YANG push Integration into Apache Kafka

YANG push Integration
into Apache Kafka
1
Zhuoyao Lin
IETF 118 - NETCONF

TABLE OF CONTENT
I. Introduction
• Motivation
• Overview
II. Problems
• Problems and
proposed solution
• Chosen solution
III. Proposal
• Augmenting the ietf-
yang-library
2

I. INTRODUCTION: Motivation
Challenge of Network Management
Networks are growing and being more and more complex to manage.
A costly task for network operators is to identify and correct network issues.
This involves three steps:
• Detect that there is an issue,
• Identify the cause of issue,
• Correct the issue
Today, it is often the customer that notifies the operator about an issue.
Then the cause detection and correction of an issue is done by expert engineers.
The challenge of the network management industry today, is to automate the detection,
identification and correction of the problem, which is the frame we are trying to achieved.
The project will be deployed into production at the end.
3

I. INTRODUCTION: Motivation
Network Analytics
Nowadays Network Analytics
requires to correlate metrics
from various sources to make
accurate detection/prediction.
There are different sources
and protocols to collect
metrics. Here we focus on the
YANG push protocol.
The metrics are the foundation
to network analytics. We must
have reliable and meaningful
data collection.
Data Mesh
The principle of Data Mesh is to
present the data as a product.
The quality of the data is
guarantee by the team preparing
the data.
In our case, the collected data
comes with the YANG model. 4
Missing Semantics
In the TSDB, YANG model
semantics might be lost. It is not
clear how to interpret the data.
An example: Temperature

YANG push Receiver
Apart from collecting YANG
data, the receiver needs to be
extended to register schema
for YANG subscription.
This is the goal of libyangpush.
YANG Schema Registry
5
A solution to keep the
semantics is to have a schema
registry. Each message in Kafka
contains a schema id pointing
to the original YANG model.
Schema will be used during
serialization and deseralization.
I. INTRODUCTION: Overview

pmacct: A multi-purpose
Network Monitoring
tool(https://github.com/pmac
ct/pmacct)
Schema registry: Confluent
pluggable schema registry has
been extended to natively
support YANG.
YANG push:
1. Subscription to YANG
Notifications for Datastore
Updates - RFC 8641
2. draft-ietf-netconf-udp-notif-
11
It is partially being supported
in the device
I. INTRODUCTION: Overview and Current State

TABLE OF CONTENT
• Motivation
• Overview
II. Problems
• Problems and
proposed solution
• Chosen solution
III. Proposal
yang-library
7
II. Problems

II. PROBLEMS
8
Problem 1:
How to store YANG model in schema
registry?
Problem 2:
How to know the subscribed model?
(On the basis that the YANG push
subscription is providing telemetry
data)
Problem 3:
How to obtain the YANG model and
its dependencies?

II. PROBLEM 1: How to store YANG model in
schema registry?
9
Simulate YANG data structure:
Schema Content <-> YANG model
code
Schema Reference <-> YANG model
dependency
Functionality of schema registry:
Validate YANG model and its
dependencies relationship

II. PROBLEM 2: How to know the subscribed model?
10
Solution 2:
Use YANG to get subscription information:
Subscribed YANG models can be known by parsing fields:
datastore-xpath-filter or datastore-subtree-filter. The
subscription is identified by a sub-id.
This information is exposed as YANG model. It can be obtain:
• As a subscription. Subscription will be synced through the
first push-update, and updated through the push-change-
update(new added, edit, remove etc.)
• Via NETCONF <get> operation. Send a <get> to obtain the
same information.
Solution 1:
Parse namespaces in the first YANG push message
It is possible to find the reverse dependency in this
way.
<subscriptions>
<subscription>
<id>6666</id>
<datastore>ds:operational</datastore>
<datastore-xpath-filter>
/a-module:a
</datastore-xpath-filter>
<encoding>encode-xml</encoding>
<periodic>
<period>30000</period>
</periodic>
</subscription>
</subscriptions>

