Kafka Connect HTTP Connector

Kafka Connect connector that enables Change Data Capture from JSON/HTTP APIs into Kafka.

This connector is for you if

• You want to (live) replicate a dataset exposed through JSON/HTTP API
• You want to do so efficiently
• You want to capture only changes, not full snapshots
• You want to do so via configuration, with no custom coding
• You want to be able to extend the connector if it comes to that

Examples

See examples, e.g.

• Jira Issues Search API
• Elasticsearch Search API

Getting Started

If your Kafka Connect deployment is automated and packaged with Maven, you can unpack the artifact on Kafka Connect plugins folder.

<plugin>
    <artifactId>maven-dependency-plugin</artifactId>
    <execution>
        <id>copy-kafka-connect-plugins</id>
        <phase>prepare-package</phase>
        <goals>
            <goal>unpack</goal>
        </goals>
        <configuration>
            <outputDirectory>${project.build.directory}/docker-build/plugins</outputDirectory>
            <artifactItems>
                <artifactItem>
                    <groupId>com.github.castorm</groupId>
                    <artifactId>kafka-connect-http</artifactId>
                    <version>0.8.11</version>
                    <type>tar.gz</type>
                    <classifier>plugin</classifier>
                </artifactItem>
            </artifactItems>
        </configuration>
    </execution>
</plugin>

Otherwise, you'll have to do it manually by downloading the package from the Releases Page.

More details on how to Install Connectors.

Source Connector

com.github.castorm.kafka.connect.http.HttpSourceConnector

Extension points

The connector can be easily extended by implementing your own version of any of the components below.

These are better understood by looking at the source task implementation:

public List<SourceRecord> poll() throws InterruptedException {

    throttler.throttle(offset.getTimestamp().orElseGet(Instant::now));

    HttpRequest request = requestFactory.createRequest(offset);

    HttpResponse response = requestExecutor.execute(request);

    List<SourceRecord> records = responseParser.parse(response);

    List<SourceRecord> unseenRecords = recordSorter.sort(records).stream()
            .filter(recordFilterFactory.create(offset))
            .collect(toList());

    confirmationWindow = new ConfirmationWindow<>(extractOffsets(unseenRecords));

    return unseenRecords;
}

public void commitRecord(SourceRecord record, RecordMetadata metadata) {
    confirmationWindow.confirm(record.sourceOffset());
}

public void commit() {
    offset = confirmationWindow.getLowWatermarkOffset()
            .map(Offset::of)
            .orElse(offset);
}

---

Timer: Throttling HttpRequest

Controls the rate at which HTTP requests are performed by informing the task, how long until the next execution is due.

http.timer

public interface Timer extends Configurable {

    Long getRemainingMillis();

    default void reset(Instant lastZero) {
        // Do nothing
    }
}

• Type: Class
• Default: com.github.castorm.kafka.connect.timer.AdaptableIntervalTimer
• Available implementations:
  • com.github.castorm.kafka.connect.timer.FixedIntervalTimer
  • com.github.castorm.kafka.connect.timer.AdaptableIntervalTimer

Throttling HttpRequest with FixedIntervalThrottler

Throttles rate of requests based on a fixed interval.

http.timer.interval.millis

Interval in between requests

• Type: Long
• Default: 60000

Throttling HttpRequests with AdaptableIntervalThrottler

Throttles rate of requests based on a fixed interval. It has, however, two modes of operation, with two different intervals:

• Up to date No new records in last poll, or there were new records, but "recently" created (shorter than interval)
• Catching up There were new records in last poll, but they were created "long ago" (longer than interval)

http.timer.interval.millis

Interval in between requests when up-to-date

• Type: Long
• Default: 60000

http.timer.catchup.interval.millis

Interval in between requests when catching up

• Type: Long
• Default: 30000

---

HttpRequestFactory: Creating a HttpRequest

The first thing our connector will need to do is creating a HttpRequest.

http.request.factory

public interface HttpRequestFactory extends Configurable {

    HttpRequest createRequest(Offset offset);
}

• Type: Class
• Default: com.github.castorm.kafka.connect.http.request.template.TemplateHttpRequestFactory
• Available implementations:
  • com.github.castorm.kafka.connect.http.request.template.TemplateHttpRequestFactory

http.offset.initial

Initial offset, comma separated list of pairs.

