json-kafka

JSON Kafka (output plugin)

Special version of JSON output plugin for Apache Kafka. The plugin is copy of the original plugin so it offers the same formatting options, however, it provides only Kafka output.

See "Kafka notes" section below for more information.

The plugin converts IPFIX flow records into JSON format and sends them to Apache Kafka. Some elements (timestamps, IP/MAC addresses, TCP flags, etc.) can be converted into human readable format. Others are represented as plain numbers.

Each flow record is converted individually and send to outputs where at least one output method must be configured. All fields in an IPFIX record are formatted as: <field name>:<value>, where the field name represents a name of a corresponding Information Element as "<Enterprise Name>:<Information Element Name>" and value is a number, string, boolean or null (if conversion from IPFIX to JSON fails or it is a zero-length field). The Enterprise Name is a name of an organization that manages the field, such as IANA, CISCO, etc.

If the field name and type is unknown and "ignoreUnknown" option is disabled, the field name use format 'enXX:idYY' where XX is an Enterprise Number and YY is an Information Element ID. See notes for instructions on how to add missing definitions of Information Elements.

Flow records with structured data types (basicList, subTemplateList and subTemplateMultiList) are supported. See the particular section below for information about the formatting of these types.

Example output

The following output is for illustration purpose only. Available IPFIX fields depend on a structure of records that your exporter sends. See documentation of the exporter, if you want to change available IPFIX fields.

{
    "@type": "ipfix.entry",
    "iana:octetDeltaCount": 3568720,
    "iana:packetDeltaCount": 2383,
    "iana:flowStartMilliseconds": "2018-05-11T19:44:29.006Z",
    "iana:flowEndMilliseconds": "2018-05-11T19:44:30.364Z",
    "iana:ingressInterface": 1,
    "iana:ipVersion": 4,
    "iana:sourceIPv4Address": "192.168.0.100",
    "iana:destinationIPv4Address": "192.168.0.101",
    "iana:ipClassOfService": 0,
    "iana:ipTTL": 62,
    "iana:protocolIdentifier": "TCP",
    "iana:tcpControlBits": ".AP.S.",
    "iana:sourceTransportPort": 47200,
    "iana:destinationTransportPort": 20810,
    "iana:egressInterface": 0,
    "iana:samplingInterval": 0,
    "iana:samplingAlgorithm": 0
}

The plugin can produce multiple record types, which are always distinguished by "@type" key:

ipfix.entry

Flow record captured by the exporter as shown in the example above. This record is always converted and it is the most common type.

ipfix.optionsEntry

Record containing additional information provided by the exporter to the collector. Usually it is used for information about the exporter configuration (such as packet sampling) and flow export statistics. Periodicity and content of the record always depends on the particular exporter. By default, record conversion is disabled, see ignoreOptions parameter in the configuration section.

ipfix.template and ipfix.optionsTemplate

Records describing structure of flow records (i.e. ipfix.entry) and options records (i.e. ipfix.optionsEntry), respectively. These types are usually useful only for exporter analytics. Periodicity of these records usually depends on the exporter and used transport protocol. For example, for connection over UDP it corresponds to "template refresh interval" configured on the exporter. However, for TCP and SCTP sessions, the template records are usually sent only once immediately after session start. By default, template record conversion is disabled, see templateInfo parameter in the configuration section.

Example configuration

Below you see a complex configuration with all available output options enabled. Don't forget to remove (or comment) outputs that you don't want to use!

<output>
    <name>JSON output</name>
    <plugin>json-kafka</plugin>
    <params>
        <tcpFlags>formatted</tcpFlags>
        <timestamp>formatted</timestamp>
        <protocol>formatted</protocol>
        <ignoreUnknown>true</ignoreUnknown>
        <ignoreOptions>true</ignoreOptions>
        <nonPrintableChar>true</nonPrintableChar>
        <octetArrayAsUint>true</octetArrayAsUint>
        <numericNames>false</numericNames>
        <splitBiflow>false</splitBiflow>
        <detailedInfo>false</detailedInfo>
        <templateInfo>false</templateInfo>

        <outputs>
            <kafka>
                <name>Send to Kafka</name>
                <brokers>127.0.0.1</brokers>
                <topic>ipfix</topic>
                <blocking>false</blocking>
                <partition>unassigned</partition>

                <!-- Zero or more additional properties -->
                <property>
                    <key>compression.codec</key>
                    <value>lz4</value>
                </property>
            </kafka>
        </outputs>
    </params>
</output>

