Skip to content

Latest commit

 

History

History
265 lines (189 loc) · 12 KB

README.md

File metadata and controls

265 lines (189 loc) · 12 KB

KeenClient-Scala

Build Status Coverage Status Maven Central Badge Scaladoc Badge


The official asynchronous Scala client for the Keen IO API.

Note: This library is in early development and does not implement all of the features of the Keen API. It is pre-1.0 in the Semantic Versioning sense: public interfaces may change without backwards compatibility. We will try to minimize breaking changes, but please consult the changelog when updating to a new release version.

Additional API features will be added over time. Contributions are welcome!


Get It

Artifacts for keen-client-scala are hosted on Maven Central. You can use them in your project with SBT thusly:

libraryDependencies += "io.keen" %% "keenclient-scala" % "0.7.0"

Note that we publish artifacts for Scala 2.10, 2.11 and 2.12, so you can either use %% to automatically pick the correct version or specify them explicitly with something like:

libraryDependencies += "io.keen" % "keenclient-scala_2.10" % "0.7.0"

Configuration

The client has a notion of access levels that reflect the Keen IO API key security model. These are represented by Scala traits called Reader, Writer, and Master. According to the level of access that your application requires, you must mix the appropriate trait(s) into your client instance when creating it, and configure your corresponding API keys. This is demonstrated in the examples.

Configuration is supported by the Typesafe config library, offering all the flexibility you could wish to provide settings through a file, environment variables, or programmatically. We recommend environment variables for your API keys at very least, to avoid storing credentials in source control. To that end, the following will be honored by default if set:

  • KEEN_PROJECT_ID
  • KEEN_READ_KEY
  • KEEN_WRITE_KEY
  • KEEN_MASTER_KEY

To configure with a file, it must be on the classpath, customarily called application.conf though you may use others--see the Typesafe config documentation for all the options and details of the file format. Our reference.conf reflects all of the settings you can configure and their default values.

For advanced needs, you may provide your own custom Config object by simply passing it to the client constructor:

import io.keen.client.scala.Client

val keen = new Client(config = myCustomConfigObject)

When using environment variables, you might like sbt-dotenv in your development setup (install it as a global plugin, and chmod 600 your .env files that contain credentials!). In production, a good service manager can set env vars for app processes with ease. On Heroku you'll be right at home.

Settings

  • keen.project-id: Your project ID.
  • keen.optional.read-key: Your project read key.
  • keen.optional.write-key: Your project write key.
  • keen.optional.master-key: Your project master key.
  • keen.queue.batch.size: Number of events to include in each batch sent by sendQueuedEvents(). Default is 500.
  • keen.queue.batch.timeout: Duration that each batch sent by sendQueuedEvents() should wait before the request times out. Default is 5 seconds.
  • keen.queue.max-events-per-collection: Maximum number of events to store for each collection. Old events are purged from the queue to make room for new events when the size of the queue exceeds this number. Default is 10000.
  • keen.queue.send-interval.events: Automatically send all queued events every time the queue reaches this number. Minimum is 100, maximum is 10000, and default is 0.
  • keen.queue.send-interval.duration: Automatically send all queued events at a specified interval. Minimum is 60 seconds, maximum is 3600, and default is 0.
  • keen.queue.shutdown-delay: Duration before client stops attempting to send events scheduled to be sent at a specific interval. Default is 30 seconds.

Use It - A Quick Taste

import io.keen.client.scala.{ Client, Writer }

// Assumes you've configured a write key as explained in Configuration above
val keen = new Client with Writer

// Publish an event!
keen.addEvent(
  collection = "collectionNameHere",
  event = """{"foo": "bar"}"""
)

// Publish lots of events!
keen.addEvents(someEvents)

// Responses are Futures - handle errors!
val resp = keen.addEvent(
  collection = "collectionNameHere",
  event = """{"foo": "bar"}"""
)

resp onComplete {
  case Success(r) => println(resp.statusCode)
  case Failure(t) => println(t.getMessage) // A Throwable
}

// Or using map
resp map { println("I succeeded!") } getOrElse { println("I failed :(") }

Batching Write Requests

The standard client is asynchronous, but it does make a discrete API call for each individual (addEvent) or bulk (addEvents) event write. Depending on your usage patterns, batching events to make fewer HTTP requests to Keen IO may be a significant optimization.

