Core Components of Policy-based Telemetry Streaming

The core components used in streaming policy-based telemetry data are:

Telemetry Policy File

A telemetry policy file is defined by the user to specify the kind of telemetry data that is generated and pushed to the receiver. The policy must be stored in a text file with a .policy extension. Multiple policy files can be defined and installed in the /telemetry/policies/ folder in the router file system.

A policy file:
  • Contains one or more collection groups; a collection group includes different types of data to be streamed at different intervals

  • Includes a period in seconds for each group

  • Contains one or more paths for each group

  • Includes metadata that contains version, description, and other details about the policy

Policy file syntax

The following example shows a sample policy file:

{
"Name": "NameOfPolicy",
 "Metadata": {
     "Version": 25,
     "Description": "This is a sample policy to demonstrate the syntax",
     "Comment": "This is the first draft",
     "Identifier": "<data that may be sent by the encoder to the mgmt stn"
},
 "CollectionGroups": {
     "FirstGroup": {
         "Period": 10,
         "Paths": [
             "RootOper.MemorySummary.Node",
             "RootOper.RIB.VRF",
             "..."
         ]
     },
     "SecondGroup": {
         "Period": 300,
         "Paths": [
             "RootOper.Interfaces.Interface"
         ]
     }
 }
}
The syntax of the policy file includes:
  • Name the name of the policy. In the previous example, the policy is stored in a file named NameOfPolicy.policy. The name of the policy must match the filename (without the .policy extension). It can contain uppercase alphabets, lower-case alphabets, and numbers. The policy name is case sensitive.

  • Metadata information about the policy. The metadata can include the version number, date, description, author, copyright information, and other details that identify the policy. The following fields have significance in identifying the policy:
    • Description is displayed in the show policies command.

    • Version and Identifier are sent to the receiver as part of the message header of the telemetry messages.

  • CollectionGroups an encoder object that maps the group names to information about them. The name of the collection group can contain uppercase alphabets, lowercase alphabets, and numbers. The group name is case sensitive.

  • Period the cadence for each collection group. The period specifies the frequency in seconds at which data is queried and sent to the receiver. The value must be within the range of 5 and 86400 seconds.

  • Paths one or more schema paths, allowed list entries or native YANG paths (for a container) for the data to be streamed and sent to the receiver. For example,

    Schema path:
    RootOper.InfraStatistics.Interface(*).Latest.GenericCounters
    YANG path:
    /Cisco-IOS-XR-infra-statsd-oper:infra-statistics/interfaces/interface=*/latest/generic-counters
    Allowed list entry:
    "RootOper.Interfaces.Interface(*)": 
    {
                   "IncludeFields": ["State"]
    }

Schema Paths

A schema path is used to specify where the telemetry data is collected. A few paths are listed in the following table for your reference:

Table 1. Schema Paths

Operation

Path

Interface Operational data

RootOper.Interfaces.Interface(*)

Packet/byte counters

RootOper.InfraStatistics.Interface(*).Latest.GenericCounters

Packet/byte rates

RootOper.InfraStatistics.Interface(*).Latest.DataRate

IPv4 packet/byte counters

RootOper.InfraStatistics.Interface(*).Latest.Protocol(['IPV4_UNICAST'])

MPLS stats

  • RootOper.MPLS_TE.Tunnels.TunnelAutoBandwidth
  • RootOper.MPLS_TE.P2P_P2MPTunnel.TunnelHead
  • RootOper.MPLS_TE.SignallingCounters.HeadSignallingCounters

QOS Stats

  • RootOper.QOS.Interface(*).Input.Statistics
  • RootOper.QOS.Interface(*).Output.Statistics

BGP Data

RootOper.BGP.Instance({'InstanceName': 'default'}).InstanceActive.DefaultVRF.Neighbor([*])

Inventory data

RootOper.PlatformInventory.Rack(*).Attributes.BasicInfo RootOper.PlatformInventory.Rack(*).Slot(*).Card(*).Sensor(*).Attributes.BasicInfo

Telemetry Encoder

The telemetry encoder encapsulates the generated data into the desired format and transmits to the receiver.

