search

getindata / flink-http-connector

Http Connector for Apache Flink. Provides sources and sinks for Datastream , Table and SQL APIs.
Apache License 2.0
160 stars 45 forks source link
data-streaming flink flink-sql flink-stream-processing java

readme

flink-http-connector

Maven Central javadoc

The HTTP TableLookup connector that allows for pulling data from external system via HTTP GET method and HTTP Sink that allows for sending data to external system via HTTP requests.

Note: The main branch may be in an unstable or even broken state during development. Please use releases instead of the main branch in order to get a stable set of binaries.

The goal for HTTP TableLookup connector was to use it in Flink SQL statement as a standard table that can be later joined with other stream using pure SQL Flink.

Currently, HTTP source connector supports only Lookup Joins (TableLookup) [1] in Table/SQL API. HttpSink supports both Streaming API (when using HttpSink built using HttpSinkBuilder) and the Table API (using connector created in HttpDynamicTableSinkFactory).

Updating the connector

In case of updating http-connector please see Breaking changes section.

Prerequisites

Runtime dependencies

This connector has few Flink's runtime dependencies, that are expected to be provided.

Installation

In order to use the flink-http-connector the following dependencies are required for both projects using a build automation tool (such as Maven or SBT) and SQL Client with SQL JAR bundles. For build automation tool reference, look into Maven Central: https://mvnrepository.com/artifact/com.getindata/flink-http-connector.

Documentation

You can read the official JavaDoc documentation of the latest release at https://javadoc.io/doc/com.getindata/flink-http-connector.

Usage

HTTP TableLookup Source

Flink SQL table definition:

Enrichment Lookup Table

CREATE TABLE Customers (
    id STRING,
    id2 STRING,
    msg STRING,
    uuid STRING,
    details ROW<
      isActive BOOLEAN,
      nestedDetails ROW<
        balance STRING
      >
    >
) WITH (
'connector' = 'rest-lookup',
'format' = 'json',
'url' = 'http://localhost:8080/client',
'asyncPolling' = 'true'
)

Data Source Table

CREATE TABLE Orders (
    id STRING,
    id2 STRING,
    proc_time AS PROCTIME()
) WITH (
'connector' = 'datagen',
'rows-per-second' = '1',
'fields.id.kind' = 'sequence',
'fields.id.start' = '1',
'fields.id.end' = '120',
'fields.id2.kind' = 'sequence',
'fields.id2.start' = '2',
'fields.id2.end' = '120'
);

Using Customers table in Flink SQL Lookup Join with Orders table:

SELECT o.id, o.id2, c.msg, c.uuid, c.isActive, c.balance FROM Orders AS o 
JOIN Customers FOR SYSTEM_TIME AS OF o.proc_time AS c ON o.id = c.id AND o.id2 = c.id2

The columns and their values used for JOIN ON condition will be used as HTTP GET parameters where the column name will be used as a request parameter name.

For Example: http://localhost:8080/client/service?id=1&uuid=2

Or for REST POST method they will be converted to Json and used as request body. In this case, json request body will look like this:

{
    "id": "1",
    "uuid": "2"
}

Http headers

It is possible to set HTTP headers that will be added to HTTP request send by lookup source connector. Headers are defined via property key gid.connector.http.source.lookup.header.HEADER_NAME = header value for example: gid.connector.http.source.lookup.header.X-Content-Type-Options = nosniff.

Headers can be set using http lookup source table DDL. In example below, HTTP request done for http-lookup table will contain three headers:

CREATE TABLE http-lookup (
  id bigint,
  some_field string
) WITH (
  'connector' = 'rest-lookup',
  'format' = 'json',
  'url' = 'http://localhost:8080/client',
  'asyncPolling' = 'true',
  'gid.connector.http.source.lookup.header.Origin' = '*',
  'gid.connector.http.source.lookup.header.X-Content-Type-Options' = 'nosniff',
  'gid.connector.http.source.lookup.header.Content-Type' = 'application/json'
)

Custom REST query

Http Lookup Source builds queries out of JOIN clauses. One can customize how those queries are built by implementing LookupQueryCreator and LookupQueryCreatorFactory interfaces. Custom implementations of LookupQueryCreatorFactory can be registered along other factories in resources/META-INF.services/org.apache.flink.table.factories.Factory file and then referenced by their identifiers in the Http Lookup Source DDL property field gid.connector.http.source.lookup.query-creator.

