Elasticsearch::Transport

This library is part of the elasticsearch-ruby package; please refer to it, unless you want to use this library standalone.


The elasticsearch-transport library provides a low-level Ruby client for connecting to an Elasticsearch cluster.

It handles connecting to multiple nodes in the cluster, rotating across connections, logging and tracing requests and responses, maintaining failed connections, discovering nodes in the cluster, and provides an abstraction for data serialization and transport.

It does not handle calling the Elasticsearch API; see the elasticsearch-api library.

The library is compatible with Ruby 1.9 or higher and with all versions of Elasticsearch since 0.90.

Features overview:

  • Pluggable logging and tracing
  • Pluggable connection selection strategies (round-robin, random, custom)
  • Pluggable transport implementation, customizable and extendable
  • Pluggable serializer implementation
  • Request retries and dead connections handling
  • Node reloading (based on cluster state) on errors or on demand

For optimal performance, use a HTTP library which supports persistent ("keep-alive") connections, such as patron or Typhoeus. Just require the library (require 'patron') in your code, and it will be automatically used.

Currently these libraries will be automatically detected and used:

Note on Typhoeus: You need to use v1.4.0 or up since older versions are not compatible with Faraday 1.0.

For detailed information, see example configurations below.

Installation

Install the package from Rubygems:

gem install elasticsearch-transport

To use an unreleased version, either add it to your Gemfile for Bundler:

gem 'elasticsearch-transport', git: 'git://github.com/elasticsearch/elasticsearch-ruby.git'

or install it from a source code checkout:

git clone https://github.com/elasticsearch/elasticsearch-ruby.git
cd elasticsearch-ruby/elasticsearch-transport
bundle install
rake install

Example Usage

In the simplest form, connect to Elasticsearch running on http://localhost:9200 without any configuration:

require 'elasticsearch/transport'

client = Elasticsearch::Client.new
response = client.perform_request 'GET', '_cluster/health'
# => #<Elasticsearch::Transport::Transport::Response:0x007fc5d506ce38 @status=200, @body={ ... } >

Full documentation is available at http://rubydoc.info/gems/elasticsearch-transport.

Configuration

The client supports many configurations options for setting up and managing connections, configuring logging, customizing the transport library, etc.

Setting Hosts

To connect to a specific Elasticsearch host:

Elasticsearch::Client.new host: 'search.myserver.com'

To connect to a host with specific port:

Elasticsearch::Client.new host: 'myhost:8080'

To connect to multiple hosts:

Elasticsearch::Client.new hosts: ['myhost1', 'myhost2']

Instead of Strings, you can pass host information as an array of Hashes:

Elasticsearch::Client.new hosts: [ { host: 'myhost1', port: 8080 }, { host: 'myhost2', port: 8080 } ]

NOTE: When specifying multiple hosts, you probably want to enable the retry_on_failure option to perform a failed request on another node (see the Retrying on Failures chapter).

Common URL parts -- scheme, HTTP authentication credentials, URL prefixes, etc -- are handled automatically:

Elasticsearch::Client.new url: 'https://username:[email protected]:4430/search'

You can pass multiple URLs separated by a comma:

Elasticsearch::Client.new urls: 'http://localhost:9200,http://localhost:9201'

Another way to configure the URL(s) is to export the ELASTICSEARCH_URL variable.

The client will automatically round-robin across the hosts (unless you select or implement a different connection selector).

Default port

The default port is 9200. Please specify a port for your host(s) if they differ from this default. Please see below for an exception to this when connecting using an Elastic Cloud ID.

Connect using an Elastic Cloud ID

If you are using Elastic Cloud, you can provide your cloud id to the client. You must supply your username and password separately, and optionally a port. If no port is supplied, port 443 will be used.

Note: Do not enable sniffing when using Elastic Cloud. The nodes are behind a load balancer so Elastic Cloud will take care of everything for you.