• Example: property1=value1, property2=value2
• Type: String
• Default: ""

Creating a HttpRequest with TemplateHttpRequestFactory

This HttpRequestFactory is based on template resolution.

http.request.method

Http method to use in the request.

• Type: String
• Default: GET

http.request.url

Http url to use in the request.

• Required
• Type: String

http.request.headers

Http headers to use in the request, , separated list of : separated pairs.

• Example: Name: Value, Name2: Value2
• Type: String
• Default: ""

http.request.params

Http query parameters to use in the request, & separated list of = separated pairs.

• Example: name=value & name2=value2
• Type: String
• Default: ""

http.request.body

Http body to use in the request.

• Type: String
• Default: ""

http.request.template.factory

public interface TemplateFactory {

    Template create(String template);
}

public interface Template {

    String apply(Offset offset);
}

Class responsible for creating the templates that will be used on every request.

• Type: Class
• Default: com.github.castorm.kafka.connect.http.request.template.freemarker.BackwardsCompatibleFreeMarkerTemplateFactory
• Available implementations:
  • com.github.castorm.kafka.connect.http.request.template.freemarker.BackwardsCompatibleFreeMarkerTemplateFactory Implementation based on FreeMarker which accepts offset properties without offset namespace (Deprecated)
  • com.github.castorm.kafka.connect.http.request.template.freemarker.FreeMarkerTemplateFactory Implementation based on FreeMarker
  • com.github.castorm.kafka.connect.http.request.template.NoTemplateFactory

Creating a HttpRequest with FreeMarkerTemplateFactory

FreeMarker templates will have the following data model available:

• offset
  • key
  • timestamp (as ISO8601 string, e.g.: 2020-01-01T00:00:00Z)
  • ... (custom offset properties)

Accessing any of the above withing a template can be achieved like this:

http.request.params=after=${offset.timestamp}

For an Epoch representation of the same string, FreeMarker built-ins should be used:

http.request.params=after=${offset.timestamp?datetime.iso?long}

For a complete understanding of the features provided by FreeMarker, please, refer to the User Manual

---

HttpClient: Executing a HttpRequest

Once our HttpRequest is ready, we have to execute it to get some results out of it. That's the purpose of the HttpClient

http.client

public interface HttpClient extends Configurable {

    HttpResponse execute(HttpRequest request) throws IOException;
}

• Type: Class
• Default: com.github.castorm.kafka.connect.http.client.okhttp.OkHttpClient
• Available implementations:
  • com.github.castorm.kafka.connect.http.client.okhttp.OkHttpClient

Executing a HttpRequest with OkHttpClient

Uses a OkHttp client.

http.client.connection.timeout.millis

Timeout for opening a connection

• Type: Long
• Default: 2000

http.client.read.timeout.millis

Timeout for reading a response

• Type: Long
• Default: 2000

http.client.connection.ttl.millis

Time to live for the connection

• Type: Long
• Default: 300000

---

HttpAuthenticator: Authenticating a HttpRequest

When executing the request, authentication might be required. The HttpAuthenticator is responsible for resolving the Authorization header to be included in the HttpRequest.

http.auth

public interface HttpAuthenticator extends Configurable {

    Optional<String> getAuthorizationHeader();
}

• Type: Class
• Default: com.github.castorm.kafka.connect.http.auth.ConfigurableHttpAuthenticator
• Available implementations:
  • com.github.castorm.kafka.connect.http.auth.ConfigurableHttpAuthenticator
  • com.github.castorm.kafka.connect.http.auth.NoneHttpAuthenticator
  • com.github.castorm.kafka.connect.http.auth.BasicHttpAuthenticator

Authenticating with ConfigurableHttpAuthenticator

Allows selecting the authentication type via configuration property

http.auth.type

Type of authentication

• Type: Enum { None, Basic }
• Default: None

Authenticating with BasicHttpAuthenticator

Allows selecting the authentication type via configuration property

http.auth.user

• Type: String
• Default: ""

http.auth.password

• Type: String
• Default: """

