Re: [Dots] DOTS telemetry Issues picked up in Interop Testing

[[e-mail]]

Hi Jon,

Please see inline.

Cheers,
Med

De : Jon Shallow [mailto:[e-mail]]
Envoyé : mardi 21 avril 2020 10:59
À : BOUCADAIR Mohamed TGI/OLN; [e-mail]
Objet : RE: [Dots] DOTS telemetry Issues picked up in Interop Testing

Hi all,

A further thought on the use of Uri-Queries to clarify the AND/OR usage.

If you only allow one query per query type and put the match list in an array, then this will be an OR of the array list (the same as we do for the target* definitions right now.  E.G. :-

Uri-Query: target_prefix=[1.2.3.4/32,4.3.2.1/32]
Gives either 1.2.3.4 or 4.3.2.1 as a valid match.

And
Uri-Query: target-prefix=[1.2.3.4/32,4.3.2.1/32]
Uri-Query: lower-port=[80,443]
Gives (either 1.2.3.4 or 4.3.2.1) and (either port 80 or 443)

[] should not include spaces and comma used as a separator.

[Med] The issue I have with this is that we will need to handle cases where both lower-port and upper-port are present. Not sure what would be the benefit of allowing multiple key values, compact uris? If that's a concern, we may consider shortened names in the query (e.g., s/target-prefix/tp, s/lower-port/lp, ..).

Not sure about :-  If [] are not used, then it should be treated as a single item to match.

The list of valid Uri-Queries may need to be extended.

A Uri-Query option can include the following
   parameters: target-prefix, lower-port, upper-port, target-protocol,
   target-fqdn, target-uri, alias-name.

should include mid and content ('c').
[Med] OK. Please note that 'c' wasn't listed because we do have the following:

  The DOTS server follows the same considerations discussed in
   Section of 4.5.3 of [I-D.ietf-dots-signal-channel] for managing DOTS
   telemetry configuration freshness and notification.  Likewise, a DOTS
   client may control the selection of configuration and non-
   configuration data nodes when sending a GET request by means of the
   'c' Uri-Query option and following the procedure specified in
   Section of 4.4.2 of [I-D.ietf-dots-signal-channel].  These
   considerations are not re-iterated in the following sections.

I added a pointer to that section.

  Do we need to be able to filter on source attributes as well?
[Med] What would be the usage? Is it the case of an administrator that is aware (using some means) that some sources are involved an attacks on other networks, then sends a request to its server to filter telemetry data bound to these source? Filtering in this case may be problematic as some attacks may be observed but are hidden by the filter on the source. Or do we want to focus on very few talkers? For example, the client would instruct the server to send data only related to the top-talker, 3 first top-talkers, etc.?

Regards

Jon



From: Dots [mailto: [e-mail]] On Behalf Of Jon Shallow
Sent: 20 April 2020 14:47
To: [e-mail]; [e-mail]
Subject: Re: [Dots] DOTS telemetry Issues picked up in Interop Testing

Hi Med,

See inline.

Regards

Jon

From: Dots [mailto: [e-mail]] On Behalf Of [e-mail]
Sent: 20 April 2020 13:54
To: Jon Shallow; [e-mail]
Subject: Re: [Dots] DOTS telemetry Issues picked up in Interop Testing

Hi Jon,

Thank you for sharing the comments.

Please see

De : Dots [mailto:[e-mail]] De la part de Jon Shallow
Envoyé : lundi 20 avril 2020 14:07
À : [e-mail]
Objet : [Dots] DOTS telemetry Issues picked up in Interop Testing

Hi All,

1)
   DOTS agents MUST NOT send pre-or-ongoing-mitigation telemetry
   messages to the same peer more frequently than once every 'telemetry-
   notify-interval' (Section 6.1).

If I do a PUT /tm-setup, followed immediately by a GET /tm-setup, should the GET fail based on the telemetry-notify-interval, or should telemetry-notify-interval apply individually to PUT. GET and DELETE?
[I may want to do a DELETE (no tsid) followed by a GET to get suitable max/min information and telemetry-notify-interval can easily be more than 1 second. ]

[Med] This rate limit does not apply to these messages; it applies only for notifications;

==
        leaf telemetry-notify-interval {
          type uint32 {
            range "1 .. 3600";
          }
          must '. >= ../../min-config-values/telemetry-notify-interval' {
            error-message
              "The value must be greater than or equal
               to the telemetry-notify-interval in the min-config-values";
          }
          units "seconds";
          description
            "Minimum number of seconds between successive
             telemetry notifications.";
        }
===

We can make the text clear:

OLD:

   DOTS agents MUST NOT send pre-or-ongoing-mitigation telemetry

   messages to the same peer more frequently than once every 'telemetry-

   notify-interval' (Section 6.1<https://tools.ietf.org/html/draft-ietf-dots-telemetry-07#section-6.1>).  If a telemetry notification is sent

   using a block-like transfer mechanism (e.g.,

   [I-D.bosh-core-new-block<https://tools.ietf.org/html/draft-ietf-dots-telemetry-07#ref-I-D.bosh-core-new-block>]), this rate limit policy MUST NOT consider

   these individual blocks as separate notifications, but as a single

   notification.

NEW:

   DOTS agents MUST NOT send pre-or-ongoing-mitigation telemetry

   notifications to the same peer more frequently than once every 'telemetry-

   notify-interval' (Section 6.1<https://tools.ietf.org/html/draft-ietf-dots-telemetry-07#section-6.1>).  If a telemetry notification is sent

   using a block-like transfer mechanism (e.g.,

   [I-D.bosh-core-new-block<https://tools.ietf.org/html/draft-ietf-dots-telemetry-07#ref-I-D.bosh-core-new-block>]), this rate limit policy MUST NOT consider

   these individual blocks as separate notifications, but as a single

   notification.

Jon> This change works for me.

2)
Use of multiple Uri-Query:

Uri-Query: target_prefix=[1.2..3.4]
Uri-Query: target_prefix=[2.3..4.5]
I think this should be OR
[Med] Yes, this is the same interpretation when we include target-prefix in the message body.

Uri-Query: target_prefix=[1.2..3.4]
Uri-Query: target_port=[80]
I think this should be an AND.
[Med] Yes, because the target-port is specifying a subset of ports bound to the target-prefix.

Perhaps we should have operator queries
Uri-Query: target_prefix=[1.2..3.4]
Uri-Query: op=OR
Uri-Query: target_prefix=[2.3..4.5]

Uri-Query: target_prefix=[1.2..3.4]
Uri-Query: op=AND
Uri-Query: target_port=[80]
[Med] I don't think this is needed. We don't specify the operator when we include the target clauses in the message body. The same rules apply. We can clarify this in the text If you think this is needed.

Jon> I think that it is worth stating that the same rules apply for clarification.

Thoughts?

3)
What happens if a Query cannot be supported.

E.g. Uri-Query: target_port=[80] and statistics on the server by a port basis  is not supported.  Do we want to either error out with a 4.0X responses, or should we include a (new additional) YANG body response indicating which Uri-Query is not supported?

[Med] Good point. The client can retrieve during telemetry setup the supported query types + server replies with 4.00 when an invalid query type is used.

Jon> I think the server stating what Queries are supported is a good idea.

Regards

Jon