Change Log

Changes between Kehaar 1.0.2 and 1.0.3

Bespoke streaming response queues are not redeclared during queue recovery. These queues are considered to be "owned" by the requester, but they have to be created by the responder. By default, connections that declare queues will recover them unless the same connection deletes them.

Changes between Kehaar 1.0.1 and 1.0.2

Fixed a bug where if a consumer was slow to process streaming results, the result stream could be truncated.

Changes between Kehaar 1.0.0 and 1.0.1

kehaar.configured can optionally use a var instead of a function for incoming-services, incoming-events, and incoming-jobs, making repl-based development simpler. Set the KEHAAR_DEBUG environment variable to true to use this behavior.

Changes between Kehaar 0.11.4 and 1.0.0

• Potentially breaking change: Kehaar no longer pulls in core.async as a dependency. Since using core.async directly in your application code has always been a requirement when using kehaar, we have switched it to :scope "provided". You must add a recent version of org.clojure/core.async to your app's dependencies to use kehaar 1.0.0+.

• The kehaar.rabbitmq namespace has a new fn named dissoc-blank-config-params-with-defaults. This is an optional helper fn whose purpose is to assist with not clobbering default values with emtpy strings and nils in your RabbitMQ connection config map (i.e. the arg you'll pass to connect-with-retries). It dissoc's any value for the following keys:

  • :username
  • :hostname
  • :vhost
  • :host
  • :port
  • :requested-heartbeat
  • :requested-channel-max
  • :connection-timeout
  • :automatically-recover

  ...where that value is either nil or a string for which clojure.string/blank? returns true.

• Kehaar will output additional helpful information if an exception is thrown while trying to realize a symbol from a kehaar.configured handler fn name. Specifically it will tell you the name of the symbol it was trying to realize.

Changes between Kehaar 0.11.3 and 0.11.4

The EDN handling changes introduced in 0.11.0 have been ripped out. In practice they caused more problems than they solved. Only Clojure core developers can fix EDN in Clojure. Here's hoping they do so.

Changes between Kehaar 0.11.2 and 0.11.3

kehaar.rabbitmq/connect-with-retries now retries to connect on any Exception subclass thrown rather than trying to catch specific ones.

Changes between Kehaar 0.11.1 and 0.11.2

Updated Langohr to 4.1.0

Changes between Kehaar 0.11.0 and 0.11.1

Initialization functions in kehaar.configured can now also take actual function and channel values where only symbols had been expected before.

Changes between Kehaar 0.10.5 and 0.11.0

Rather than spinning up the maximum number of desired processing threads for each handler, kehaar now spins up threads as needed for the work coming in. It then allows them to be released if the workload decreases for a time. It still respects the thread count param and will never spin up more threads than that. This should prevent leaving idle threads hanging around forever just because they might be needed for workload spikes.

EDN reading and writing is now safer. Kehaar encodes all values going over RabbitMQ into EDN using clojure.core/pr-str and then calls clojure.edn/read-string on them when coming off of the queue on the other side. In 0.11.0, kehaar will now convert instances of java.util.regex.Pattern to their string representations before encoding them into EDN. This is necessary because clojure.edn/read-string will choke on the #"..." form they otherwise get encoded as. On the reading side, instead of clojure.edn/read-string's behavior of throwing exceptions on unknown tags, kehaar will now simply read in the tagged values as is. This should prevent exceptions coming from kehaar because of regexes or tagged representations without registered data readers in your payloads.

Changes between Kehaar 0.10.4 and 0.10.5

Updated dependencies:

• clojure from 1.7.0 to 1.8.0
• core.async from 0.2.385 to 0.3.443
• langohr from 3.2.0 to 3.7.0

Changes between Kehaar 0.10.3 and 0.10.4

kehaar.rabbitmq/connect-with-retries can now retry connection failures that throw java.util.concurrent.TimeoutException exceptions in addition to the java.net.ConnectException exceptions it has always been able to retry.

Changes between Kehaar 0.10.2 and 0.10.3

A Clojure 1.9-friendly ns form in kehaar.response-queues.

Changes between Kehaar 0.10.1 and 0.10.2

When RabbitMQ renames response queues kehaar sets up for external services after a node restart, kehaar now reacts, ensuring that future calls to the external service use the new name for the reply-to queue.

Changes between Kehaar 0.10.0 and 0.10.1

Added prefetch limits to rabbit channels created when setting up streaming external services.

Changes between Kehaar 0.9.0 and 0.10.0

A couple of improvements to streaming:

• A streaming handler function may now return a core.async channel that results will be placed on.
• Streaming responders now send the first threshold values over the shared reply queue, switching over to a bespoke queue after threshold + 1 values.

Due to changes in the messages that coordinate streaming behind the scenes, both the service producing streaming responses and its consumers must use Kehaar >= 0.10.0.

Changes between Kehaar 0.8.1 and 0.9.0

Added jobs. See README, kehaar.jobs, and kehaar.configured for details.

Changes between Kehaar 0.8.0 and 0.8.1

Added thread count option to init-incoming-event!.

Changes between Kehaar 0.7.2 and 0.8.0

kehaar.configured

A new namespace with functions that allow for a more declarative description of your system.

Changes between Kehaar 0.7.1 and 0.7.2

Fixed the 3-arity version of external-service-fire-and-forget to call the 5-arity version of itself.

Changes between Kehaar 0.7.0 and 0.7.1

Fire and forget

Added external-service-fire-and-forget and async->fire-and-forget-fn, used much like their counterparts without "fire-and-forget" in their names to send messages to an external service but without waiting for a response.