---

HttpResponseParser: Parsing a HttpResponse

Once our HttpRequest has been executed, as a result we'll have to deal with a HttpResponse and translate it into the list of SourceRecords expected by Kafka Connect.

http.response.parser

public interface HttpResponseParser extends Configurable {

    List<SourceRecord> parse(HttpResponse response);
}

• Type: Class
• Default: com.github.castorm.kafka.connect.http.response.PolicyHttpResponseParser
• Available implementations:
  • com.github.castorm.kafka.connect.http.response.PolicyHttpResponseParser
  • com.github.castorm.kafka.connect.http.response.KvHttpResponseParser

Parsing with PolicyHttpResponseParser

Vets the HTTP response deciding whether the response should be processed, skipped or failed. This decision is delegated to a HttpResponsePolicy. When the decision is to process the response, this processing is delegated to a secondary HttpResponseParser.

HttpResponsePolicy: Vetting a HttpResponse

http.response.policy

public interface HttpResponsePolicy extends Configurable {

    HttpResponseOutcome resolve(HttpResponse response);

    enum HttpResponseOutcome {
        PROCESS, SKIP, FAIL
    }
}

• Type: Class
• Default: com.github.castorm.kafka.connect.http.response.StatusCodeHttpResponsePolicy
• Available implementations:
  • com.github.castorm.kafka.connect.http.response.StatusCodeHttpResponsePolicy

http.response.policy.parser

• Type: Class
• Default: com.github.castorm.kafka.connect.http.response.KvHttpResponseParser
• Available implementations:
  • com.github.castorm.kafka.connect.http.response.KvHttpResponseParser

Vetting with StatusCodeHttpResponsePolicy

Does response vetting based on HTTP status codes in the response and the configuration below.

http.response.policy.codes.process

Comma separated list of code ranges that will result in the parser processing the response

• Example: 200..205, 207..210
• Type: String
• Default: 200..299

http.response.policy.codes.skip

Comma separated list of code ranges that will result in the parser skipping the response

• Example: 300..305, 307..310
• Type: String
• Default: 300..399

Parsing with KvHttpResponseParser

Parses the HTTP response into a key-value SourceRecord. This process is decomposed in two steps:

• Parsing the HttpResponse into a KvRecord
• Mapping the KvRecord into a SourceRecord

http.response.record.parser

public interface KvRecordHttpResponseParser extends Configurable {

    List<KvRecord> parse(HttpResponse response);
}

• Type: Class
• Default: com.github.castorm.kafka.connect.http.response.jackson.JacksonKvRecordHttpResponseParser
• Available implementations:
  • com.github.castorm.kafka.connect.http.response.jackson.JacksonKvRecordHttpResponseParser

http.response.record.mapper

public interface KvSourceRecordMapper extends Configurable {

    SourceRecord map(KvRecord record);
}

• Type: Class
• Default: com.github.castorm.kafka.connect.http.record.SchemedKvSourceRecordMapper
• Available implementations:
  • com.github.castorm.kafka.connect.http.record.SchemedKvSourceRecordMapper Maps key to a Struct schema with a single property key, and value to a Struct schema with a single property value
  • com.github.castorm.kafka.connect.http.record.StringKvSourceRecordMapper Maps both key and value to a String schema

Parsing with JacksonKvRecordHttpResponseParser

Uses Jackson to look for the records in the response.

http.response.list.pointer

JsonPointer to the property in the response body containing an array of records

• Example: /items
• Type: String
• Default: /

http.response.record.pointer

JsonPointer to the individual record to be used as kafka record body. Useful when the object we are interested in is under a nested structure

• Type: String
• Default: /

http.response.record.offset.pointer

Comma separated list of key=/value pairs where the key is the name of the property in the offset, and the value is the JsonPointer to the value being used as offset for future requests. This is the mechanism that enables sharing state in between HttpRequests. HttpRequestFactory implementations receive this Offset.