An encoder calls the streaming Telemetry API to:
  • Specify policies to be explicitly defined

  • Register all policies of interest

Telemetry supports two types of encoders:
  • JavaScript Object Notation (JSON) encoder

    This encoder is packaged with the IOS XR software and provides the default method of streaming telemetry data. It can be configured by CLI and XML to register for specific policies. Configuration is grouped into policy groups, with each policy group containing one or more policies and one or more destinations. JSON encoding is supported over only TCP transport service.

    JSON encoder supports two encoding formats:
    • Restconf-style encoding is the default JSON encoding format.

    • Embedded-keys encoding treats naming information in the path as keys.

  • Google Protocol Buffers (GPB) encoder

    This encoder provides an alternative encoding mechanism, streaming the data in GPB format over UDP or TCP. It can be configured by CLI and XML and uses the same policy files as those of JSON. Additionally, a GPB encoder requires metadata in the form of compiled .proto files to translate the data into GPB format.

    GPB encoder supports two encoding formats:
    • Compact encoding stores data in a compact GPB structure that is specific to the policy that is streamed. This format is available over both UDP and TCP transport services. A .proto file must be generated for each path in the policy file to be used by the receiver to decode the resulting data.

    • Key-value encoding stores data in a generic key-value format using a single .proto file. The encoding is self-describing as the keys are contained in the message. This format is available over UDP and TCP transport service. A .proto file is not required for each policy file because the receiver can interpret the data.

TCP Header

Streaming data over a TCP connection either with a JSON or a GPB encoder and having it optionally compressed by zlib ensures that the stream is flushed at the end of each batch of data. This helps the receiver to decompress the data received. If data is compressed using zlib, the compression is done at the policy group level. The compressor resets when a new connection is established from the receiver because the decompressor at the receiver has an empty initial state.

Header of each TCP message:

Type

Flags

Length

Message

4 bytes

4 bytes

  • default - Use 0x0 value to set no flags.

  • zlib compression - Use 0x1 value to set zlib compression on the message.

4 bytes

Variable

where:
  • The Type is encoded as a big-endian value.

  • The Length (in bytes) is encoded as a big-endian value.

  • The flags indicates modifiers (such as compression) in big-endian format.

  • The message contains the streamed data in either JSON or GPB object.

Type of messages:

Type

Name

Length

Value

1

Reset Compressor

0

No value

2

JSON Message

Variable

JSON message (any format)

3

GPB compact

Variable

GPB message in compact format

4

GPB key-value

Variable

GPB message in key-value format

JSON Message Format

JSON messages are sent over TCP and use the header message described in TCP Header.

The message consists of the following JSON objects:

{
"Policy": "<name-of-policy>",
"Version": "<policy-version>",
"Identifier": "<data from policy file>"
"CollectionID": <id>,
"Path": <Policy Path>,  
"CollectionStartTime": <timestamp>,
"Data": { … object as above … },
"CollectionEndTime": <timestamp>,
}
where:
  • Policy, Version and Identifier are specified in the policy file.

  • CollectionID is an integer that allows messages to be grouped together if data for a single path is split over multiple messages.

  • Path is the base path of the corresponding data as specified in the policy file.

  • CollectionStartTime and CollectionEndTime are the timestamps that indicate when the data was collected

The JSON message reflects the hierarchy of the router's data model. The hierarchy consists of:
  • containers: a container has nodes that can be of different types.

  • tables: a table also contains nodes, but the number of child nodes may vary, and they must be of the same type.

  • leaf node: a leaf contains a data value, such as integer or string.

The schema objects are mapped to JSON are in this manner:
  • Each container maps to a JSON object. The keys are strings that represent the schema names of the nodes; the values represent the values of the nodes.