Added an optional argument to incoming-service called ignore-no-reply-to, which causes it to no longer log warnings when a message comes without the :reply-to metadata set. Good for decreasing the noise in a service you expect to be used by the "fire and forget" functions above.

Changes between Kehaar 0.6.0 and 0.7.0

Configurable number of threads to handle messages with

start-responder!, start-streaming-responder!, and start-event-handler! can now take an extra argument: a number of threads to use to pull messages from their input channels. The threads will be created and be waiting on new messages immediately. The default number of threads is 10.

Sleeping on nack

When rabbit=>async nacks a message, it sleeps for one second before attempting to take another message. Eventually this will be configurable.

Together with the limited number of threads for message handlers, this better allows for backpressure from core.async to RabbitMQ.

Changes between Kehaar 0.5.0 and 0.6.0

Streaming responders

Added start-streaming-responder! and streaming-external-service to the kehaar.wire-up namespace for starting and consuming streaming responses. Those functions are used for streaming responders in place of start-responder! and external-service respectively.

A streaming responder function merely needs to return a sequence (lazy, if you like) and the values from that sequence will be sent across RabbitMQ to a core.async channel on the consumer's side.

Changes between Kehaar 0.4.0 and 0.5.0

kehaar.rabbitmq/connect-with-retries

Added kehaar.rabbitmq/connect-with-retries function to make connecting to RabbitMQ brokers more robust.

Backpressure

This is a major change. It renames and reworks the existing functionality.

The main motivation for this was to make kehaar more robust when faced with errors. EDN parse errors would cause the entire handler to crash. And kehaar would read as many messages as it could at a time, potentially overloading the server with no recourse for backpressure.

This rewrite solves those two problems while renaming and reworking the abstractions. The main difference is that the functions which pump messages to and from rabbit <=> core.async now pass the payload and metadata. This simplified the code a lot since we often need the metadata.

Event handlers and service responders now run in their own threads. call start-event-handler! and start-responder! in server initialization to start those. The return value of event handlers is ignored, but responders can return any value or a channel which will contain the value. This lets things remain asynchronous.

This is a breaking change. Many "lower-level" function in core are very different. Other functions have been moved, renamed, and their arguments are different.

Example project

An example project has been added at /example, to demonstrate how to use the various functions in the wire-up namespace.

Changes between Kehaar 0.3.0 and 0.4.0

kehaar.core/ch->response-fn

In 0.2.0 ch->response-fn started returning promises instead of core.async channels. That is now reverted back to core.async channels. This is a breaking change.

kehaar.wire-up

The kehaar.wire-up namespace contains a higher-level interface for declaring and setting up queues.

Updated Clojure dependency to 1.7.0.

It came out.

Changes between Kehaar 0.2.1 and 0.3.0

Add kehaar.core namespace

Since single-level namespaces are not recommended in Clojure, we have moved the code that was in kehaar to a new kehaar.core ns. This is a breaking change. You will need to update all your (ns ... (:require [kehaar])) forms in your code to look more like this instead: (ns ... (:require [kehaar.core])).

Ack-on-take in rabbit->async

Previously kehaar auto-acked every incoming RabbitMQ message. Now it acks only when something successfully consumes the message from the core.async channel that it is forwarded to.

Logging

There are now debug-level log messages when Kehaar consumes a RabbitMQ message and when it forwards them on to core.async channels.

There are also warn-level log messages when it tries to take from a closed core.async channel.

This adds a dependency on clojure.tools.logging 0.3.1.

Updated dependencies

Use Clojure 1.7.0-RC2

Changes between Kehaar 0.2.0 and 0.2.1

nil? checks when taking values from async channels

Pulling from a core.async channel will return nil if the channel is closed, so now we check for that and stop trying to handle those messages and stop trying to pull more.

Updated dependencies

Use Clojure 1.7.0-beta3 and Langohr 3.2.0.

Changes between Kehaar 0.1.0 and 0.2.0

kehaar/async->rabbit

kehaar/async->rabbit no longer declares the queue it operates on. This is a breaking change. Queues must now be already declared. This allows, for example, the queue to be a server-named, exclusive, auto-deleted queue.

kehaar/rabbit->async

The old kehaar/rabbit->async function has been renamed kehaar/rabbit->async-handler-fn, and kehaar/rabbit->async now takes the RabbitMQ queue and the async channel, handling the subscription for you. This is a breaking change.

rabbit->async and async->rabbit now appropriately mirror the other.

kehaar/rabbit->async-handler-fn

kehaar/rabbit->async-handler-fn now blocks if the async channel's buffer is full, providing the opportunity for some back pressure.

kehaar/ch->response-fn

kehaar/ch->response-fn now returns promises instead of async channels for the caller to wait on. This is a breaking change.

kehaar/wire-up-service

kehaar/wire-up-service no longer declares the queue it operates on either. This is a breaking change. Queues must now already be declared. Additionally, internally, it uses a server-named, exclusive, auto-deleted queue.

kehaar/responder

kehaar/simple-responder has been renamed kehaar/fn->handler-fn and a new function kehaar/responder has been made which takes a RabbitMQ channel and queue and a function to apply to all messages, replying on the reply-to queue with the result. This is a breaking change.

Tests using RabbitMQ

There are now tests which use RabbitMQ, however they are not run by default with lein run. In order to run the RabbitMQ tests, start rabbitmq-server with its default configuration and run lein test :rabbit-mq. To run all tests, run lein test :all.

Travis CI has been updated to run those tests as well.