Special properties:

• key is used as record's identifier, used for de-duplication and topic partition routing
• timestamp is used as record's timestamp, used for de-duplication and ordering

One of the roles of the offset, even if not required for preparing the next request, is helping in deduplication of already seen records, by providing a sense of progress, assuming consistent ordering. (e.g. even if the response returns some repeated results in between requests because they have the same timestamp, anything prior to the last seen offset will be ignored). see OffsetFilterFactory

• Example: id=/itemId
• Type: String
• Default: ""

http.response.record.timestamp.parser

Class responsible for converting the timestamp property captured above into a java.time.Instant.

• Type: String
• Default: com.github.castorm.kafka.connect.http.response.timestamp.EpochMillisOrDelegateTimestampParser
• Available implementations:
  • com.github.castorm.kafka.connect.http.response.timestamp.EpochMillisTimestampParser Implementation that captures the timestamp as an epoch millis long
  • com.github.castorm.kafka.connect.http.response.timestamp.EpochMillisOrDelegateTimestampParser Implementation that tries to capture as epoch millis or delegates to another parser in case of failure
  • com.github.castorm.kafka.connect.http.response.timestamp.DateTimeFormatterTimestampParser Implementation based on based on a DateTimeFormatter
  • com.github.castorm.kafka.connect.http.response.timestamp.NattyTimestampParser Implementation based on Natty parser
  • com.github.castorm.kafka.connect.http.response.timestamp.RegexTimestampParser Implementation that extracts substring from timestamp column and parse it

http.response.record.timestamp.parser.pattern

When using DateTimeFormatterTimestampParser, a custom pattern can be specified

• Type: String
• Default: yyyy-MM-dd'T'HH:mm:ss[.SSS]X

http.response.record.timestamp.parser.zone

Timezone of the timestamp. Accepts ZoneId valid identifiers

• Type: String
• Default: UTC

http.response.record.timestamp.parser.regex

When using RegexTimestampParser, a custom regex pattern can be specified

• Type: String
• Default: .*

http.response.record.timestamp.parser.regex.delegate

When using RegexTimestampParser, a delegate class to parse timestamp

• Type: Class
• Default: DateTimeFormatterTimestampParser

---

Mapping a KvRecord into SourceRecord with SimpleKvSourceRecordMapper

Once we have our KvRecord we have to translate it into what Kafka Connect is expecting: SourceRecords

Embeds the record properties into a common simple envelope to enable schema evolution. This envelope simply contains a key and a value properties with customizable field names.

Here is also where we'll tell Kafka Connect to what topic and on what partition do we want to send our record.

** It's worth noticing there are projects out there that allow you to infer the schema from your json document. (e.g. expandjsonsmt)

kafka.topic

Name of the topic where the record will be sent to

• Required
• Type: String
• Default: ""

http.record.schema.key.property.name

Name of the key property in the key-value envelope

• Type: String
• Default: key

http.record.schema.value.property.name

Name of the value property in the key-value envelope

• Type: String
• Default: value

---

SourceRecordSorter: Sorting SourceRecords

Some Http resources not designed for CDC, return snapshots with most recent records first. In this cases de-duplication is especially important, as subsequent request are likely to produce similar results. The de-duplication mechanisms offered by this connector are order-dependent, as they are usually based on timestamps.

To enable de-duplication in cases like this, we can instruct the connector to assume a specific order direction, either ASC, DESC, or IMPLICIT, where implicit figures it out based on records' timestamps.

http.record.sorter

public interface SourceRecordSorter extends Configurable {

    List<SourceRecord> sort(List<SourceRecord> records);
}

• Type: Class
• Default: com.github.castorm.kafka.connect.http.record.OrderDirectionSourceRecordSorter
• Available implementations:
  • com.github.castorm.kafka.connect.http.record.OrderDirectionSourceRecordSorter

http.response.list.order.direction

Order direction of the results in the response list.