Elasticsearch::Client.new(cloud_id: 'name:bG9jYWxob3N0JGFiY2QkZWZnaA==', user: 'elastic', password: 'changeme')

Authentication

You can pass the authentication credentials, scheme and port in the host configuration hash:

Elasticsearch::Client.new hosts: [
  { host: 'my-protected-host',
    port: '443',
    user: 'USERNAME',
    password: 'PASSWORD',
    scheme: 'https'
  } ]

... or simply use the common URL format:

Elasticsearch::Client.new url: 'https://username:[email protected]:9200'

To pass a custom certificate for SSL peer verification to Faraday-based clients, use the transport_options option:

Elasticsearch::Client.new url: 'https://username:[email protected]:9200',
                          transport_options: { ssl: { ca_file: '/path/to/cacert.pem' } }

You can also use API Key authentication:

Elasticsearch::Client.new(
  host: host,
  transport_options: transport_options,
  api_key: credentials
)

Where credentials is either the base64 encoding of id and api_key joined by a colon or a hash with the id and api_key:

Elasticsearch::Client.new(
  host: host,
  transport_options: transport_options,
  api_key: {id: 'my_id', api_key: 'my_api_key'}
)

Logging

To log requests and responses to standard output with the default logger (an instance of Ruby's Logger class), set the log argument to true:

Elasticsearch::Client.new(log: true)

You can also use ecs-logging. ecs-logging is a set of libraries that allows you to transform your application logs to structured logs that comply with the Elastic Common Schema (ECS):

logger = EcsLogging::Logger.new($stdout)
Elasticsearch::Client.new(logger: logger)

To trace requests and responses in the Curl format, set the trace argument:

Elasticsearch::Client.new(trace: true)

You can customize the default logger or tracer:

    client.transport.logger.formatter = proc { |s, d, p, m| "#{s}: #{m}\n" }
    client.transport.logger.level = Logger::INFO

Or, you can use a custom ::Logger instance:

Elasticsearch::Client.new(logger: Logger.new(STDERR))

You can pass the client any conforming logger implementation:

require 'logging' # https://github.com/TwP/logging/

log = Logging.logger['elasticsearch']
log.add_appenders Logging.appenders.stdout
log.level = :info

client = Elasticsearch::Client.new(logger: log)

Custom HTTP Headers

You can set a custom HTTP header on the client's initializer:

client = Elasticsearch::Client.new(
  transport_options: {
    headers:
      {user_agent: "My App"}
  }
)

You can also pass in headers as a parameter to any of the API Endpoints to set custom headers for the request:

client.search(index: 'myindex', q: 'title:test', headers: {user_agent: "My App"})

Identifying running tasks with X-Opaque-Id

The X-Opaque-Id header allows to track certain calls, or associate certain tasks with the client that started them (more on the Elasticsearch docs). To use this feature, you need to set an id for opaque_id on the client on each request. Example:

client = Elasticsearch::Client.new
client.search(index: 'myindex', q: 'title:test', opaque_id: '123456')

The search request will include the following HTTP Header:

X-Opaque-Id: 123456

You can also set a prefix for X-Opaque-Id when initializing the client. This will be prepended to the id you set before each request if you're using X-Opaque-Id. Example:

client = Elasticsearch::Client.new(opaque_id_prefix: 'eu-west1')
client.search(index: 'myindex', q: 'title:test', opaque_id: '123456')

The request will include the following HTTP Header:

X-Opaque-Id: eu-west1_123456

Setting Timeouts

For many operations in Elasticsearch, the default timeouts of HTTP libraries are too low. To increase the timeout, you can use the request_timeout parameter:

Elasticsearch::Client.new request_timeout: 5*60

You can also use the transport_options argument documented below.