A default implementation that builds an "ordinary" GET query, i.e. adds ?joinColumn1=value1&joinColumn2=value2&... to the URI of the endpoint,

For body based queries such as POST/PUT requests, the (GenericGetQueryCreator) is provided as a default query creator. This implementation uses Flink's json-format to convert RowData object into Json String.

The GenericGetQueryCreator allows for using custom formats that will perform serialization to Json. Thanks to this, users can create their own logic for converting RowData to Json Strings suitable for their HTTP endpoints and use this logic as custom format with HTTP Lookup connector and SQL queries. To create a custom format user has to implement Flink's SerializationSchema and SerializationFormatFactory interfaces and register custom format factory along other factories in resources/META-INF.services/org.apache.flink.table.factories.Factory file. This is common Flink mechanism for providing custom implementations for various factories.

In order to use custom format, user has to specify option 'lookup-request.format' = 'customFormatName', where customFormatName is the identifier of custom format factory.

Additionally, it is possible to pass query format options from table's DDL. This can be done by using option like so: 'lookup-request.format.customFormatName.customFormatProperty' = 'propertyValue', for example 'lookup-request.format.customFormatName.fail-on-missing-field' = 'true'.

It is important that customFormatName part match SerializationFormatFactory identifier used for custom format implementation. In this case, the fail-on-missing-field will be passed to SerializationFormatFactory::createEncodingFormat( DynamicTableFactory.Context context, ReadableConfig formatOptions) method in ReadableConfig object.

With default configuration, Flink-Json format is used for GenericGetQueryCreator, all options defined in json-format can be passed through table DDL. For example 'lookup-request.format.json.fail-on-missing-field' = 'true'. In this case, format identifier is json.

Timeouts

Lookup Source is guarded by two timeout timers. First one is specified by Flink's AsyncIO operator that executes AsyncTableFunction. The default value of this timer is set to 3 minutes and can be changed via table.exec.async-lookup.timeout option.

The second one is set per individual HTTP requests by HTTP client. Its default value is set currently to 30 seconds and can be changed via gid.connector.http.source.lookup.request.timeout option.

Flink's current implementation of AsyncTableFunction does not allow specifying custom logic for handling Flink AsyncIO timeouts as it is for Java API. Because of that, if AsyncIO timer passes, Flink will throw TimeoutException which will cause job restart.

The HTTP request timeouts on the other hand will not cause Job restart. In that case, exception will be logged into application logs. To avoid job restart on timeouts caused by Lookup queries, the value of gid.connector.http.source.lookup.request.timeout should be smaller than table.exec.async-lookup.timeout.

HTTP Sink

The following example shows the minimum Table API example to create a HttpDynamicSink that writes JSON values to an HTTP endpoint using POST method, assuming Flink has JAR of JSON serializer installed:

CREATE TABLE http (
  id bigint,
  some_field string
) WITH (
  'connector' = 'http-sink',
  'url' = 'http://example.com/myendpoint',
  'format' = 'json'
)

Then use INSERT SQL statement to send data to your HTTP endpoint:

INSERT INTO http VALUES (1, 'Ninette'), (2, 'Hedy')

Due to the fact that HttpSink sends bytes inside HTTP request's body, one can easily swap 'format' = 'json' for some other format.

Other examples of usage of the Table API can be found in some tests.

Request submission

Starting from version 0.10 HTTP Sink by default submits events in batch. Before version 0.10 the default and only submission type was single. This is a breaking compatibility change.

The submission mode can be changed using gid.connector.http.sink.writer.request.mode property using single or batch as property value.

Batch submission mode

In batch mode, a number of events (processed elements) will be batched and submitted in one HTTP request. In this mode, HTTP PUT/POST request's body contains a Json array, where every element of this array represents individual event.