Parameters

Formatting parameters:

tcpFlags

Convert TCP flags to common textual representation (formatted, e.g. ".A..S.") or to a number (raw). [values: formatted/raw, default: formatted]

timestamp

Convert timestamp to ISO 8601 textual representation (all timestamps in UTC and milliseconds, e.g. "2018-01-22T09:29:57.828Z") or to a unix timestamp (all timestamps in milliseconds). [values: formatted/unix, default: formatted]

protocol

Convert protocol identification to formatted style (e.g. instead 6 writes "TCP") or to a number. [values: formatted/raw, default: formatted]

ignoreUnknown

Skip unknown Information Elements (i.e. record fields with unknown name and data type). If disabled, data of unknown elements are formatted as unsigned integer or hexadecimal values. For more information, see octetArrayAsUint option. [values: true/false, default: true]

ignoreOptions

Skip non-flow records used for reporting metadata about IPFIX Exporting and Metering Processes (i.e. records described by Options Templates). [values: true/false, default: true]

nonPrintableChar

Ignore non-printable characters (newline, tab, control characters, etc.) in IPFIX strings. If disabled, these characters are escaped on output. [values: true/false, default: true]

octetArrayAsUint

Converter each IPFIX field with octetArray type (including IPFIX fields with unknown definitions) as unsigned integer if the size of the field is less or equal to 8 bytes. Fields with the size above the limit are always converted as string representing hexadecimal value, which is typically in network byte order (e.g. "0x71E1"). Keep on mind, that there might be octetArray fields with variable length that might be interpreted differently based on their size. If disabled, octetArray fields are never interpreted as unsigned integers. [values: true/false, default: true]

numericNames

Use only short identification of Information Elements (i.e. "enXX:idYY"). If enabled, the short version is used even if the definition of the field is known. This option can help to create a shorter JSON records with key identifiers which are independent on the internal configuration. [values: true/false, default: false]

splitBiflow

In case of Biflow records, split the record to two unidirectional flow records. Non-biflow records are unaffected. [values: true/false, default: false]

detailedInfo

Add additional information about the IPFIX message (such as export time, Observation Domain ID, IP address of the exporter, etc.) to which each record belongs. Additional fields starts with "ipfix:" prefix. [values: true/false, default: false]

templateInfo

Convert Template and Options Template records. See the particular section below for information about the formatting of these records. [values: true/false, default: false]

---

Output types: At least one output must be configured. Multiple kafka outputs can be used at the same time if the outputs are not in collision with each other.

kafka

Send data to Kafka i.e. Kafka producer.

name

Identification name of the output. Used only for readability.

brokers

Initial list of brokers as a CSV list of broker "host" or "host:port".

topic

Kafka topic to produce to.

partition

Partition number to produce to. If the value is "unassigned", then the default random distribution is used. [default: "unassigned"]

brokerVersion

Older broker versions (before 0.10.0) provide no way for a client to query for supported protocol features making it impossible for the client to know what features it may use. As a workaround a user may set this property to the expected broker version and the client will automatically adjust its feature set. [default: <empty>]

blocking

Enable blocking on produce. If disabled and a cluster is down or not able to retrieve records fast enough, some flow records may be dropped. On the other hand, if enabled, no records are dropped. However, if the cluster is slow or not accessible at all, the plugin waits (i.e. blocks) until data are send. This can significantly slow down or block(!) the whole collector and other output plugins [true/false, default: false]

performanceTuning

By default, the connection provided by librdkafka is not optimized for high throughput required for transport of JSON records. This option adds optional library parameters, which reduces messaging overhead and significantly improves throughput. In particular, Kafka message capacity is increased and maximal buffering interval is prolonged. These options can be overwritten by user defined properties. [true/false, default: true]

property

Additional configuration properties of librdkafka library as key/value pairs. Multiple <property> parameters, which can improve performance, can be defined. See the project website for the full list of supported options. Keep on mind that some options might not be available in all versions of the library.

Notes

If one or more Information Element definitions are missing, you can easily add them. All definitions are provided by libfds library. See the project website for help.

If a flow record contains multiple occurrences of the same Information Element, their values are stored into a single name/value pair as JSON array. Order of the values in the array corresponds to their order in the flow record.