Randomizing Hosts

If you pass multiple hosts to the client, it rotates across them in a round-robin fashion, by default. When the same client would be running in multiple processes (eg. in a Ruby web server such as Thin), it might keep connecting to the same nodes "at once". To prevent this, you can randomize the hosts collection on initialization and reloading:

Elasticsearch::Client.new hosts: ['localhost:9200', 'localhost:9201'], randomize_hosts: true

Retrying on Failures

When the client is initialized with multiple hosts, it makes sense to retry a failed request on a different host:

Elasticsearch::Client.new hosts: ['localhost:9200', 'localhost:9201'], retry_on_failure: true

By default, the client will retry the request 3 times. You can specify how many times to retry before it raises an exception by passing a number to retry_on_failure:

Elasticsearch::Client.new hosts: ['localhost:9200', 'localhost:9201'], retry_on_failure: 5

These two parameters can also be used together:

Elasticsearch::Client.new hosts: ['localhost:9200', 'localhost:9201'], retry_on_status: [502, 503], retry_on_failure: 10

Reloading Hosts

Elasticsearch by default dynamically discovers new nodes in the cluster. You can leverage this in the client, and periodically check for new nodes to spread the load.

To retrieve and use the information from the Nodes Info API on every 10,000th request:

Elasticsearch::Client.new hosts: ['localhost:9200', 'localhost:9201'], reload_connections: true

You can pass a specific number of requests after which the reloading should be performed:

Elasticsearch::Client.new hosts: ['localhost:9200', 'localhost:9201'], reload_connections: 1_000

To reload connections on failures, use:

Elasticsearch::Client.new hosts: ['localhost:9200', 'localhost:9201'], reload_on_failure: true

The reloading will timeout if not finished under 1 second by default. To change the setting:

Elasticsearch::Client.new hosts: ['localhost:9200', 'localhost:9201'], sniffer_timeout: 3

NOTE: When using reloading hosts ("sniffing") together with authentication, just pass the scheme, user and password with the host info -- or, for more clarity, in the http options:

Elasticsearch::Client.new host: 'localhost:9200',
                          http: { scheme: 'https', user: 'U', password: 'P' },
                          reload_connections: true,
                          reload_on_failure: true

Connection Selector

By default, the client will rotate the connections in a round-robin fashion, using the Elasticsearch::Transport::Transport::Connections::Selector::RoundRobin strategy.

You can implement your own strategy to customize the behaviour. For example, let's have a "rack aware" strategy, which will prefer the nodes with a specific attribute. Only when these would be unavailable, the strategy will use the other nodes:

class RackIdSelector
  include Elasticsearch::Transport::Transport::Connections::Selector::Base

  def select(options={})
    connections.select do |c|
      # Try selecting the nodes with a `rack_id:x1` attribute first
      c.host[:attributes] && c.host[:attributes][:rack_id] == 'x1'
    end.sample || connections.to_a.sample
  end
end

Elasticsearch::Client.new hosts: ['x1.search.org', 'x2.search.org'], selector_class: RackIdSelector

Transport Implementations

By default, the client will use the Faraday HTTP library as a transport implementation.

It will auto-detect and use an adapter for Faraday based on gems loaded in your code, preferring HTTP clients with support for persistent connections.

To use the Patron HTTP, for example, just require it:

require 'patron'

Then, create a new client, and the Patron gem will be used as the "driver":

client = Elasticsearch::Client.new

client.transport.connections.first.connection.builder.adapter
# => Faraday::Adapter::Patron

10.times do
  client.nodes.stats(metric: 'http')['nodes'].values.each do |n|
    puts "#{n['name']} : #{n['http']['total_opened']}"
  end
end

# => Stiletoo : 24
# => Stiletoo : 24
# => Stiletoo : 24
# => ...

To use a specific adapter for Faraday, pass it as the adapter argument:

client = Elasticsearch::Client.new adapter: :net_http_persistent

client.transport.connections.first.connection.builder.handlers
# => [Faraday::Adapter::NetHttpPersistent]

