Erlang NSQ client

This library implements a client to the NSQ Messaging Queue it does it's best to follow the guidelines provided by the project.

Missing features

• Support for direct non discovery connections.
• TCP discovery (using HTTP currently).
• Proper behavior when max_in_flight is smaller then the total number of connections.
• Support for UTF8 hostnames (seems to be a Erlang limitation).
• Protocol negation with IDENTIFY.
  • max_ready_count delegation from the server.

OTP Compatibility

OTP 21.0 or above is required since 0.3.0, If you need a version of ensq which runs on an older OTP release, we suggest you use ensq 0.2.0.

Design

The Erlang process model fits the nsq client concept quite well so:

• The top level 'process' are one (or more) topics and discovery server combinations.
  • they can have none, one or multiple a channel + handler combinations.
    • each discovered producer spawns one process for channel/handler that opens a own TCP connection.
  • they can have none, one or multiple target hosts or host groups for PUB
    • Each PUB host spawns one sender.
    • Messages are send in a round robin principle.
    • Messages to host groups are send to all members.

                                 topic*
                                 /     \
                                /       \
                               /         \
                              /           \
                      discovery_server     \
                           /    \           \
                          /      \          target*(PUB)
                         /        \
                        /          \
                      server      server2
                       / \          |
                      /   \         |
                     |     |        |
                 channel* channel* channel*


* - seperate process

API

Using ensq:init

One way to connect to nsq is using

Setting up topics separately

%% Setting up a topic separately.
DiscoveryServers = [{"localhost", 4161}], %% Discovery Server
Channels = [{<<"test">>,                  %% Channel name
			  ensq_debug_callback},       %% Callback Module
			  {<<"test2">>, ensq_debug_callback}],
ensq_topic:discover(
  test,                                   %% Topic
  DiscoveryServers,                       %% Discovery servers to use
  Channels,                               %% Channels to join.
  [{"localhost", 4150}]).                 %% Targets for SUBing


%% Sending a message to a topic
ensq:send(test, <<"hello there!">>).

Non registered channels

By default channels are registered processes that way it's possible to do things like ensq:send(test, ...) but for some situations registering the process and creating an atom for the name is not desirable to prevent poisoning the atom table. A dynamically created and destroyed channel would be an example of this. Passing a binary instead of a atom binary to ensq_topic:discover/4 will solve this issue but loose the advantage of being able to call the process by name, it still is possible to either save the PID on creation or retrieve it from ensq:list/1.

Configuration

• max_in_flight maximum number of messages the server can handle at once (300).
• discovery_inteval Delay between discovery requests in milliseconds (60000).
• discover_jitter Jitter in percent added or subtracted of the discovery interval, a value of 10 gives a total jitter of 20% by adding between -10% and +10% (10).
• max_retry_delay The maximum delay for retries in milliseconds (10000).
• retry_inital Initial retry value in milliseconds (1000).
• retry_inc_type The algorithm the retry interval progresses, it can either be linear making the retry delay Attempt*retry_inital milliseconds or quadratic making the retry delay Attempt*Attempt*retry_inital