class Rightscale::HttpConnection

HttpConnection maintains a persistent HTTP connection to a remote server. Each instance maintains its own unique connection to the HTTP server. HttpConnection makes a best effort to receive a proper HTTP response from the server, although it does not guarantee that this response contains a HTTP Success code.

On low-level errors (TCP/IP errors) HttpConnection invokes a reconnect and retry algorithm. Note that although each HttpConnection object has its own connection to the HTTP server, error handling is shared across all connections to a server. For example, if there are three connections to www.somehttpserver.com, a timeout error on one of those connections will cause all three connections to break and reconnect. A connection will not break and reconnect, however, unless a request becomes active on it within a certain amount of time after the error (as specified by HTTP_CONNECTION_RETRY_DELAY). An idle connection will not break even if other connections to the same server experience errors.

A HttpConnection will retry a request a certain number of times (as defined by HTTP_CONNNECTION_RETRY_COUNT). If all the retries fail, an exception is thrown and all HttpConnections associated with a server enter a probationary period defined by HTTP_CONNECTION_RETRY_DELAY. If the user makes a new request subsequent to entering probation, the request will fail immediately with the same exception thrown on probation entry. This is so that if the HTTP server has gone down, not every subsequent request must wait for a connect timeout before failing. After the probation period expires, the internal state of the HttpConnection is reset and subsequent requests have the full number of potential reconnects and retries available to them.

Constants

HTTP_CONNECTION_OPEN_TIMEOUT

Throw a Timeout::Error if a connection isn’t established within this number of seconds

HTTP_CONNECTION_READ_TIMEOUT

Throw a Timeout::Error if no data have been read on this connnection within this number of seconds

HTTP_CONNECTION_RETRY_COUNT

Number of times to retry the request after encountering the first error

HTTP_CONNECTION_RETRY_DELAY

Length of the post-error probationary period during which all requests will fail

Attributes

http[RW]
logger[RW]
params[RW]
server[RW]

Public Class Methods

blank?(obj) click to toggle source
# File lib/right_http_connection.rb, line 422
def self.blank?(obj)
  case obj
  when NilClass, FalseClass
    true
  when TrueClass, Numeric
    false
  when Array, Hash
    obj.empty?
  when String
    obj.empty? || obj.strip.empty?
  else
    # "", "   ", nil, [], and {} are blank
    if obj.respond_to?(:empty?) && obj.respond_to?(:strip)
      obj.empty? or obj.strip.empty?
    elsif obj.respond_to?(:empty?)
      obj.empty?
    else
      !obj
    end
  end
end
new(params={}) click to toggle source

Params hash:

:user_agent => 'www.HostName.com'    # String to report as HTTP User agent
:ca_file    => 'path_to_file'        # A path of a CA certification file in PEM format. The file can contain several CA certificates.
:logger     => Logger object         # If omitted, HttpConnection logs to STDOUT
:exception  => Exception to raise    # The type of exception to raise if a request repeatedly fails. RuntimeError is raised if this parameter is omitted.
:http_connection_retry_count         # by default == Rightscale::HttpConnection.params[:http_connection_retry_count]
:http_connection_open_timeout        # by default == Rightscale::HttpConnection.params[:http_connection_open_timeout]
:http_connection_read_timeout        # by default == Rightscale::HttpConnection.params[:http_connection_read_timeout]
:http_connection_retry_delay         # by default == Rightscale::HttpConnection.params[:http_connection_retry_delay]
# File lib/right_http_connection.rb, line 128
def initialize(params={})
  @params                                = params
  @params[:http_connection_retry_count]  ||= @@params[:http_connection_retry_count]
  @params[:http_connection_open_timeout] ||= @@params[:http_connection_open_timeout]
  @params[:http_connection_read_timeout] ||= @@params[:http_connection_read_timeout]
  @params[:http_connection_retry_delay]  ||= @@params[:http_connection_retry_delay]
  @http                                  = nil
  @server                                = nil
  @logger                                = get_param(:logger) ||
      (RAILS_DEFAULT_LOGGER if defined?(RAILS_DEFAULT_LOGGER))
  unless @logger
    @logger = Logger.new(STDOUT)
    @logger.level = Logger::INFO
  end
end
params() click to toggle source