To pass options to the Faraday::Connection constructor, use the transport_options key:

client = Elasticsearch::Client.new transport_options: {
  request: { open_timeout: 1 },
  headers: { user_agent:   'MyApp' },
  params:  { :format => 'yaml' },
  ssl:     { verify: false }
}

To configure the Faraday instance directly, use a block:

require 'patron'

client = Elasticsearch::Client.new(host: 'localhost', port: '9200') do |f|
  f.response :logger
  f.adapter  :patron
end

You can use any standard Faraday middleware and plugins in the configuration block. You can also initialize the transport class yourself, and pass it to the client constructor as the transport argument:

require 'patron'

transport_configuration = lambda do |f|
  f.response :logger
  f.adapter  :patron
end

transport = Elasticsearch::Transport::Transport::HTTP::Faraday.new \
  hosts: [ { host: 'localhost', port: '9200' } ],
  &transport_configuration

# Pass the transport to the client
#
client = Elasticsearch::Client.new transport: transport

Instead of passing the transport to the constructor, you can inject it at run time:

# Set up the transport
#
faraday_configuration = lambda do |f|
  f.instance_variable_set :@ssl, { verify: false }
  f.adapter :excon
end

faraday_client = Elasticsearch::Transport::Transport::HTTP::Faraday.new \
  hosts: [ { host: 'my-protected-host',
             port: '443',
             user: 'USERNAME',
             password: 'PASSWORD',
             scheme: 'https'
          }],
  &faraday_configuration

# Create a default client
#
client = Elasticsearch::Client.new

# Inject the transport to the client
#
client.transport = faraday_client

You can also use a bundled Curb based transport implementation:

require 'curb'
require 'elasticsearch/transport/transport/http/curb'

client = Elasticsearch::Client.new transport_class: Elasticsearch::Transport::Transport::HTTP::Curb

client.transport.connections.first.connection
# => #<Curl::Easy http://localhost:9200/>

It's possible to customize the Curb instance by passing a block to the constructor as well (in this case, as an inline block):

transport = Elasticsearch::Transport::Transport::HTTP::Curb.new \
  hosts: [ { host: 'localhost', port: '9200' } ],
  & lambda { |c| c.verbose = true }

client = Elasticsearch::Client.new transport: transport

You can write your own transport implementation easily, by including the Elasticsearch::Transport::Transport::Base module, implementing the required contract, and passing it to the client as the transport_class parameter -- or injecting it directly.

Serializer Implementations

By default, the MultiJSON library is used as the serializer implementation, and it will pick up the "right" adapter based on gems available.

The serialization component is pluggable, though, so you can write your own by including the Elasticsearch::Transport::Transport::Serializer::Base module, implementing the required contract, and passing it to the client as the serializer_class or serializer parameter.

Exception Handling

The library defines a number of exception classes for various client and server errors, as well as unsuccessful HTTP responses, making it possible to rescue specific exceptions with desired granularity.

The highest-level exception is Elasticsearch::Transport::Transport::Error and will be raised for any generic client or server errors.

Elasticsearch::Transport::Transport::ServerError will be raised for server errors only.

As an example for response-specific errors, a 404 response status will raise an Elasticsearch::Transport::Transport::Errors::NotFound exception.

Finally, Elasticsearch::Transport::Transport::SnifferTimeoutError will be raised when connection reloading ("sniffing") times out.

Development and Community

For local development, clone the repository and run bundle install. See rake -T for a list of available Rake tasks for running tests, generating documentation, starting a testing cluster, etc.

Bug fixes and features must be covered by unit tests. Integration tests are written in Ruby 1.9 syntax.

Github's pull requests and issues are used to communicate, send bug reports and code contributions.

The Architecture

Development

To work on the code, clone and bootstrap the main repository first -- please see instructions in the main README.

To run tests, launch a testing cluster and use the Rake tasks:

time rake test:unit
time rake test:integration

Use COVERAGE=true before running a test task to check coverage with Simplecov.

License

This software is licensed under the Apache 2 license.