Though you can certainly implement your own queueing and batching via addEvents, the library includes automated batching support sufficient for many use cases. This is available in the BatchWriterClient, shown below.

import io.keen.client.scala.BatchWriterClient

val keen = new BatchWriterClient

// Queue an event
keen.queueEvent("collectionNameHere", """{"foo": "bar"}""")

Queuing is handled via an in-memory queue that lives as long as the Client does. The behavior of the queue and automated sending of events is configurable in conf/application.conf as outlined below.

Sending queued events manually

keen.sendQueuedEvents()

Sending queued events every time the queue reaches 100 events

Set keen.queue.send-interval.events equal to 100 in conf/application.conf.

Sending queued events every 5 minutes

Set keen.queue.send-interval.duration equal to 5 minutes in conf/application.conf.

Note that send-interval.events takes precedence when both send-interval.events and send-interval.duration contain values greater than zero.

Using batch sizes

Setting a specific batch size will help optimize your experience when sending events. It's recommended that you set keen.queue.batch.size to something that makes sense for your application (default is 500). Note that a batch size of 5000 is the upper bound of what you should shoot for. Anything higher and your request has a good chance of being rejected due to payload size limitations.

Failed events

Events that fail to be sent for whatever reason (payload rejection, network issue, etc.) are not removed from the queue. They will remain in the queue until they are either manually removed or the client shuts down.

Shutdown

BatchWriterClient will attempt one last time to send all queued events when shutdown() is called just in case you forget to send the events yourself. Make sure you call shutdown() before you exit otherwise all events that remain in the queue upon termination will be lost.

Caveats and Tradeoffs

Bear in mind that BatchWriterClient uses an in-memory event queue implementation by default and has an API that favors fire-and-forget usage; it is difficult to handle errors in a granular fashion and ensure that writes are never lost if queue bounds are exceeded, etc. If you're recording something like internal metrics, the possibility of occasional missing data points may be an acceptable tradeoff for the optimization of batching.

If it is business-critical that some of your writes are never lost, you may wish to use the discrete API so that you can handle the Future result of each call with exacting care. Or if you require batching, you can implement a custom persistent EventStore queue for control over retries, and/or your own batching Client subclass tailored to your needs.

BatchWriterClient is a Client with Writer, so you can call the discrete addEvent/addEvents methods on an instance and as usual these send immediately without queueing. Thus you can selectively use batching for some types of writes and make discrete calls for more critical ones.

Dependencies

The client's default HTTP adapter is built on the spray HTTP toolkit, which is Akka-based and asynchronous. A Dispatch-based adapter is also available. At this time spray (and thus Akka) is a hard dependency, but if there is demand we may consider designating it as "provided" so that you may opt for the Dispatch adapter (or a custom one for your preferred HTTP client) and avoid pulling in Akka dependencies if you wish, or to avoid version conflicts. Please share your feedback if you find the spray deps burdensome.

With either adapter, API calls will return a uniform Future[Response] type, where Response is an io.keen.client.scala.Response. Instances have statusCode and body attributes that you may inspect to act on errors. An example of choosing the Dispatch adapter is shown below.

The client also depends on grizzled-slf4j for logging.

It is cross-compiled for 2.10, 2.11 and 2.12 Scala versions. If you are interested in support for other versions or discover any binary compatibility problems, please share your feedback.

Using the Dispatch adapter

To use the Dispatch HTTP adapter instead of the Spray default, specify the following override when instantiating a Client instance:

import io.keen.client.scala.{ Client, HttpAdapterDispatch }

val keen = new Client {
  override val httpAdapter = new HttpAdapterDispatch
}

BatchWriterClient works the same way.

JSON

Presently this library does not do any JSON parsing. It works with strings only. It is assumed that you will parse the JSON returned and pass stringified JSON into any methods that require it. Feedback is welcome!

We understand that Scala users value the language's strong type system. Again, we wish to avoid unwanted dependencies given that there are so many JSON parsing libraries out there. We'd eventually like to offer rich types through JSON adapters with optional deps.

Hack On It

Unit tests can be run with the standard SBT commands test, testQuick, etc.

The test suite includes integration tests which require keys and access to Keen IO's API. If you have set keys through environment variables or configuration as described above, you may run these with:

$ sbt it:test

Only use a dedicated dummy account for this purpose, data could be destroyed that you didn't expect!