• Type: Enum { ASC, DESC, IMPLICIT }
• Default: IMPLICIT

---

SourceRecordFilterFactory: Filtering out SourceRecord

There are cases when we'll be interested in filtering out certain records. One of these would be de-duplication.

http.record.filter.factory

public interface SourceRecordFilterFactory extends Configurable {

    Predicate<SourceRecord> create(Offset offset);
}

• Type: Class
• Default: com.github.castorm.kafka.connect.http.record.OffsetRecordFilterFactory
• Available implementations:
  • com.github.castorm.kafka.connect.http.record.OffsetRecordFilterFactory
  • com.github.castorm.kafka.connect.http.record.OffsetTimestampRecordFilterFactory
  • com.github.castorm.kafka.connect.http.record.PassthroughRecordFilterFactory

Filtering out SourceRecord with OffsetTimestampRecordFilterFactory

De-duplicates based on Offset's timestamp, filtering out records with earlier or the same timestamp. Useful when timestamp is used to filter the HTTP resource, but the filter does not have full timestamp precision. Assumptions:

• Records are ordered by timestamp
• No two records can contain the same timestamp (to whatever precision the HTTP resource uses)

If the latter assumption cannot be satisfied, check OffsetRecordFilterFactory to try and prevents data loss.

Filtering out SourceRecord with OffsetRecordFilterFactory

De-duplicates based on Offset's timestamp, key and any other custom property present in the Offset, filtering out records with earlier timestamps, or when in the same timestamp, only those up to the last seen Offset properties. Useful when timestamp alone is not unique but together with some other Offset property is. Assumptions:

• Records are ordered by timestamp
• There is an Offset property that uniquely identify records (e.g. key)
• There won't be new items preceding already seen ones

---

Development

Building

mvn package

Debugging

Using Pre-configured docker setup

You can easily run a Kafka Connect cluster with kafka-connect-http pre-installed by executing:

mvn verify -Pdebug -DskipTests

It'll run dockerized versions of kafka and kafka-connect which you can access via REST API or attach debuggers to the url printed in console:

Kafka Connect testcontainers infra is ready
  Rest API: http://localhost:33216
  Debug agent: localhost:33217

Right after, it'll allow you to specify the file path to your connector's json configuration:

Introduce the path to your connector JSON configuration file:

It'll subscribe to the corresponding kafka topic, printing every message going through the output topic of your connector.

Using Kafka Connect standalone

These instructions are phrased in terms of the steps needed when using IntelliJ, but other integrated development environments are likely to be similar.

Point the Kafka stand-alone plugin.path at the module compile Output path. Assuming you are using the default Maven project import, this is the ./target directory, so the config/connect-standalone.properties file would contain the line

plugin.path=<directory where git clone was executed>/kafka-connect-http/kafka-connect-http/target

In the Run/Debug Configurations dialog, create a new Remote JVM Debug configuration with the mode Attach to remote JVM. When remote debugging, some Java parameters need to be specified when the program is executed. Fortunately there are hooks in the Kafka shell scripts to accommodate this. The Remote JVM Debug configuration specifies the needed Command line arguments for remote JVM. In the terminal console where you execute the connect command line, define KAFKA_DEBUG and JAVA_DEBUG_OPTS as:

export KAFKA_DEBUG=true
export JAVA_DEBUG_OPTS=-agentlib:jdwp=transport=dt_socket,server=y,suspend=n,address=*:5005

Place a suitable breakpoint in the kafka-connect-http code, e.g. in HttpSourceTask.start(), and launch the standalone connect program:

bin/connect-standalone.sh config/connect-standalone.properties plugins/<kafka-connect-http properties file>

Click the Debug icon in IntelliJ and ensure the debugger console says Connected to the target VM, address: 'localhost:5005', transport: 'socket' and the breakpoint you placed becomes checked. The program should now break when the breakpoint is hit.

Running the tests

mvn test

Releasing

• Update CHANGELOG.md and README.md files.
• Prepare release:

mvn release:clean release:prepare

Contributing

Contributions are welcome via pull requests, pending definition of code of conduct, please just follow existing conventions.

Versioning

We use SemVer for versioning.

License

This project is licensed under the Apache 2.0 License - see the LICENSE.txt file for details

Built With

• Maven - Dependency Management
• Kafka Connect - The framework for our connectors
• OkHttp - HTTP Client
• Jackson - Json deserialization
• FreeMarker - Template engine
• Natty - Date parser

Acknowledgments

• Inspired by llofberg/kafka-connect-rest