If you've already installed https://zookeeper.apache.org[Zookeeper], http://kafka.apache.org/[Kafka], and {link-kafka-docs}.html#connect[Kafka Connect], then using one of Debezium's connectors is easy.
Simply download one or more connector plugin archives (see below), extract their files into your Kafka Connect environment, and add the parent directory of the extracted plugin(s) to https://docs.confluent.io/current/connect/userguide.html#installing-plugins[Kafka Connect's plugin path].
If not the case yet, specify the plugin path in your worker configuration (e.g. _connect-distributed.properties_) using the `plugin.path` property.
As an example, let's assume you have downloaded the Debezium MySQL connector archive and extracted its contents to _/kafka/connect/debezium-connector-mysql_.
Then you'd specify the following in the worker config:
[source]
----
plugin.path=/kafka/connect
----
Restart your Kafka Connect process to pick up the new JARs.
NOTE: All above links are to nightly snapshots of the Debezium master branch. If you are looking for non-snapshot versions, please select the appropriate version in the top right.
If immutable containers are your thing, then check out https://quay.io/organization/debezium[Debezium's container images] (https://hub.docker.com/r/debezium/[alternative source] on DockerHub) for Apache Kafka, Kafka Connect and Apache Zookeeper, with the different Debezium connectors already pre-installed and ready to go. Our xref:tutorial.adoc[tutorial] even walks you through using these images, and this is a great way to learn what Debezium is all about.
Of course you also can run Debezium on Kubernetes and xref:operations/openshift.adoc[OpenShift].
By default, the directory _/kafka/connect_ is used as plugin directory by the Debezium Docker image for Kafka Connect.
So any additional connectors you may wish to use should be added to that directory.
Alternatively, you can add further directories to the plugin path by specifying the `KAFKA_CONNECT_PLUGINS_DIR` environment variable when starting the container
When using the Docker image for Kafka Connect provided by Confluent, you can specify the `CONNECT_PLUGIN_PATH` environment variable to achieve the same.
Not that Java 8 or later is required to run the Debezium connectors.
If you want to try latest and fresh or verify a bug fix you are interested in, then use plugins from https://oss.sonatype.org/content/repositories/snapshots/io/debezium/[oss.sonatype.org] or view the xref:master@install.adoc[master branch] version of this document for direct links to each connector's plugin artifact.
(if you wish to only keep the _last_ change event for a given record);
in this case the `min.compaction.lag.ms` and `delete.retention.ms` topic-level settings in Apache Kafka should be configured,
so that consumers have enough time to receive all events and delete markers;
specifically, these values should be larger than the maximum downtime you anticipate for the sink connectors,
e.g. when updating them
** Replicated in production
** Single partition
*** You can relax the single partition rule but your application must handle out-of-order events for different rows in database (events for a single row are still totally ordered). If multiple partitions are used, Kafka will determine the partition by hashing the key by default. Other partition strategies require using SMTs to set the partition number for each record.
== Using the Debezium Libraries
Although Debezium is intended to be used as turnkey services, all of JARs and other artifacts are available in http://search.maven.org/#search%7Cga%7C1%7Cg%3A%22io.debezium%22[Maven Central].
We do provide a small library so applications can xref:operations/embedded.adoc[embed any Kafka Connect connector] and consume data change events read directly from the source system. This provides a light weight system (since Zookeeper, Kafka, and Kafka Connect services are not needed), but as a consequence it is not as fault tolerant or reliable since the application must manage and maintain all state normally kept inside Kafka's distributed and replicated logs. It's perfect for use in tests, and with careful consideration it may be useful in some applications.