An example of Http Sink batch request body containing data for three events:

[
  {
    "id": 1,
    "first_name": "Ninette",
    "last_name": "Clee",
    "gender": "Female",
    "stock": "CDZI",
    "currency": "RUB",
    "tx_date": "2021-08-24 15:22:59"
  },
  {
    "id": 2,
    "first_name": "Rob",
    "last_name": "Zombie",
    "gender": "Male",
    "stock": "DGICA",
    "currency": "GBP",
    "tx_date": "2021-10-25 20:53:54"
  },
  {
    "id": 3,
    "first_name": "Adam",
    "last_name": "Jones",
    "gender": "Male",
    "stock": "DGICA",
    "currency": "PLN",
    "tx_date": "2021-10-26 20:53:54"
  }
]

By default, batch size is set to 500 which is the same as Http Sink's maxBatchSize property and has value of 500. The `maxBatchSize' property sets maximal number of events that will by buffered by Flink runtime before passing it to Http Sink for processing.

In order to change submission batch size use gid.connector.http.sink.request.batch.size property. For example:

Streaming API:

HttpSink.<String>builder()
      .setEndpointUrl("http://example.com/myendpoint")
      .setElementConverter(
          (s, _context) -> new HttpSinkRequestEntry("POST", s.getBytes(StandardCharsets.UTF_8)))
      .setProperty("gid.connector.http.sink.request.batch.size", "50")
      .build();

SQL:

CREATE TABLE http (
  id bigint,
  some_field string
) WITH (
  'connector' = 'http-sink',
  'url' = 'http://example.com/myendpoint',
  'format' = 'json',
  'gid.connector.http.sink.request.batch.size' = '50'
)

Single submission mode

In this mode every processed event is submitted as individual HTTP POST/PUT request.

Streaming API:

HttpSink.<String>builder()
      .setEndpointUrl("http://example.com/myendpoint")
      .setElementConverter(
          (s, _context) -> new HttpSinkRequestEntry("POST", s.getBytes(StandardCharsets.UTF_8)))
      .setProperty("gid.connector.http.sink.writer.request.mode", "single")
      .build();

SQL:

CREATE TABLE http (
  id bigint,
  some_field string
) WITH (
  'connector' = 'http-sink',
  'url' = 'http://example.com/myendpoint',
  'format' = 'json',
  'gid.connector.http.sink.writer.request.mode' = 'single'
)

Http headers

It is possible to set HTTP headers that will be added to HTTP request send by sink connector. Headers are defined via property key gid.connector.http.sink.header.HEADER_NAME = header value for example: gid.connector.http.sink.header.X-Content-Type-Options = nosniff. Properties can be set via Sink builder or Property object:

HttpSink.<String>builder()
      .setEndpointUrl("http://example.com/myendpoint")
      .setElementConverter(
          (s, _context) -> new HttpSinkRequestEntry("POST", s.getBytes(StandardCharsets.UTF_8)))
      .setProperty("gid.connector.http.sink.header.X-Content-Type-Options", "nosniff")
      .build();

or

Properties properties = Properties();
properties.setProperty("gid.connector.http.sink.header.X-Content-Type-Options", "nosniff");

HttpSink.<String>builder()
      .setEndpointUrl("http://example.com/myendpoint")
      .setElementConverter(
          (s, _context) -> new HttpSinkRequestEntry("POST", s.getBytes(StandardCharsets.UTF_8)))
      .setProperties(properties)
      .build();

In Table/SQL API, headers can be set using http sink table DDL. In example below, HTTP request done for http table will contain three headers:

CREATE TABLE http (
  id bigint,
  some_field string
) WITH (
  'connector' = 'http-sink',
  'url' = 'http://example.com/myendpoint',
  'format' = 'json',
  'gid.connector.http.sink.header.Origin' = '*',
  'gid.connector.http.sink.header.X-Content-Type-Options' = 'nosniff',
  'gid.connector.http.sink.header.Content-Type' = 'application/json'
)

Note that when using OIDC, it adds an Authentication header with the bearer token; this will override an existing Authorization header specified in configuration.

Custom request/response callback

HTTP status code handler

Http Sink and Lookup Source connectors allow defining list of HTTP status codes that should be treated as errors. By default all 400s and 500s response codes will be interpreted as error code.

