Contributing a Patch to a Logstash Plugin
editContributing a Patch to a Logstash Plugin
editThis section discusses the information you need to know to successfully contribute a patch to a Logstash plugin.
Each plugin defines its own configuration options. These control the behaviour of the plugin to some degree. Configuration option definitions commonly include:
- Data validation
- The default value
- Any required flags
Plugins are subclasses of a Logstash base class. A plugin’s base class defines common configuration and methods.
Input Plugins
editInput plugins ingest data from an external source. Input plugins are always associated with a codec. An input plugin
always has an associated codec plugin. Input and codec plugins operate in conjuction to create a Logstash event and add
that event to the processing queue. An input codec is a subclass of the LogStash::Inputs::Base
class.
Table 1. Input API
|
Required. This API sets up resources for the plugin, typically the connection to the external source. |
|
Required. This API fetches or listens for source data, typically looping until stopped. Must handle errors inside the loop. Pushes any created events to the queue object specified in the method argument. Some inputs may receive batched data to minimize the external call overhead. |
|
Optional. Stops external connections and cleans up. |
Codec Plugins
editCodec plugins decode input data that has a specific structure, such as JSON input data. A codec plugin is a subclass of
LogStash::Codecs::Base
.
Table 2. Codec API
|
Identical to the API of the same name for input plugins. |
|
Must be implemented. Used to create an Event from the raw data given in the method argument. Must handle errors. The caller must provide a Ruby block. The block is called with the created Event. |
|
Required. Used to create a structured data object from the given Event. May handle errors. This method calls a block that was previously stored as @on_event with two arguments: the original event and the data object. |
Filter Plugins
editA mechanism to change, mutate or merge one or more Events. A filter plugin is a subclass of the LogStash::Filters::Base
class.
Table 3. Filter API
|
Identical to the API of the same name for input plugins. |
|
Required. May handle errors. Used to apply a mutation function to the given event. |
Output Plugins
editA mechanism to send an event to an external destination. This process may require serialization. An output plugin is a
subclass of the LogStash::Outputs::Base
class.
Table 4. Output API
|
Identical to the API of the same name for input plugins. |
|
Required. Must handle errors. Used to prepare the given event for transmission to the external destination. Some outputs may buffer the prepared events to batch transmit to the destination. |
Process
editA bug or feature is identified. An issue is created in the plugin repository. A patch is created and a pull request (PR) is submitted. After review and possible rework the PR is merged and the plugin is published.
The Community Maintainer Guide explains, in more detail, the process of getting a patch accepted, merged and published. The Community Maintainer Guide also details the roles that contributors and maintainers are expected to perform.
Testing Methodologies
editTest Driven Development
editTest Driven Development, colloquially known as TDD, describes a methodology for using tests to guide evolution of source code. For our purposes, we are only going to use a part of it, that is, before writing the fix - we create tests that illustrate the bug by failing. We stop when we have written enough code to make the tests pass and submit the fix and tests as a patch. It is not necessary to write the tests before the fix, but it is very easy to write a passing test afterwards that may not actually verify that the fault is really fixed especially if the fault can be triggered via multiple execution paths or varying input data.
The RSpec Framework
editLogstash uses Rspec, a Ruby testing framework, to define and run the test suite. What follows is a summary of various sources.
-
Rspec Example
1 # encoding: utf-8 2 require "logstash/devutils/rspec/spec_helper" 3 require "logstash/plugin" 4 5 describe "outputs/riemann" do 6 describe "#register" do 7 let(:output) do 8 LogStash::Plugin.lookup("output", "riemann").new(configuration) 9 end 10 11 context "when no protocol is specified" do 12 let(:configuration) { Hash.new } 13 14 it "the method completes without error" do 15 expect {output.register}.not_to raise_error 16 end 17 end 18 19 context "when a bad protocol is specified" do 20 let(:configuration) { {"protocol" => "fake"} } 21 22 it "the method fails with error" do 23 expect {output.register}.to raise_error 24 end 25 end 26 27 context "when the tcp protocol is specified" do 28 let(:configuration) { {"protocol" => "tcp"} } 29 30 it "the method completes without error" do 31 expect {output.register}.not_to raise_error 32 end 33 end 34 end 35 36 describe "#receive" do 37 let(:output) do 38 LogStash::Plugin.lookup("output", "riemann").new(configuration) 39 end 40 41 context "when operating normally" do 42 let(:configuration) { Hash.new } 43 let(:event) do 44 data = {"message"=>"hello", "@version"=>"1", 45 "@timestamp"=>"2015-06-03T23:34:54.076Z", 46 "host"=>"vagrant-ubuntu-trusty-64"} 47 LogStash::Event.new(data) 48 end 49 50 before(:example) do 51 output.register 52 end 53 54 it "should accept the event" do 55 expect { output.receive event }.not_to raise_error 56 end 57 end 58 end 59 end
Describe blocks (lines 5, 6 and 36 in Example 1).