For higher performance, it is advisable to use non-formatted conversion of IPFIX data types. In that case, you should prefer, for example, timestamps as numbers over ISO 8601 strings and numeric identifiers of fields as they are usually shorted.

Structured data types

Flow records can be extended with structured data types (as described in RFC6313). Each of these types are formatted as JSON objects with "@type" field which helps to distinguish its formatting. Moreover, as the standard describes, the semantic of the list is also included.

Converted basicList contains "fieldID" with the Information Element identifier of zero or more Information Element(s) contained in the list. All values are stored as a JSON array in "data" field.

{
    "example:blField": {
        "@type": "basicList",
        "semantic": "anyOf",
        "fieldID": "iana:octetDeltaCount",
        "data": [23, 34, 23]
    }
}

Converted subTemplateList contains only additional "data" field with array of zero or more JSON objects. As all nested JSON object are described by the same IPFIX Template, it's guaranteed the their structure is also the same. The "semantic" field indicates the relationship among the different JSON objects.

{
    "example:stlField": {
        "@type": "subTemplateList",
        "semantic": "allOf",
        "data": [
            {"keyA.1": "value1", "keyA.2": "value2"},
            {"keyB.1": "value1", "keyB.2": "value2"},
        ]
    }
}

Converted subTemplateMultiList is similar to the previous type, however, sub-records can be even more nested. The "data" field contains a JSON array with zero or more nested JSON arrays. Each nested array contains zero or more JSON objects and it's guaranteed that JSON objects in the same array have the same structure. The "semantic" field indicates top-level relationship among the nested arrays.

{
    "example:stmlField": {
        "@type": "subTemplateMultiList",
        "semantic": "allOf",
        "data" : [
            [
                {"keyA.1": "value1", "keyA.2": "value2"},
                {"keyB.1": "value1", "keyB.2": "value2"},
            ],
            [
                {"idA.1": "something", "idB.1": 123}
            ]
        ]
    }
}

Keep on mind that all structures can be nested in each other.

Template and Options Template records

These records are converted only if <templateInfo> option is enabled.

Template record describes structure of flow records, and its "@type" is "ipfix.template". Converted Template record contains "ipfix:templateId" field, which is unique to the Transport Session and Observation Domain, and "ipfix:fields", which is an array of JSON objects specifying fields of flow records.

{
    "@type": "ipfix.template",
    "ipfix:templateId": 49171,
    "ipfix:fields": [
        {
            "ipfix:elementId": 16399,
            "ipfix:enterpriseId": 6871,
            "ipfix:fieldLength": 1
        }, {
            "ipfix:elementId": 184,
            "ipfix:enterpriseId": 29305,
            "ipfix:fieldLength": 4
        }, {
            ...
        }
    ]
}

Options Template record describes structure of additional information for the collector, such as sampling configuration, export statistics, etc. The description is converted to a record with "@type" equal to "ipfix.optionsEntry". Converted Options Template record is similar to the previous type, however, it contains additional "ipfix:scopeCount" field, which specifies number of scope fields in the record. A scope count of N specifies that the first N field specifiers are scope fields.

{
    "@type": "ipfix.optionsTemplate",
    "ipfix:templateId": 53252,
    "ipfix:scopeCount": 1,
    "ipfix:fields": [
        {
            "ipfix:elementId": 130,
            "ipfix:enterpriseId": 0,
            "ipfix:fieldLength": 4
        }, {
            "ipfix:elementId": 103,
            "ipfix:enterpriseId": 6871,
            "ipfix:fieldLength": 4
        }, {
            ...
        }
    ]
}

Kafka notes

The plugin depends on librdkafka which unfortunately redefines some common symbols (e.g. thrd_create) when built without support for C11 threads. This is usually not a problem, however, expected return codes of thread functions differs from common glibc/musl implementation.

As the JSON plugin and librdkafka library are loaded after collector start, the common version of symbols are used and the redefined ones are ignored. Later when librdkafka tries to create its internal processing threads, it fails as thrd_create() returned unexpected return code. This issue can be resolved by using RTLD_DEEPBIND flag of dlopen() function when loading the plugin. However, this flag is a GNU extension, which is not supported by all C libraries, and it can also break some other standard functions (e.g. C++ iostream).

Affected versions of the library are available in official package repositories of multiple distributions. Therefore, we created this special version of JSON output which instructs the collector core to use the RTLD_DEEPBIND flag.

This issue should be fixes since librdkafka v1.4.2. For more information see librdkafka issue #2681 on Github.