  • JSON objects are also used to represent tables. In this case, the keys are based on naming information that is converted to string format. Two options are provided for encoding the naming information:
    • The default is restconf-style encoding, where naming parameters are contained within the child node to which it refers.

    • The embedded-keys option uses the naming information as keys in a JSON dictionary, with the corresponding child node forming the value.

  • Leaf data types are mapped in this manner:
    • Simple strings, integers, and booleans are mapped directly.

    • Enumeration values are stored as the string representation of the value.

    • Other simple data types, such as IP addresses, are mapped as strings.

Example: Rest-conf Encoding

For example, consider the path -
Interfaces(*).Counters.Protocols(“IPv4”) 
This has two naming parameters - the interface name and the protocol name - and represents a container holding leaf nodes which are packet and byte counters. This would be represented as follows:
{
  "Interfaces": [
    {
      "Name": "GigabitEthernet0/0/0/1"
      "Counters": {
        "Protocols": [
          {
            "ProtoName": "IPv4",
            "CollectionTime": 12345678,
            "InputPkts": 100,
            "InputBytes": 200,
          }
        ]
      }
    },{
      "Name": "GigabitEthernet0/0/0/2"
      "Counters": {
        "Protocols": [
          {
            "ProtoName": "IPv4",
            "CollectionTime": 12345678,
            "InputPkts": 400,
            "InputBytes": 500,
          }
        ]
      }
    }
  ]
}             
A naming parameter with multiple keys, for example Foo.Destination(IPAddress=10.1.1.1, Port=2000) would be represented as follows:
{
  "Foo": 
    {
    "Destination": [
      {
        "IPAddress": 10.1.1.1,
        "Port": 2000,
        "CollectionTime": 12345678,
        "Leaf1": 100,    
      }
    ]
  }
}  

Example: Embedded Keys Encoding

The embedded-keys encoding treats naming information in the path as keys in the JSON dictionary. The key name information is lost and there are extra levels in the hierarchy but it is clearer which data constitutes the key which may aid collectors when parsing it. This option is provided primarily for backwards-compatibility with 6.0.
{
  "Interfaces": {
      "GigabitEthernet0/0/0/1": {
        "Counters": {
          "Protocols": {
            "IPv4": {
              "CollectionTime": 12345678,
              "InputPkts": 100,
              "InputBytes": 200,
            }
          }
        }
      },
      "GigabitEthernet0/0/0/2": {
        "Counters": {
          "Protocols": {
            "IPv4": {
              "CollectionTime": 12345678,
              "InputPkts": 400,
              "InputBytes": 500,
            }
          }
        }
      }
    }
  }
}
A naming parameter with multiple keys, for example Foo.Destination(IPAddress=10.1.1.1, Port=2000), would be represented by nesting each key in order:
{
  "Foo": 
    {
    "Destination": {
        10.1.1.1: {
          2000: {
            Leaf1": 100,
          }
        }    
      }
    ]
  }
}  

GPB Message Format

The output of the GPB encoder consists entirely of GPBs and allows multiple tables in a single packet for scalability.

GPB (Google Protocol Buffer) encoder requires metadata in the form of compiled .proto files. A .proto file describes the GPB message format, which is used to stream data.

For UDP, the data is simply a GPB. Only the compact format is supported so the message can be interpreted as a TelemetryHeader message.

For TCP, the message body is either a Telemetry message or a TelemetryHeader message, depending on which of the following encoding types is configured:
  • Compact GPB format stores data in a compressed and non-self-describing format. A .proto file must be generated for each path in the policy file to be used by the receiver to decode the resulting data.

  • Key-value GPB format uses a single .proto file to encode data in a self-describing format. This encoding does not require a .proto file for each path. The data on the wire is much larger because key names are included.