describe(string){block} -> nil describe(Class){block} -> nil
With RSpec, we are always describing the plugin method behavior. The describe block is added in logical sections and can accept either an existing class name or a string. The string used in line 5 is the plugin name. Line 6 is the register method, line 36 is the receive method. It is a RSpec convention to prefix instance methods with one hash and class methods with one dot.
Context blocks (lines 11, 19, 27 and 41).
context(string){block} -> nil
In RSpec, context blocks define sections that group tests by a variation. The string should start with the word when
and then detail the variation. See line 11. The tests in the content block should should only be for that variation.
Let blocks (lines 7, 12, 20, 28, 37, 42 and 43).
let(symbol){block} -> nil
In RSpec, let
blocks define resources for use in the test blocks. These resources are reinitialized for every test
block. They are available as method calls inside the test block. Define let
blocks in describe
and context
blocks,
which scope the let
block and any other nested blocks.
You can use other let
methods defined later within the let
block body. See lines 7-9, which define the output resource
and use the configuration method, defined with different variations in lines 12, 20 and 28.
Before blocks (line 50).
before(symbol){block} -> nil - symbol is one of :suite, :context, :example, but :all and :each are synonyms for :suite and :example respectively.
In RSpec, before
blocks are used to further set up any resources that would have been initialized in a let
block.
You cannot define let
blocks inside before
blocks.
You can also define after
blocks, which are typically used to clean up any setup activity performed by a before
block.
It blocks (lines 14, 22, 30 and 54).
it(string){block} -> nil
In RSpec, it
blocks set the expectations that verify the behavior of the tested code. The string should not start with
it or should, but needs to express the outcome of the expectation. When put together the texts from the enclosing
describe, context
and it
blocks should form a fairly readable sentence, as in lines 5, 6, 11 and 14:
outputs/riemann #register when no protocol is specified the method completes without error
Readable code like this make the goals of tests easy to understand.
Expect method (lines 15, 23, 31, 55).
expect(object){block} -> nil
In RSpec, the expect method verifies a statement that compares an actual result to an expected result. The expect
method
is usually paired with a call to the to
or not_to
methods. Use the block form when expecting errors or observing for
changes. The to
or not_to
methods require a matcher
object that encapsulates the expected value. The argument form
of the expect
method encapsulates the actual value. When put together the whole line tests the actual against the
expected value.
Matcher methods (lines 15, 23, 31, 55).
raise_error(error class|nil) -> matcher instance be(object) -> matcher instance eq(object) -> matcher instance eql(object) -> matcher instance for more see http://www.relishapp.com/rspec/rspec-expectations/docs/built-in-matchers
In RSpec, a matcher is an object generated by the equivalent method call (be, eq) that will be used to evaluate the expected against the actual values.
Putting it all together
editThis example fixes an issue in the ZeroMQ output plugin. The issue does not require knowledge of ZeroMQ.
The activities in this example have the following prerequisites:
- A minimal knowledge of Git and Github. See the Github boot camp.
- A text editor.
-
A JRuby runtime
environment. The
chruby
tool manages Ruby versions. - JRuby 1.7.22 or later.
-
The
bundler
andrake
gems installed. - ZeroMQ installed.
- In Github, fork the ZeroMQ output plugin repository.
