitoring?"

I also think this approach would be a better fit as just a rating system. That 
said, we probably could flag the existing modules a bit better (there is some 
status on the main overview page, but it is not visible e.g. in the README). 
https://kamailio.org/docs/modules/5.2.x/

They have been some more comments in favor of the cookbook approach. Is there 
somebody from the community interested in continuing on this, maybe creating a 
first draft document or something in this regards?

> [1] We use the term "cookbook" differently in this project, but this
> usage is closer to the commonplace conventional one.
> 
> On Thu, Mar 14, 2019 at 08:35:15AM -0400, Alex Balashov wrote:
> > Hi,
> > 
> > A few off-the-cuff thoughts, in no particular order:
> > 
> > 1. Kamailio does have hundreds of modules of all kinds, and some sort of
> > guide for which ones to use when, or some other schema which would serve
> > to provide some conceptual organisation for the modules, is probably a
> > desirable documentation objective.
> > 
> > 2. I am nevertheless wary of any system which purports to "rank" or
> > "rate" components.
> > 
> > One obvious reason is that popularity isn't a very good metric for
> > whether something is appropriate for or applicable to a particular
> > purpose. There is a reason the expression "degenerated into a popularity
> > contest" exists in our industry. This is all the more true given
> > (arguably) Kamailio's somewhat unique status as a "toolkit" or a
> > "framework" for building certain kinds of systems and services; it means
> > some of the most useful components in many scenarios might be either
> > obscure, or so broadly general that a highly fact-dependent /
> > situation-specific logic of its relationship to a given scenario is hard
> > to tease out. Would you "recommend" the TM module? Is it widely used?
> > 
> > :-)
> > 
> > This goes to another, more general problem with ranking components, and
> > that is that the module ecosystem is a wildly eclectic bag of things of
> > very different scope. Without a more rigid preconceived taxonomy to
> > create the right mental categories, meaningful comparisons between
> > modules are difficult to draw, as would be theoretically required for
> > some kind of ranking or any system which purports to raise the profile
> > of some components over others.
> > 
> > Kamailio modules fall into at least a few identifiable categories--this
> > is just a half-hearted improvisational stab at it, and probably not very
> > nuanced:
> > 
> > a. "Essential" / "core" - while these are technically modules, their
> > feature set is so universal and critical to most non-trivial Kamailio
> > implementations that they are substantially equivalent to "core"
> > functionality. Modules like 'rr', 'tm' and 'pv' clearly fall into this
> > category. One cannot really do anything worthy of remark with Kamailio
> > without them, excluding some exotic cases.
> > 
> > b. Broad and holistic systems - complete categories of broad SIP server
> > functionality implemented in more or less one module (usrloc +
> > registrar, presence/pua, auth*, dispatcher, etc.);
> > 
> > c. Dependencies of other higher-level modules - the relationship of
> > 'usrloc' to 'registrar' is suitably described by this; if you're using
> > 'usrloc', you're almost certainly using 'registrar', and there are
> > really no meaningful standalone uses of 'usrloc', nor does it expose any
> > route script functions. At the same time, the list of things one may
> > wish to interrogate via RPC/management interfaces are split between
> > 'usrloc' and 'registrar' in ways that make sense to a programmer but
> > which users might find arbitrary. In any case, do you give 'usrloc' a
> > "thumbs up"? What does that even mean? :-)
> > 
> > Some modules are hybrids; they have some standalone functionality but
> > are most commonly inputs into a higher-level usage, such as 'xhttp' and
> > its relationship to 'jsonrpcs', or the various presence_* or pua_*
> > subcomponents, some of the auth_* components, ims_*, etc.
> > 
> > d. Pure API dependencies of other modules - these expose internal APIs
> > used by other modules and provide no standalone functionality others.
> > One can debate whether 'usrloc' falls into this category, but certainly,
> > 'dmq' is a good example of this, as are the various 'db_*' connectors,
> > and maybe 'keepalive'.
> > 
> > e. Niche programmatic functionality - 'htable', 'cfgutils', 'jansson',
> > 'async', etc. This category entails additional programmatic constructs
> > relevant to the "software engineering" aspect of writing config/route
> > script or how processing is done, but do not provide any external
> > services or interfaces as such.
> > 
> > f. Broadly environmental - subtly alter the overall way that SIP
> > messages are processed, or add support for new kinds of transports, etc.
> > This would include 'tls', 'ws', 'outbound', and the 'topo{h,s}' modules.
> > 
> > g. Language / API / interpreter connectors - 'app_*' modules.
> > 
> > h. Highly specific functionality - most other modules fall into this.
> > But some of the functionality is fairly high-level, e.g. 'rtpengine',
> > while others are more low-level and closer to category E, such as
> > 'mqueue'.
> > 
> > 
> > Anyway, the point here is that I don't think a vehicle to provide
> > community guidance about which components to use, or which would purport
> > to rank them somehow, can be meaningfully separated from the equally
> > vital need to create some kind of taxonomy of modules or, more
> > generally, adequate categories of thought within which such
> > conversations should take place. Kamailio documentation lacks a clear
> > and distinct "metaphysics" in this sense.
> > 
> > -- Alex
> > 
> > On Thu, Mar 14, 2019 at 09:55:41AM +0100, Daniel-Constantin Mierla wrote:
> > > Hello,
> > > 
> > > starting to continue the discussions here on mailing lists about some of
> > > the approached topics during the last IRC meeting -- there will be
> > > couple of them.
> > > 
> > > The one now is about finding and deploying a ranking/rating system that
> > > could eventually help community members to make decisions easier in
> > > regard to what components to use in their voip systems.
> > > 
> > > The need was exemplified by user verticelo with the dificulty to decide
> > > what KEMI scripting language to use. While maintaining all the app_xyz
> > > are expected to be easy, at least I know that for those I created, the
> > > arguments were also from the perspective of the community. Like what
> > > others are using, so one can expect good assistance, hints and share of
> > > knowledge via community forums. This can be also extended to related
> > > tools, like what people use for shared IP high availability of kamailio,
> > > preferred database servers, ...
> > > 
> > > This email is to ask if those that didn't participate to IRC devel
> > > meeting find such system useful and, if there is a positive feedback, is
> > > anyone aware of some OSS that we can deploy on kamailio.org for such
> > > purpose? It should be something that allows posting a topic (title and
> > > short description) along with a list of answers/options (each can be
> > > again like a title and short description) and provide a way to rate the
> > > options (like, thumbs up, starts, ...), eventually allowing also
> > > comments for each option. I guess it sounds a bit like stackoverflow ...
> > > 
> > > Being users related, the email is sent only to sr-users, let's keep the
> > > discussion on this mailing list.

Cheers,

Henning

-- 
Henning Westerholt - https://skalatan.de/blog/
Kamailio services - https://skalatan.de/services
Kamailio security assessment - https://skalatan.de/de/assessment