II. PROBLEM 3: How to obtain the YANG model
and its dependencies?
11
Solution 1:
On-demand Downloading
There are two possible implementation solutions,
each with pros and cons
The idea of on-demand downloading is to send get-
schema request to get the YANG models based on the
model name we obtain. For the main model in the
subscription, we can obtain its name using method in
“identify model”.
Import and include can parsed from the YANG code.
Deviate can be known by subscribing to /ietf-yang-
library:modules-state.
Pros: schema is update-to-date
Cons: Cannot handle augment
Solution 2:
Get-all-schema
The main idea of get-all-schemas is to get all models,
store them in the disk, and analyse their full
dependencies (import, include, deviate and augment)
at the beginning of connection.
Pros: Can handle all dependencies
Cons: Downloading all schemas takes a lot of time

12
Algorithm for finding dependency
Expend on DFS to traverse the YANG model dependency
tree, for the main model: sequentially search for its
import, include, augment and deviate. For its
dependency model, we only search for their import and
include.
The model with the greatest depth will be put into
register list first, then the top level module, in order to
ensure that each register model has the dependency to
refers to in the schema registry.
A hash map is used throughout the traverse with the
index being djb2 hash of the model name to make sure
that the model will not be put into register list twice.
The YANG source can be obtained via NETCONF <get-schema>.
Now we need to obtain the full schema and dependencies.
II. PROBLEM 3: How to obtain the YANG model
and its dependencies?

II. PROBLEMS: An example for the chosen solution
13
module: a-module
+--rw a
+--rw a-instance* [name]
| +--rw name string
| +--rw state? string
+--rw d:y
+--rw d:y-leaf? e:e-enum
The schema for a-module
require to register a-modules, e-
module and d-module.
a-module
d-module
augment
e-module
import
Schema registration Order:
a-module, reference{}
e-module, reference{}
d-module, reference{e-module, a-module}
a-module, reference{d-module, e-module}
<subscriptions>
<subscription>
<id>6666</id>
<datastore>ds:operational</datastore>
<datastore-xpath-filter>
/a-module:a
</datastore-xpath-filter>
<encoding>encode-xml</encoding>
<periodic>
<period>30000</period>
</periodic>
</subscription>
</subscriptions>

TABLE OF CONTENT
I. Introduction
• Motivation
• Overview
III. Proposal
yang-library
14
II. Problems
• Problems and
proposed solution
• Chosen solution

III. Proposal: Augmenting YANG model
ietf-yang-library
15
Problem 3:
How to obtain the YANG model and its dependencies?
Explanation:
As for the current chosen solution, we cannot invalidate
the model stored in disk when they have been updated
in the device. As a way to get the most up-to-date
reverse dependency model, the on-demand
downloading only support checking for deviations
currently. However, it is reasonable to augment to make
the ietf-yang-library to also support storing the
augmentations.
In this way, we can get rid of having to use the get-all-
schemas solution, while we can also handle the reverse
dependencies easily.

IV. Conclusion & Future Work
16
Demo is presented during IETF117 Hackathon:
https://wiki.ietf.org/en/meeting/117/hackathon#net
work-telemetry-yang-push-integration-into-apache-
kafka
Demonstrate the interaction with a working YANG
schema registry during the demo
Github repository for libyangpush:
https://github.com/network-analytics/libyangpush.git
Outcomes
• Deploy the functionalities into Swisscom lab.
Future Work
• Integration with pmacct
• Test with device as YANG push becomes
better supported
• Augment YANG model ietf-yang-library to
support the record of augmentations(to
support the on-demand downloading)
Gaps to fill
Industry Post:
https://github.com/graf3net/draft-daisy-kafka-yang-integration/blob/main/draft-daisy-kafka-
yang-integration-05.md
https://www.linkedin.com/pulse/network-analytics-ietf-115-london-thomas-graf/
https://www.linkedin.com/pulse/network-analytics-ietf-116-yokohama-thomas-graf/
https://www.linkedin.com/pulse/network-analytics-ietf-117-san-francisco-thomas-graf/

YANG push Integration into Apache Kafka

Recommended

More Related Content

Similar to YANG push Integration into Apache Kafka (20)

More from ThomasGraf42 (20)

Recently uploaded (20)

YANG push Integration into Apache Kafka