-
On your local machine, clone the fork to a known folder such as
logstash/
. -
Open the following files in a text editor:
-
logstash-output-zeromq/lib/logstash/outputs/zeromq.rb
-
logstash-output-zeromq/lib/logstash/util/zeromq.rb
-
logstash-output-zeromq/spec/outputs/zeromq_spec.rb
-
-
According to the issue, log output in server mode must indicate
bound
. Furthermore, the test file contains no tests.Line 21 of
util/zeromq.rb
reads@logger.info("0mq: #{server? ? 'connected' : 'bound'}", :address => address)
-
In the text editor, set the file encoding and require
zeromq.rb
for the filezeromq_spec.rb
by adding the following lines:# encoding: utf-8 require "logstash/outputs/zeromq" require "logstash/devutils/rspec/spec_helper"
-
The desired error message should read:
LogStash::Outputs::ZeroMQ when in server mode a 'bound' info line is logged
To properly generate this message, add a
describe
block with the fully qualified class name as the argument, a context block, and anit
block.describe LogStash::Outputs::ZeroMQ do context "when in server mode" do it "a 'bound' info line is logged" do end end end
-
To add the missing test, use an instance of the ZeroMQ output and a substitute logger. This examle uses an RSpec feature called test doubles as the substitute logger.
Add the following lines to
zeromq_spec.rb
, afterdescribe LogStash::Outputs::ZeroMQ do
and beforecontext "when in server mode" do
:let(:output) { described_class.new("mode" => "server", "topology" => "pushpull" } let(:tracer) { double("logger") }
-
Add the body to the
it
block. Add the following five lines after the linecontext "when in server mode" do
:
Allow the double to receive |
|
Make the output use the test double. |
|
Set an expectation on the test to receive an |
|
Call |
|
Call |
At the end of the modifications, the relevant code section reads:
# encoding: utf-8 require "logstash/outputs/zeromq" require "logstash/devutils/rspec/spec_helper" describe LogStash::Outputs::ZeroMQ do let(:output) { described_class.new("mode" => "server", "topology" => "pushpull") } let(:tracer) { double("logger") } context "when in server mode" do it "a ‘bound’ info line is logged" do allow(tracer).to receive(:debug) output.logger = tracer expect(tracer).to receive(:info).with("0mq: bound", {:address=>"tcp://127.0.0.1:2120"}) output.register output.do_close end end end
To run this test:
- Open a terminal window
- Mavigate to the cloned plugin folder
-
The first time you run the test, run the command
bundle install
-
Run the command
bundle exec rspec
Assuming all prerequisites were installed correctly, the test fails with output similar to:
Using Accessor#strict_set for specs Run options: exclude {:redis=>true, :socket=>true, :performance=>true, :couchdb=>true, :elasticsearch=>true, :elasticsearch_secure=>true, :export_cypher=>true, :integration=>true, :windows=>true} LogStash::Outputs::ZeroMQ when in server mode a ‘bound’ info line is logged (FAILED - 1) Failures: 1) LogStash::Outputs::ZeroMQ when in server mode a ‘bound’ info line is logged Failure/Error: output.register Double "logger" received :info with unexpected arguments expected: ("0mq: bound", {:address=>"tcp://127.0.0.1:2120"}) got: ("0mq: connected", {:address=>"tcp://127.0.0.1:2120"}) # ./lib/logstash/util/zeromq.rb:21:in `setup' # ./lib/logstash/outputs/zeromq.rb:92:in `register' # ./lib/logstash/outputs/zeromq.rb:91:in `register' # ./spec/outputs/zeromq_spec.rb:13:in `(root)' # /Users/guy/.gem/jruby/1.9.3/gems/rspec-wait-0.0.7/lib/rspec/wait.rb:46:in `(root)' Finished in 0.133 seconds (files took 1.28 seconds to load) 1 example, 1 failure Failed examples: rspec ./spec/outputs/zeromq_spec.rb:10 # LogStash::Outputs::ZeroMQ when in server mode a ‘bound’ info line is logged Randomized with seed 2568
To correct the error, open the util/zeromq.rb
file in your text editor and swap the positions of the words connected
and bound
on line 21. Line 21 now reads:
@logger.info("0mq: #{server? ? 'bound' : 'connected'}", :address => address)
Run the test again with the bundle exec rspec
command.
The test passes with output similar to:
Using Accessor#strict_set for specs Run options: exclude {:redis=>true, :socket=>true, :performance=>true, :couchdb=>true, :elasticsearch=>true, :elasticsearch_secure=>true, :export_cypher=>true, :integration=>true, :windows=>true} LogStash::Outputs::ZeroMQ when in server mode a ‘bound’ info line is logged Finished in 0.114 seconds (files took 1.22 seconds to load) 1 example, 0 failures Randomized with seed 45887
Commit the changes to git and Github.
Your pull request is visible from the Pull Requests section of the original Github repository. The plugin maintainers review your work, suggest changes if necessary, and merge and publish a new version of the plugin.