Skip to content

Latest commit

 

History

History
190 lines (154 loc) · 9.35 KB

README.md

File metadata and controls

190 lines (154 loc) · 9.35 KB
Raw

DSH tenant example

This example shows how tenants on the DSH platform should interact with the DSH data streams. The particular points to pay attention to are:

  • getting Kafka configuration from the kafka configuration service
  • key and data envelopes for serialization/deserialization on public streams (the stream.<something> streams).
  • partitioning for public streams.

Kafka configuration

On startup, every DSH container that wants to interact with the data streams needs to fetch its kafka configuration from the kafka configuration service (pikachu.dsh.marathon.mesos).

Fetching the Kafka configuration must happen on container startup. It depends on a limited-lifetime token that is configured in the container's environment on startup. If you wait for the token to expire, there is no way for the container to ever fetch its Kafka configuration.

The full flow for fetching the Kafka configuration is a rather involved multi-step process. Documenting it here in its entirety will lead us too far, but the good news is that this example contains a fully working shell script that does all the work for you. You can copy it into your own containers and invoke it from the entrypoint.

See the get_signed_certificate.sh script.

Consumer groups

In order to consume data from Kafka, you need to set a consumer group for your consumers. How you choose your consumer group has a profound impact on the operation of your program:

  • if you choose a unique consumer group name, your program will see all data from all partitions of the topics you subscribe to.
  • if you share a consumer group with other programs, each program will see a subset of the total data on the kafka topics you subscribe to. Kafka balances the partitions among all members of a consumer group automatically.

Consumer group enforcement

The DSH platform enforces restrictions on the consumer groups each container is allowed to join. When you retrieve the Kafka configuration from the Kafka configuration service, it will include a list of consumer group names you are allowed to use. To cater for the two options described above, the list is split in two parts:

  • private consumer groups are unique to this particular container, and can be used to consume all data on a Kafka topic.
  • shared consumer groups are shared among all instances of this container, to allow you to spin up multiple instances and balance the load of a (set of) Kafka topics among them.

Note that the DSH platform does not currently offer provisions for sharing a consumer group among instances of different containers.

See below for a description of how the consumer group names are encoded in the datastreams.properties file.

Setting the consumer group

To set a specific consumer group, pass the group.id property to the Kafka Consumer constructor.

Configuration data

The configuration data fetched by the get_signed_certificates.sh script consists of an SSL certificate to be used for connection to Kafka on the one hand, and a Java properties file that contains the necessary security settings, a list of bootstrap servers and a description of the streams you have access to on the other hand.

On startup, load the properties from the datastreams.properties file generated by the script, and pass them along to any Kafka consumer or producer you create.

To learn more about the streams you have access to, you can inspect the contents of the properties file. All configuration entries for a given stream are prefixed with datastream.<streamname>. Note that the streamname includes the stream type prefix, so concrete examples are datastream.stream.foo for a public stream called "foo" and datastream.internal.bar for an internal stream named "bar".

For each stream, the following configuration values are defined:

  • datastream.<streamname>.cluster: identifies the Kafka cluster carrying this data stream (right now, the only supported value is "tt")
  • datastream.<streamname>.partitioner: partitioning scheme used on this stream (see below). Allowed values are: "default-partitioner", "topic-level-partitioner" and "topic-char-depth-partitioner".
  • datastream.<streamname>.canretain: indicates whether this stream supports retained MQTT messages. This is only applicable to public streams. Allowed values are "true" and "false"
  • datastream.<streamname>.partitions: number of Kafka partitions for the stream
  • datastream.<streamname>.replication: Kafka replication factor for the stream
  • datastream.<streamname>.partitioningDepth: the partitioning depth configuration parameter for the topic level and character depth partitioners. This value has no meaning for the default partitioner.
  • datastream.<streamname>.read: the regex pattern you can use to subscribe to this stream on Kafka. You should always subscribe with a regex pattern, to ensure you subscribe to all Kafka topics that underly this data stream. If this configuration value is empty, it means you do not have read access to this stream.
  • datastream.<streamname>.write: the Kafka topic name you can use to produce data on the stream. If this configuration value is empty, it means that you do not have write permissions for this stream.

In addition to the stream descriptions, the datastreams.properties file defines the following helpful values:

  • consumerGroups.private: a comma-separated list of "private" consumer group names your container is allowed to use. A private consumer group is one that is not shared with any other container in the system, even other instances of the same container.
  • consumerGroups.shared: a comma-separated list of "shared" consumer group names your container is allowed to use. A shared consumer group is one that is available for use in all instances of the same container.
  • envelope.streams: a comma-separated list of public streams that do NOT yet have envelopes. Use this value to configure your (de)serializers to expect the right kind of data (enveloped or non-enveloped) on each topic. Note that the list contains only the middle part of the stream name, not the stream prefix or tenant name suffix.

Envelopes on public data streams

Public data streams (the stream.<something> streams) have specific requirements with respect to the structure of the key and payload values. In essence, the "plain" key and payload are wrapped in envelope structures. The envelope structures are defined as Protobuf messages. The definition can be found in src/main/proto/envelope.proto.

Note: at this moment, the DSH platform is in a transition phase where the use of envelopes on the public topics is gradually phased in. Right now, envelopes MUST be used on all Kafka topics ending in .dsh, and MAY NOT be used on any other topics. The example code deals with this transparently, and allows you to write your application code as if envelopes are mandatory everywhere.

Note: while the usage of envelopes on the public streams is strictly regulated and enforced, you are free to choose whether you want to use envelopes on the other stream types (scratch, internal). You are not forced to do it, and also not forbidden to if you feel it is useful to do so.

Envelope definition

The envelope definition can be found here.

Envelope usage

The Kafka serializers and deserializers for enveloped keys and data can be found in the KeyEnvelopeSerializer, KeyEnvelopeDeserializer, DataEnvelopeSerializer and DataEnvelopeDeserializer classes. As noted above, these implementations deal automatically with enveloped (.dsh) and non-enveloped (all other) topics. If you reuse these implementations in your own containers, take care to remove this logic when envelopes become mandatory for all public data streams.

Convenience methods

For your convenience, the serializer classes define some utility methods that facilitate wrapping of plain keys and payloads in the envelopes:

  • KeyEnvelopeSerializer.setIdentifier() is used to set the static identifier (tenant id and free-form publisher id) for every message produced by this container.
  • KeyEnvelopeSerializer.wrap() wraps a plain string key in an envelope, with optional qos and retain settings (defaults are QoS 0 and non-retained).
  • DataEnvelopeSerializer.wrap() wraps a plain byte array payload in a data envelope.

Partitioning for public streams

If your container publishes data on a public stream, it needs to abide by the stream contract, which defines (amongst others) the partitioning scheme for this stream.

Currently, there are three supported partitioning schemes:

  • topic level partitioning: the first n levels of the topic path (i.e. the plain string key) are hashed using the Murmur2 hashing scheme
  • character level partitioning: the first n characters of the topic path are hashed using the Murmur2 hashing scheme
  • default partitioning: Murmur2 hash of the entire topic path

As of now, the only partitioning scheme that is used in practice on the public streams is the topic level partitioner. Therefore, that is the only partitioner for which we have included an example implementation, in the TopicLevelPartitioner class.

The TopicLevelPartitioner class will figure out the partitioning depth (the aforementioned n parameter) from the configuration values you fetched from the Kafka configuration service, and passed to your Kafka producer constructor.