Query the global (class-level) parameters:

:user_agent => 'www.HostName.com'    # String to report as HTTP User agent
:ca_file    => 'path_to_file'        # Path to a CA certification file in PEM format. The file can contain several CA certificates.  If this parameter isn't set, HTTPS certs won't be verified.
:logger     => Logger object         # If omitted, HttpConnection logs to STDOUT
:exception  => Exception to raise    # The type of exception to raise
                                     # if a request repeatedly fails. RuntimeError is raised if this parameter is omitted.
:http_connection_retry_count         # by default == Rightscale::HttpConnection::HTTP_CONNECTION_RETRY_COUNT
:http_connection_open_timeout        # by default == Rightscale::HttpConnection::HTTP_CONNECTION_OPEN_TIMEOUT
:http_connection_read_timeout        # by default == Rightscale::HttpConnection::HTTP_CONNECTION_READ_TIMEOUT
:http_connection_retry_delay         # by default == Rightscale::HttpConnection::HTTP_CONNECTION_RETRY_DELAY
# File lib/right_http_connection.rb, line 101
def self.params
  @@params
end
params=(params) click to toggle source

Set the global (class-level) parameters

# File lib/right_http_connection.rb, line 106
def self.params=(params)
  @@params = params
end

Public Instance Methods

close(reason='') click to toggle source
# File lib/right_http_connection.rb, line 418
def close(reason='')
  finish
end
finish(reason = '') click to toggle source
# File lib/right_http_connection.rb, line 410
def finish(reason = '')
  if @http && @http.started?
    reason = ", reason: '#{reason}'" unless self.class.blank?(reason)
    @logger.debug("Closing #{@http.use_ssl? ? 'HTTPS' : 'HTTP'} connection to #{@http.address}:#{@http.port}#{reason}")
    @http.finish
  end
end
get_param(name) click to toggle source
# File lib/right_http_connection.rb, line 144
def get_param(name)
  @params[name] || @@params[name]
end
local_read_size=(newsize) click to toggle source

Set the maximum size (in bytes) of a single read from local data sources like files. This can be used to tune the performance of, for example, a streaming PUT of a large buffer.

# File lib/right_http_connection.rb, line 172
def local_read_size=(newsize)
  Net::HTTPGenericRequest.local_read_size=(newsize)
end
local_read_size?() click to toggle source

Query for the maximum size (in bytes) of a single read from local data sources like files. This is important, for example, in a streaming PUT of a large buffer.

# File lib/right_http_connection.rb, line 165
def local_read_size?
  Net::HTTPGenericRequest.local_read_size?
end
request(request_params, &block) click to toggle source

Send HTTP request to server

request_params hash:
:server   => 'www.HostName.com'   # Hostname or IP address of HTTP server
:port     => '80'                 # Port of HTTP server
:protocol => 'https'              # http and https are supported on any port
:request  => 'requeststring'      # Fully-formed HTTP request to make

Raises RuntimeError, Interrupt, and params (if specified in new).

# File lib/right_http_connection.rb, line 322
def request(request_params, &block)
  # We save the offset here so that if we need to retry, we can return the file pointer to its initial position
  mypos = get_fileptr_offset(request_params)
  loop do
    # if we are inside a delay between retries: no requests this time!
    if error_count > @params[:http_connection_retry_count] &&
        error_time + @params[:http_connection_retry_delay] > Time.now
      # store the message (otherwise it will be lost after error_reset and
      # we will raise an exception with an empty text)
      banana_message_text = banana_message
      @logger.warn("#{err_header} re-raising same error: #{banana_message_text} " +
                       "-- error count: #{error_count}, error age: #{Time.now.to_i - error_time.to_i}")
      exception = get_param(:exception) || RuntimeError
      raise exception.new(banana_message_text)
    end

    # try to connect server(if connection does not exist) and get response data
    begin
      request_params[:protocol] ||= (request_params[:port] == 443 ? 'https' : 'http')

      request                   = request_params[:request]
      request['User-Agent']     = get_param(:user_agent) || ''