This behavior can be changed by using below properties in table definition (DDL) for Sink and Lookup Source or passing it via `setProperty' method from Sink's builder. The property names are:

TLS (more secure replacement for SSL) and mTLS support

Both Http Sink and Lookup Source connectors support HTTPS communication using TLS 1.2 and mTLS. To enable Https communication simply use https protocol in endpoint's URL.

To specify certificate(s) to be used by the server, use gid.connector.http.security.cert.server connector property; the value is a comma separated list of paths to certificate(s), for example you can use your organization's CA Root certificate, or a self-signed certificate.

Note that if there are no security properties for a https url then, the JVMs default certificates are used - allowing use of globally recognized CAs without the need for configuration.

You can also configure the connector to use mTLS. For this simply use gid.connector.http.security.cert.client and gid.connector.http.security.key.client connector properties to specify paths to the certificate and private key. The key MUST be in PCKS8 format. Both PEM and DER keys are allowed.

All properties can be set via Sink's builder .setProperty(...) method or through Sink and Source table DDL.

For non production environments it is sometimes necessary to use Https connection and accept all certificates. In this special case, you can configure connector to trust all certificates without adding them to keystore. To enable this option use gid.connector.http.security.cert.server.allowSelfSigned property setting its value to true.

Basic Authentication

The connector supports Basic Authentication using a HTTP Authorization header. The header value can be set via properties, similarly as for other headers. The connector converts the passed value to Base64 and uses it for the request. If the used value starts with the prefix Basic, or gid.connector.http.source.lookup.use-raw-authorization-header is set to 'true', it will be used as header value as is, without any extra modification.

OIDC Bearer Authentication

The connector supports Bearer Authentication using a HTTP Authorization header. The OAuth 2.0 rcf mentions Obtaining Authorization and an authorization grant. OIDC makes use of this authorisation grant in a Token Request by including a OAuth grant type and associated properties, the response is the token response.

If you want to use this authorization then you should supply the Token Request body in application/x-www-form-urlencoded encoding in configuration property gid.connector.http.security.oidc.token.request. See grant extension for an example of a customised grant type token request. The supplied token request will be issued to the token end point, whose url should be supplied in configuration property gid.connector.http.security.oidc.token.endpoint.url. The returned access token is then cached and used for subsequent requests; if the token has expired then a new one is requested. There is a property gid.connector.http.security.oidc.token.expiry.reduction, that defaults to 1 second; new tokens will be requested if the current time is later than the cached token expiry time minus gid.connector.http.security.oidc.token.expiry.reduction.

