Please see the 4.0.0.pre.1 README for instructions on use.

An EventMachine based Doozer client

(For pre-releases, use --pre)

$ gem install fraggle

Connecting to a cluster

Use Fraggle.connect. It takes an optional doozer uri (String). If no parameters are given, it will use the DOOZER_URI` environment variable if present, otherwise it will default to the uri containing the default doozer addresses with IP 127.0.0.1 and ports 8046, 8041, 8042, 8043.

simple example

require 'rubygems'
require 'eventmachine'
require 'fraggle'

EM.run do
  # In the event of a lost connection, fraggle will attempt
  # other doozers until one accepts or it runs out of options; A NoAddrs
  # exception will be raised if that later happens.

  Fraggle.connect do |c, err|
    if err
      raise err.message
    end

    c.rev do |v|
      c.get(v, "/foo") do |e, err|
        if err
          err.code   # => nil
          err.detail # => nil
        else
          e.value    # => nil
          e.rev      # => 0
          e.missing? # => true
        end

        p [:get, e, err]
      end

      ## Obtain the current revision the store is at and watch from then on for
      ## any SET or DEL to /foo.
      c.wait(v, "/foo") do |e, err|
        # The event has:
        # ------------------------
        if err
          err.code   # => nil
          err.detail # => nil
        else
          e.path       # => "/foo"
          e.value      # => "zomg!"
          e.rev        # => 123
          e.set?       # => true
          e.del?       # => false
        end

        p [:wait, e, err]
      end
    end

    ## Setting a key (this will trigger the watch above)
    f = Proc.new do |e, err|
      p [:e, e, err]

      # This case statement is undesirable.  We're working
      # on better error handling.
      case (err && err.code rescue err)
      when Fraggle::Connection::DisconnectedError
        # Fraggle (for now) does not attempt a non-idempotent request.  This means
        # Fraggle will hand off the error to the user if there is a SET or DEL
        # with rev 0 (missing) and delete it during the time we may be
        # disconnected.
        #
        # In this scenario, there are no other clients that can exist that will
        # attempt to set this "lock" if it's missing then delete it.  It is safe
        # for us to resend the request if we were disconnected from the previous
        # server before a response.
        #
        # See High-Availability in the README for more information about this.
        #
        c.set(0, "/foo", "zomg!", &f)
      when Fraggle::REV_MISMATCH
        p :not_it
      when nil
          # Success!
        p [:it, e]
      else
        fail "something bad happened: "   err.inspect
      end
    end

    c.set(0, "/foo", "zomg!", &f)
  end
end

Fraggle read commands take a rev. If no rev is given, Doozer will reply with the most up-to-date data. If you need to do multiple reads at certain point in time for consistency, use the rev command.

c.rev do |v|
  c.get(v, "/a") { ... }
  c.get(v, "/b") { ... }
  c.get(v, "/c") { ... }
end

This also means you can go back in time or into the future!

# This will not yield until the data store is at revision 100,000
c.get(100_000, "/a") { ... }

NOTE: Doozer's data store is a persistent data structure. You can reference the stores history as far back as it is configured to hold it. The default is 360,000 revisions. See data model for more information.

Fraggle has mechanisms to gracefully deal with connection loss. They are:

Resend / Connection loss

When a connection is lost and Fraggle successfully reconnects to another Doozer node, Fraggle will resend most pending requests to the new connection. This means you will not miss events; Even events that happened while you were disconnected! All read commands will pick up where they left off. This is valuable to understand because it means you don't need to code for failure on reads; Fraggle gracefully handles it for you.

Write commands will be resent if their rev is greater than 0. These are idempotent requests. A rev of 0 will cause that request's error callback to be invoked with a Fraggle::Connection::Disconnected response. You will have to handle these yourself because Fraggle cannot know whether or not it's safe to retry on your behalf.

attempt

Before fraggle will attempt a new address after connection loss, it calls the block given to Fraggle::Client#attempt. If the block returns false, Fraggle will not attempt that address or anymore. The block is called with on parameter addr, which is the address being attempted.

Example:

c = Fraggle.connect

c.attempt do |addr|
  addr =~ /^127\.*$/ # don't connect to localhost doozers
end

The default attempt is Proc.new {|_| true }

Each command below behaves according to the proto spec, respectively. Their blks are called with two parameters, a Fraggle::Response as the first or a Fraggle::Connection::ResponseError as the second if a response is returned from the server.

set(rev, path, value, &blk)

del(rev, path, &blk)

get(rev, path, &blk)

wait(rev, path, &blk)

rev(&blk)

stat(rev, path, &blk)

watch(rev, path, &blk)

Watches path (a glob pattern) for changes, from rev in history on. Its blk is called with a Fraggle::Response for each event.

getdir(rev, path, off=0, lim=MaxInt64, ents=[], &blk)

Behaves like getdir but collects ents, starting at off until all or lim entries are read. When done, blk is called with the result (an Array) as the first parameter or a Fraggle::Connection::Response as the second. Depending on the response, one or the other will be set and the other with be nil.

walk(rev, path, off=0, lim=MaxInt64, ents=[], &blk)

Like getdir, but but path is a glob pattern and each result contains a path, value, and rev.

Clone

$ git clone http://github.com/ha/fraggle.git

Test

$ gem install turn

$ turn

Mailing List

Please join the Doozer mailing list for help: http://groups.google.com/forum/#!forum/doozer