In the following example, the policy group, alpha uses the default configuration of compact encoding and UDP transport. The policy group, beta uses compressed TCP and key-value encoding. The policy group, gamma uses compact encoding over uncompressed TCP.
telemetry policy-driven encoder gpb
  policy group alpha
    policy foo
    destination ipv4 192.168.1.1 port 1234
    destination ipv4 10.0.0.1 port 9876
  policy group beta
    policy bar
    policy whizz
    destination ipv4 10.20.30.40 port 3333
    transport tcp
      compression zlib
  policy group gamma
    policy bang
    destination ipv4 10.11.1.1 port 4444
    transport tcp 
      encoding-format gpb-compact

Compact GPB Format

The compact GPB format is intended for streaming large volumes of data at frequent intervals. The format minimizes the size of the message on the wire. Multiple tables can be sent in in a single packet for scalability.

Note


The tables can be split over multiple packets but fragmenting a row is not supported. If a row in the table is too large to fit in a single UDP frame, it cannot be streamed. Instead either switch to TCP, increase the MTU, or modify the .proto file.


The following .proto file shows the header, which is common to all packets sent by the encoder:
message TelemetryHeader {
  optional uint32 encoding = 1

  optional string policy_name = 2;
  optional string version = 3;
  optional string identifier = 4;

  optional uint64 start_time = 5;
  optional uint64 end_time = 6;
  
  repeated TelemetryTable tables = 7;
		}

message TelemetryTable {
  optional string policy_path = 1;
  repeated bytes  row = 2;
		}
where:
  • encoding is used by receivers to verify that the packet is valid.

  • policy name, version and identifier are metadata taken from the policy file.

  • start time and end time indicate the duration when the data is collected.

  • tables is a list of tables within the packet. This format indicates that it is possible to receive results for multiple schema paths in a single packet.

  • For each table:
    • policy path is the schema path.

    • row is one or more byte arrays that represents an encoded GPB.

Key-value GPB Format

The self-describing key-value GPB format uses a generic .proto file. This file encodes data as a sequence of key-value pairs. The field names are included in the output for the receiver to interpret the data.

The following .proto file shows the field containing the key-value pairs:
message Telemetry {
  uint64   collection_id = 1;
  string   base_path = 2;	
  string   subscription_identifier = 3;
  string   model_version = 4;
  uint64   collection_start_time = 5;
  uint64   msg_timestamp = 6;
  repeated TelemetryField fields = 14;
  uint64   collection_end_time = 15;
}

message TelemetryField {
  uint64         timestamp = 1;
  string         name = 2;
  bool           augment_data = 3;
  oneof value_by_type {
    bytes          bytes_value = 4;
    string         string_value = 5;
    bool           bool_value = 6;
    uint32         uint32_value = 7;
    uint64         uint64_value = 8;
    sint32         sint32_value = 9;
    sint64         sint64_value = 10;  
    double         double_value = 11;
    float          float_value = 12;
  }
  repeated TelemetryField fields = 15;
}
where:
  • collection_id, base_path, collection_start_time and collection_end_time provide streaming details.

  • subscription_identifier is a fixed value for cadence-driven telemetry. This is used to distinguish from event-driven data.

  • model_version contains a string used for the version of the data model, as applicable.

Telemetry Receiver

A telemetry receiver is used as a destination to store streamed data.

A sample receiver that handles both JSON and GPB encodings is available in the Github repository.

A copy of the cisco.proto file is required to compile code for a GPB receiver. The cisco.proto file is available in the Github repository.

If you are building your own collector, use the standard protoc compiler. For example, for the GPB compact encoding:
protoc --python_out . -I=/sw/packages/protoc/current/google/include/:. generic_counters.proto ipv4_counters.proto
where:
  • --python_out <out_dir> specifies the location of the resulting generated files. These files are of the form <name>_pb2.py.

  • -I <import_path> specifies the path to look for imports. This must include the location of descriptor.proto from Google. (in /sw/packages) and cisco.proto and the .proto files that are compiled.

All files shown in the above example are located in the local directory.