Restrictions at this time

Table API Connector Options

HTTP TableLookup Source

Option Required Description/Value
connector required The Value should be set to rest-lookup
format required Flink's format name that should be used to decode REST response, Use json for a typical REST endpoint.
url required The base URL that should be use for GET requests. For example http://localhost:8080/client
asyncPolling optional true/false - determines whether Async Polling should be used. Mechanism is based on Flink's Async I/O.
lookup-method optional GET/POST/PUT (and any other) - determines what REST method should be used for lookup REST query. If not specified, GET method will be used.
lookup.cache optional Enum possible values: NONE, PARTIAL. The cache strategy for the lookup table. Currently supports NONE (no caching) and PARTIAL (caching entries on lookup operation in external API).
lookup.partial-cache.max-rows optional The max number of rows of lookup cache, over this value, the oldest rows will be expired. lookup.cache must be set to PARTIAL to use this option. See the following Lookup Cache section for more details.
lookup.partial-cache.expire-after-write optional The max time to live for each rows in lookup cache after writing into the cache. Specify as a Duration. lookup.cache must be set to PARTIAL to use this option. See the following Lookup Cache section for more details.
lookup.partial-cache.expire-after-access optional The max time to live for each rows in lookup cache after accessing the entry in the cache. Specify as a Duration. lookup.cache must be set to PARTIAL to use this option. See the following Lookup Cache section for more details.
lookup.partial-cache.cache-missing-key optional This is a boolean that defaults to true. Whether to store an empty value into the cache if the lookup key doesn't match any rows in the table. lookup.cache must be set to PARTIAL to use this option. See the following Lookup Cache section for more details.
lookup.max-retries optional The max retry times if the lookup failed; default is 3. See the following Lookup Cache section for more details.
gid.connector.http.lookup.error.code optional List of HTTP status codes that should be treated as errors by HTTP Source, separated with comma.
gid.connector.http.lookup.error.code.exclude optional List of HTTP status codes that should be excluded from the gid.connector.http.lookup.error.code list, separated with comma.
gid.connector.http.security.cert.server optional Comma separated paths to trusted HTTP server certificates that should be added to the connectors trust store.
gid.connector.http.security.cert.client optional Path to trusted certificate that should be used by connector's HTTP client for mTLS communication.
gid.connector.http.security.key.client optional Path to trusted private key that should be used by connector's HTTP client for mTLS communication.
gid.connector.http.security.cert.server.allowSelfSigned optional Accept untrusted certificates for TLS communication.
gid.connector.http.security.oidc.token.request optional OIDC Token Request body in application/x-www-form-urlencoded encoding
gid.connector.http.security.oidc.token.endpoint.url optional OIDC Token Endpoint url, to which the token request will be issued
gid.connector.http.security.oidc.token.expiry.reduction optional OIDC tokens will be requested if the current time is later than the cached token expiry time minus this value.
gid.connector.http.source.lookup.request.timeout optional Sets HTTP request timeout in seconds. If not specified, the default value of 30 seconds will be used.
gid.connector.http.source.lookup.request.thread-pool.size optional Sets the size of pool thread for HTTP lookup request processing. Increasing this value would mean that more concurrent requests can be processed in the same time. If not specified, the default value of 8 threads will be used.
gid.connector.http.source.lookup.response.thread-pool.size optional Sets the size of pool thread for HTTP lookup response processing. Increasing this value would mean that more concurrent requests can be processed in the same time. If not specified, the default value of 4 threads will be used.
gid.connector.http.source.lookup.use-raw-authorization-header optional If set to 'true', uses the raw value set for the Authorization header, without transformation for Basic Authentication (base64, addition of "Basic " prefix). If not specified, defaults to 'false'.
gid.connector.http.source.lookup.request-callback optional Specify which HttpLookupPostRequestCallback implementation to use. By default, it is set to slf4j-lookup-logger corresponding to Slf4jHttpLookupPostRequestCallback.

HTTP Sink

Option Required Description/Value
connector required Specify what connector to use. For HTTP Sink it should be set to 'http-sink'.
format required Specify what format to use.
url required The base URL that should be use for HTTP requests. For example http://localhost:8080/client.
insert-method optional Specify which HTTP method to use in the request. The value should be set either to POST or PUT.
sink.batch.max-size optional Maximum number of elements that may be passed in a batch to be written downstream.
sink.requests.max-inflight optional The maximum number of in flight requests that may exist, if any more in flight requests need to be initiated once the maximum has been reached, then it will be blocked until some have completed.
sink.requests.max-buffered optional Maximum number of buffered records before applying backpressure.
sink.flush-buffer.size optional The maximum size of a batch of entries that may be sent to the HTTP endpoint measured in bytes.
sink.flush-buffer.timeout optional Threshold time in milliseconds for an element to be in a buffer before being flushed.
gid.connector.http.sink.request-callback optional Specify which HttpPostRequestCallback implementation to use. By default, it is set to slf4j-logger corresponding to Slf4jHttpPostRequestCallback.
gid.connector.http.sink.error.code optional List of HTTP status codes that should be treated as errors by HTTP Sink, separated with comma.
gid.connector.http.sink.error.code.exclude optional List of HTTP status codes that should be excluded from the gid.connector.http.sink.error.code list, separated with comma.
gid.connector.http.security.cert.server optional Path to trusted HTTP server certificate that should be add to connectors key store. More than one path can be specified using , as path delimiter.
gid.connector.http.security.cert.client optional Path to trusted certificate that should be used by connector's HTTP client for mTLS communication.
gid.connector.http.security.key.client optional Path to trusted private key that should be used by connector's HTTP client for mTLS communication.
gid.connector.http.security.cert.server.allowSelfSigned optional Accept untrusted certificates for TLS communication.
gid.connector.http.sink.request.timeout optional Sets HTTP request timeout in seconds. If not specified, the default value of 30 seconds will be used.
gid.connector.http.sink.writer.thread-pool.size optional Sets the size of pool thread for HTTP Sink request processing. Increasing this value would mean that more concurrent requests can be processed in the same time. If not specified, the default value of 1 thread will be used.
gid.connector.http.sink.writer.request.mode optional Sets Http Sink request submission mode. Two modes are available to select, single and batch which is the default mode if option is not specified.
gid.connector.http.sink.request.batch.size optional Applicable only for gid.connector.http.sink.writer.request.mode = batch. Sets number of individual events/requests that will be submitted as one HTTP request by HTTP sink. The default value is 500 which is same as HTTP Sink maxBatchSize