      # (re)open connection to server if none exists or params has changed
      unless @http &&
          @http.started? &&
          @server == request_params[:server] &&
          @port == request_params[:port] &&
          @protocol == request_params[:protocol]
        start(request_params)
      end

      # Detect if the body is a streamable object like a file or socket.  If so, stream that
      # bad boy.
      setup_streaming(request)
      response = @http.request(request, &block)

      error_reset
      eof_reset
      return response

      # We treat EOF errors and the timeout/network errors differently.  Both
      # are tracked in different statistics blocks.  Note below that EOF
      # errors will sleep for a certain (exponentially increasing) period.
      # Other errors don't sleep because there is already an inherent delay
      # in them; connect and read timeouts (for example) have already
      # 'slept'.  It is still not clear which way we should treat errors
      # like RST and resolution failures.  For now, there is no additional
      # delay for these errors although this may change in the future.

      # EOFError means the server closed the connection on us.
    rescue EOFError => e
      @logger.debug("#{err_header} server #{@server} closed connection")
      @http = nil

      # if we have waited long enough - raise an exception...
      if raise_on_eof_exception?
        exception = get_param(:exception) || RuntimeError
        @logger.warn("#{err_header} raising #{exception} due to permanent EOF being received from #{@server}, error age: #{Time.now.to_i - eof_time.to_i}")
        raise exception.new("Permanent EOF is being received from #{@server}.")
      else
        # ... else just sleep a bit before new retry
        sleep(add_eof)
        # We will be retrying the request, so reset the file pointer
        reset_fileptr_offset(request, mypos)
      end
    rescue Exception => e # See comment at bottom for the list of errors seen...
      @http = nil
      # if ctrl+c is pressed - we have to reraise exception to terminate proggy
      if e.is_a?(Interrupt) && !(e.is_a?(Errno::ETIMEDOUT) || e.is_a?(Timeout::Error))
        @logger.debug("#{err_header} request to server #{@server} interrupted by ctrl-c")
        raise
      elsif e.is_a?(ArgumentError) && e.message.include?('wrong number of arguments (5 for 4)')
        # seems our net_fix patch was overriden...
        exception = get_param(:exception) || RuntimeError
        raise exception.new('incompatible Net::HTTP monkey-patch')
      end
      # oops - we got a banana: log it
      error_add(e.message)
      @logger.warn("#{err_header} request failure count: #{error_count}, exception: #{e.inspect}")

      # We will be retrying the request, so reset the file pointer
      reset_fileptr_offset(request, mypos)

    end
  end
end
socket_read_size=(newsize) click to toggle source

Set the maximum size (in bytes) of a single read from the underlying socket. For bulk transfer, especially over fast links, this is value is critical to performance.

# File lib/right_http_connection.rb, line 158
def socket_read_size=(newsize)
  Net::BufferedIO.socket_read_size=(newsize)
end
socket_read_size?() click to toggle source

Query for the maximum size (in bytes) of a single read from the underlying socket. For bulk transfer, especially over fast links, this is value is critical to performance.

# File lib/right_http_connection.rb, line 151
def socket_read_size?
  Net::BufferedIO.socket_read_size?
end

Private Instance Methods

add_eof() click to toggle source

Adds new EOF timestamp. Returns the number of seconds to wait before new conection retry:

0.5, 1, 2, 4, 8
# File lib/right_http_connection.rb, line 220
def add_eof
  (@@eof[@server] ||= []).unshift Time.now
  0.25 * 2 ** @@eof[@server].size
end
banana_message() click to toggle source

Error message stuff…

# File lib/right_http_connection.rb, line 209
def banana_message
  return "#{@server} temporarily unavailable: (#{error_message})"
end
eof_reset() click to toggle source

Reset a list of EOFs for this server. This is being called when we have got an successful response from server.

# File lib/right_http_connection.rb, line 238
def eof_reset
  @@eof.delete(@server)
end
eof_time() click to toggle source

Returns first EOF timestamp or nul if have no EOFs being tracked.

