Hi Gyan,
These use cases may be very reasonable and important, but TCP is complex and there are a lot of subtle issues once one looks behind the scenes.
Gyan> Understood. In these particular use cases we do have to look in detail behind the scenes.
Before actually writing corresponding YANG models, the relevant players would have to speak up, e.g., in TCPM, and come up with specific proposals. As far as I can tell, this has not happened so far.
Gyan> I think from the OPSDIR review POV as it relates to Routing area operations we would like to have some concrete follow up from TCPM on this topic. I understand the goal of this yang model however if we are able to expand the goal beyond the TCP MIB to incorporate some requirements would that be possible? We would have to get the relevant players in TCPM to speak as you stated or would the authors of this draft be willing to take on the new work?
I have to emphasize once again that draft-ietf-tcpm-yang-tcp does not prevent further YANG models. The document actually states that pretty explicitly.
Gyan> Understood. There is a chance do to priorities that the future Yang model may not come to fruition. Also I think if we expand the scope of this draft to encompass what myself and Robert are asking, I think it would make the draft much much more useful to all.
Michael
From: Gyan Mishra <hayabusagsm@xxxxxxxxx>
Sent: Monday, July 4, 2022 10:11 PM
To: Robert Raszuk <robert@xxxxxxxxxx>
Cc: Last Call <last-call@xxxxxxxx>; Scharf, Michael <Michael.Scharf@xxxxxxxxxxxxxxx>; draft-ietf-tcpm-yang-tcp.all@xxxxxxxx; ops-dir@xxxxxxxx; tcpm@xxxxxxxx; touch@xxxxxxxxxxxxxx
Subject: Re: [Last-Call] [tcpm] Opsdir telechat review of draft-ietf-tcpm-yang-tcp-07
Hi Michael
A possible good example of a use case by router vendors of use of the detailed visibility into the TCP socket in the Yang model is an issue that has caused outages across the internet related to BGP TCP O window where the receive window was stuck state and could not write to the receive buffer and so the BGP session remained in UP state resulting in a major internet outage.
Operators are now moving towards BGP based MSDC for massive scalability and no IGP (OSPF or ISIS) for scalability and stability. As a result of the motivation and change operationally towards BGP, TCP and all the socket details is now that much more important to operators as well as now an significant interest to most vendors.
As well with micro services and Kubernetes with the data center fabric being moved to compute nodes running hundreds of BGP sessions.
That is the POV we are coming from related to the inner workings and details of TCP Yang model now applies to router and switch vendors but as well also compute nodes.
Regards
Gyan
On Mon, Jul 4, 2022 at 3:52 PM Robert Raszuk <robert@xxxxxxxxxx> wrote:
Hi Michael,
Actually I used the URG flag example as this is used by one of the key features in one of the major vendor's OS. Ability to see this flag to be reported is IMO useful in this very application.
> on-path middleboxes.
That is not my concern at all. My focus is to use YANG on the endpoints to avoid need to recreate TCP state via TAP captures. Like some of the good analyzers allow you to do.
> many OS kernels don’t use YANG at all.
True - but is this the right argument ? Those will not benefit irrespective of how small or big YANG model will be shipped.
> One could write a lot in a YANG model, but who would actually implement that?
I would count on network vendors to implement it. And that is my personal area of interest. Otherwise I would not care to comment :)
Thx,
R.
On Mon, Jul 4, 2022 at 9:39 PM Scharf, Michael <Michael.Scharf@xxxxxxxxxxxxxxx> wrote:
Hi Robert,
the TCP Urgent Flag is discussed in RFC 6093 and probably not a good example for a TCP-feature relevant for modern applications (RFC 6093 stated more than 10 years ago “new applications SHOULD NOT employ the TCP urgent mechanism”).
A modern TCP implementation actually has several windows and running TCP code either measures them in bytes or in segments. That results in quite some differences. So, even for TCP windows there is no simple way to model the actual behavior of widely deployed running code.
And the algorithms of a modern TCP stack can imply more than 100 parameters. Due to the complexity it is basically impossible to draw the line between “elementary” parameters and implementation-specific ones.
All that was discussed in TCPM, and the WG consensus was not to boil the ocean. The very narrow scope of draft-ietf-tcpm-yang-tcp is a result of that discussion in TCPM. I have tried my best to explain the rationale inside the document.
It may be possible to publish a more comprehensive TCP YANG model as a follow-up specification. But the first step would be to convince TCPM that this is feasible and that relevant stacks would indeed implement that YANG model.
Michael
From: Robert Raszuk <robert@xxxxxxxxxx>
Sent: Monday, July 4, 2022 9:15 PM
To: Scharf, Michael <Michael.Scharf@xxxxxxxxxxxxxxx>
Cc: touch@xxxxxxxxxxxxxx; Gyan Mishra <hayabusagsm@xxxxxxxxx>; Last Call <last-call@xxxxxxxx>; draft-ietf-tcpm-yang-tcp.all@xxxxxxxx; ops-dir@xxxxxxxx; tcpm@xxxxxxxx
Subject: Re: [Last-Call] [tcpm] Opsdir telechat review of draft-ietf-tcpm-yang-tcp-07
Hi,
> Any application can decide to configure TCP parameters as far as possible in the given operation
> system, e.g., via the sockets API. That is orthogonal to the internals of the TCP implementation and the TCP protocol.