Lookup Cache

The HTTP Client connector can be used in lookup join as a lookup source (also known as a dimension table).

By default, the lookup cache is not enabled. You can enable it by setting lookup.cache to PARTIAL. The scope of the cache is per job, so long-running jobs can benefit from this caching.

The lookup cache is used to improve the performance of temporal joins. By default, the lookup cache is not enabled, so all the API requests are sent on the network. When the lookup cache is enabled, Flink looks in the cache first, and only sends requests on the network when there is no cached value, then the cache is updated with the returned rows. The oldest rows in this cache are expired when the cache hits the max cached rows lookup.partial-cache.max-rows or when the row exceeds the max time to live specified by lookup.partial-cache.expire-after-write or lookup.partial-cache.expire-after-access.

By default, flink caches the empty query result for the primary key. You can toggle this behaviour by setting lookup.partial-cache.cache-missing-key to false.

Build and deployment

To build the project locally you need to have maven 3 and Java 11+.

Project build command: mvn package.
Detailed test report can be found under target/site/jacoco/index.xml.

Demo application

Note: This demo works only for Flink-1.15x.

You can test this connector using simple mock http server provided with this repository and Flink SQL-client. The mock server can be started from IDE (currently only this way) by running HttpStubApp::main method. It will start HTTP server listening on http://localhost:8080/client

Steps to follow:

Create Http Connector Lookup Table:

CREATE TABLE Customers (
    id STRING,
    id2 STRING,
    msg STRING,
    uuid STRING,
    details ROW<
      isActive BOOLEAN,
      nestedDetails ROW<
        balance STRING
      >
    >
) WITH (
'connector' = 'rest-lookup',
'format' = 'json',
'url' = 'http://localhost:8080/client', 
'asyncPolling' = 'true'
);

Submit SQL Select query to join both tables:

SELECT o.id, o.id2, c.msg, c.uuid, c.isActive, c.balance FROM Orders AS o JOIN Customers FOR SYSTEM_TIME AS OF o.proc_time AS c ON o.id = c.id AND o.id2 = c.id2;

As a result, you should see a table with joined records like so: join-result

The msg column shows parameters used with REST call for given JOIN record.

Implementation

HTTP Source

Implementation of an HTTP source connector is based on Flink's TableFunction and AsyncTableFunction classes.
To be more specific we are using a LookupTableSource. Unfortunately Flink's new unified source interface [2] cannot be used for this type of source. Issue was discussed on Flink's user mailing list - https://lists.apache.org/thread/tx2w1m15zt5qnvt924mmbvr7s8rlyjmw

Implementation of an HTTP Sink is based on Flink's AsyncSinkBase introduced in Flink 1.15 [3, 4].

Http Response to Table schema mapping

The mapping from Http Json Response to SQL table schema is done via Flink's Json Format [5].

Breaking changes

TODO

HTTP TableLookup Source

HTTP Sink

[1] https://nightlies.apache.org/flink/flink-docs-release-1.15/docs/dev/table/sql/queries/joins/#lookup-join
[2] https://nightlies.apache.org/flink/flink-docs-release-1.15/docs/dev/datastream/sources/
[3] https://cwiki.apache.org/confluence/display/FLINK/FLIP-171%3A+Async+Sink
[4] https://nightlies.apache.org/flink/flink-docs-release-1.15/api/java/org/apache/flink/connector/base/sink/AsyncSinkBase.html
[5] https://nightlies.apache.org/flink/flink-docs-master/docs/connectors/table/formats/json/
[6] https://nightlies.apache.org/flink/flink-docs-master/docs/dev/table/sqlclient/