# File lib/right_http_connection.rb, line 226
def eof_time
  @@eof[@server] && @@eof[@server].last
end
err_header() click to toggle source
# File lib/right_http_connection.rb, line 213
def err_header
  return "#{self.class.name} :"
end
error_add(message) click to toggle source

add an error for a server

# File lib/right_http_connection.rb, line 199
def error_add(message)
  @@state[@server] = {:count => error_count+1, :time => Time.now, :message => message}
end
error_count() click to toggle source

number of consecutive errors seen for server, 0 all is ok

# File lib/right_http_connection.rb, line 184
def error_count
  @@state[@server] ? @@state[@server][:count] : 0
end
error_message() click to toggle source

message for last error for server, “” if all is ok

# File lib/right_http_connection.rb, line 194
def error_message
  @@state[@server] ? @@state[@server][:message] : ""
end
error_reset() click to toggle source

reset the error state for a server (i.e. a request succeeded)

# File lib/right_http_connection.rb, line 204
def error_reset
  @@state.delete(@server)
end
error_time() click to toggle source

time of last error for server, nil if all is ok

# File lib/right_http_connection.rb, line 189
def error_time
  @@state[@server] && @@state[@server][:time]
end
get_fileptr_offset(request_params) click to toggle source
# File lib/right_http_connection.rb, line 252
def get_fileptr_offset(request_params)
  request_params[:request].body.pos
rescue Exception => e
  # Probably caught this because the body doesn't support the pos() method, like if it is a socket.
  # Just return 0 and get on with life.
  0
end
raise_on_eof_exception?() click to toggle source

Returns true if we are receiving EOFs during last @params seconds and there were no successful response from server

# File lib/right_http_connection.rb, line 232
def raise_on_eof_exception?
  self.class.blank?(@@eof[@server]) ? false : ((Time.now.to_i-@params[:http_connection_retry_delay]) > @@eof[@server].last.to_i)
end
reset_fileptr_offset(request, offset = 0) click to toggle source
# File lib/right_http_connection.rb, line 260
def reset_fileptr_offset(request, offset = 0)
  if (request.body_stream && request.body_stream.respond_to?(:pos))
    begin
      request.body_stream.pos = offset
    rescue Exception => e
      @logger.warn("Failed file pointer reset; aborting HTTP retries." +
                       " -- #{err_header} #{e.inspect}")
      raise e
    end
  end
end
setup_streaming(request) click to toggle source

Detects if an object is ‘streamable’ - can we read from it, and can we know the size?

# File lib/right_http_connection.rb, line 243
def setup_streaming(request)
  if (request.body && request.body.respond_to?(:read))
    body                   = request.body
    request.content_length = body.respond_to?(:lstat) ? body.lstat.size : body.size
    request.body_stream    = request.body
    true
  end
end
start(request_params) click to toggle source

Start a fresh connection. The object closes any existing connection and opens a new one.

# File lib/right_http_connection.rb, line 274
def start(request_params)
  # close the previous if exists
  finish
  # create new connection
  @server   = request_params[:server]
  @port     = request_params[:port]
  @protocol = request_params[:protocol]

  @logger.debug("Opening new #{@protocol.upcase} connection to #@server:#@port")
  @http              = Net::HTTP.new(@server, @port)
  @http.open_timeout = @params[:http_connection_open_timeout]
  @http.read_timeout = @params[:http_connection_read_timeout]

  if @protocol == 'https'
    verifyCallbackProc = Proc.new { |ok, x509_store_ctx|
      code = x509_store_ctx.error
      msg  = x509_store_ctx.error_string
      #debugger
      @logger.warn("##### #{@server} certificate verify failed: #{msg}") unless code == 0
      true
    }
    @http.use_ssl      = true
    @http.verify_mode  = OpenSSL::SSL::VERIFY_NONE # Looks like Ruby 1.9 defaults to VERIFY_PEER which doesn't work well
    ca_file            = get_param(:ca_file)
    if ca_file
      @http.verify_mode     = OpenSSL::SSL::VERIFY_PEER
      @http.verify_callback = verifyCallbackProc
      @http.ca_file         = ca_file
    end
  end
  # open connection
  @http.start
end