While clients running on top of TCP can configure its parameters I would at least expect to be able to report such values (local and remote) when using the TCP YANG model. For example I can not find the Urgent Flag in the current YANG model. Same for elementary window size of any given connection, same for connection duration, .
Inability to do so to me sounds like a half baked model. IMHO it is not ready to be even declared as MVP.
Many thx,
Robert
On Mon, Jul 4, 2022 at 6:06 PM Scharf, Michael <Michael.Scharf@xxxxxxxxxxxxxxx> wrote:
Joe, all,
„separate protocol specific YANG model” could be the YANG model for BGP, or for any other TCP-based application.
Any application can decide to configure TCP parameters as far as possible in the given operation system, e.g., via the sockets API. That is orthogonal to the internals of the TCP implementation and the TCP protocol. The app configuration can be done in YANG or by other means. For the TCP stack, that does not matter.
As far as I understand Gyan, the concerns regarding draft-ietf-tcpm-yang-tcp are sorted out already.
@all: Please speak up if specific changes are needed in draft-ietf-tcpm-yang-tcp. The authors will have to focus on the IESG feedback.
Thanks
Michael
From: touch@xxxxxxxxxxxxxx <touch@xxxxxxxxxxxxxx>
Sent: Monday, July 4, 2022 4:38 PM
To: Gyan Mishra <hayabusagsm@xxxxxxxxx>
Cc: Last Call <last-call@xxxxxxxx>; draft-ietf-tcpm-yang-tcp.all@xxxxxxxx; ops-dir@xxxxxxxx; tcpm@xxxxxxxx
Subject: Re: [tcpm] Opsdir telechat review of draft-ietf-tcpm-yang-tcp-07
—
Dr. Joe Touch, temporal epistemologist
On Jul 3, 2022, at 10:16 PM, Gyan Mishra <hayabusagsm@xxxxxxxxx> wrote:
Hi Joe, authors et all
I reviewed the feedback from my earlier review in March and as this model is geared towards BGP primary.
To address all of my concerns would be complicated for this Yang model, so the plan is that a separate protocol specific yang model would be a follow on to address all of my concerns.
First, there should NEVER be two different YANG models for BGP routers vs. other routers or hosts. TCP is TCP is TCP. If that is an assumption for moving this document forward, TCPM should have a longer discussion about that point specifically.
Second, my observations about your requests below stand, regardless of when/where current or future authors might be considering them.
Joe
On Mon, Jul 4, 2022 at 12:44 AM touch@xxxxxxxxxxxxxx <touch@xxxxxxxxxxxxxx> wrote:
FWIW:
> On Jul 3, 2022, at 9:04 PM, Gyan Mishra via Datatracker <noreply@xxxxxxxx> wrote:
>
> Reviewer: Gyan Mishra
> Review result: Not Ready
>
> This draft provides the Yang data mode for TCP.
>
> The draft is well written and is almost ready publication. I verified the FSM
> state machine and all states are listed.
>
> Minor issues:
> None
>
> Major issues:
> None
>
> Nits:
> I reviewed the TCP Yang data model and has a question related to the FSM state
> machine.
>
> Would it be possible to specify the TCP Header flags SYN, FIN, ACK, RST of BFD
> FSM finite state machine Events and Transition. I think this would be very
> helpful for the TCP Yang model FSM state machine. For each state you could
> specify the flags set.
These issues appear to have been raised by you in March during last call review. Some have been addressed by others before; I’ll add my input.
The YANG model represents information about the current TCP connection. It is not (and should not be confused with) a specification of the protocol.
Further, flags are associated with messages that cause state transitions, not states (i.e., the FSM is a Mealy machine, not a Moore machine). There is no “flags set for each state”.
> http://tcpipguide.com/free/t_TCPOperationalOverviewandtheTCPFiniteStateMachineF-2.htm
That page has errors and is not consistent with RFC793 (or it’s pending -bis update). E.g., FIN stands for “finis” (latin for “end”), not “finish”.
> I think the TCP TCB (TCP Control Block) is missing in the Yang model. This is
> important for troubleshooting TCP connection state.
RFC793 (and -bis) indicate that the STATUS command, which might return similar information, is optional.
If there is connection information returned, I do not think it should be the TCB; that is an implementation-dependent parameter, not a universal property of TCP connections. As others have stated in previous responses to you review, the common subset of the TCB is already contained.
I.e., I think the YANG model represents TCP information. It is not - and should not be confused with - a troubleshooting tool.
Joe--
_______________________________________________
tcpm mailing list
tcpm@xxxxxxxx
https://www.ietf.org/mailman/listinfo/tcpm
--
last-call mailing list
last-call@xxxxxxxx
https://www.ietf.org/mailman/listinfo/last-call--
-- last-call mailing list last-call@xxxxxxxx https://www.ietf.org/mailman/listinfo/last-call
