ConCoord is a novel coordination service that provides replication and synchronization support for large-scale distributed systems. ConCoord employs an object-oriented approach, in which the system actively creates and maintains live replicas for user-provided objects. Through ConCoord, the clients are able to access these replicated objects transparently as if they are local objects. ConCoord uses these replicated objects to implement coordination and synchronization constructs in large-scale distributed systems, in effect establishing a transparent way of providing a coordination service.

Authors

• Deniz Altinbuken (deniz@systems.cs.cornell.edu)
• Emin Gun Sirer (egs@systems.cs.cornell.edu)

Version

1.0.0

Date

2013-04-13

The minimum requirements for ConCoord are:

- python 2.7.2 or later
- dnspython-1.9.4
- msgpack-python

To build and install ConCoord, type:

sudo python setup.py install

To use ConCoord and distribute your Python objects first you need to create the local Python object. Once the object is created, ConCoord will automatically create a proxy object and you will be able to do remote method invocations when your object is distributed. For these remote method invocations, a round of Paxos Consensus Protocol is run by ConCoord to provide consistency between different copies of the object.

To walk you through the ConCoord approach, we will use the example Counter coordination object we provided. In the source distribution you can locate the Counter object under concoord/object/counter.py. Once you install ConCoord, you can create coordination objects and save them anywhere in your filesystem. To demonstrate this, we will save counter.py under /foo/counter.py:

class Counter:
  def __init__(self, value=0):
   self.value = value

  def decrement(self):
    self.value -= 1

  def increment(self):
    self.value += 1

  def getvalue(self):
    return self.value

  def __str__(self):
    return "The counter value is %d" % self.value

Once you have created the object, update your PYTHONPATH accordingly, so that the objects can be found and imported:

$ export PYTHONPATH=$PYTHONPATH:/foo/

Clients will use a proxy object to do method calls on the ConCoord object. To create a proxy for the ConCoord object, run the following command:

$ concoord object -o counter.Counter

Usage: concoord object [-h] [-o OBJECTNAME] [-t SECURITYTOKEN] [-p PROXYTYPE] [-s] [-v]

where,

-h, --help show this help message and exit -o OBJECTNAME, --objectname OBJECTNAME client object dotted name module.Class -t SECURITYTOKEN, --token SECURITYTOKEN security token -p PROXYTYPE, --proxytype PROXYTYPE 0:BASIC, 1:BLOCKING, 2:CLIENT-SIDE BATCHING, 3:SERVER-SIDE BATCHING -s, --safe safety checking on/off -v, --verbose verbose mode on/off

This script will create a proxy file under the directory that the object resides (i.e. /foo/):

/foo/counterproxy.py := the proxy that can be used by the client

IMPORTANT NOTE: ConCoord objects treat __init__ functions specially in two ways:

1) When Replicas go live, the object is instantiated calling the

__init__ without any arguments. Therefore, while implementing coordination objects, the __init__ method should be implemented to be able to run without explicit arguments. You can use defaults to implement an __init__ method that accepts arguments.

2) In the proxy created, the __init__ function is used to initialize

the Client-Replica connection. This way, multiple clients can connect to a ConCoord instance without reinitializing the object. During proxy generation, the original __init__ function is renamed as __concoordinit__, to reinitialize the object the user can call __concoordinit__ at any point.

To distribute your object you should start at least one replica and one acceptor.

To start the replica node, use the following command:

$ concoord replica -o counter.Counter

To start replica nodes to join an active ConCoord instance, use the following command to connect to a bootstrap replica with ipaddr:port:

$ concoord replica -o counter.Counter -b ipaddr:port

To start an acceptor node that connects to the bootrstrap replica at ipaddr:port, use the following command:

$ concoord acceptor -b ipaddr:port

Note that you can specify the port and the address of any node with options -p and -a respectively. The nodes can also be run in the debug and interactive modes or with a logger with the commands shown below:

Usage: concoord [-h] [-a ADDR] [-p PORT] [-b BOOTSTRAP] [-o OBJECTNAME]

[-l LOGGER] [-d] [-i]

where,

-h, --help show this help message and exit -a ADDR, --addr ADDR addr for the node -p PORT, --port PORT port for the node -b BOOTSTRAP, --boot BOOTSTRAP address:port:type triple for the bootstrap peer -o OBJECTNAME, --objectname OBJECTNAME client object dotted name -l LOGGER, --logger LOGGER logger address -d, --debug debug on/off -i, --interactive interactive shell on/off

You can dynamically locate nodes in a given ConCoord instance using DNS queries if the instance includes nameserver nodes. There are three ways you can run a ConCoord Nameserver.

• Standalone Nameserver Keeps track of the view and responds to DNS queries itself. Requires su privileges to bind to Port 53.
• Slave Nameserver Keeps track of the view and updates a master nameserver that answers to DNS queries on behalf of the slave nameserver. Requires an active master nameserver.
• Route53 Nameserver Keeps track of the view and updates an Amazon Route53 account. Amazon Route53 answers to DNS queries on behalf of the slave nameserver. Requires a ready-to-use Amazon Route53 account.

Before starting a standalone nameserver node manually, first make sure that you have at least one replica and one acceptor running. Once your replica and acceptor nodes are set up, you can start the nameserver to answer queries for counterdomain as follows:

$ sudo concoord nameserver -n counterdomain -o counter.Counter -b ipaddr:port -t 1

When you set up the nameserver delegations, you can send queries for counterdomain and see the most current set of nodes as follows:

$ dig -t a counterdomain                   # returns set of Replicas

$ dig -t srv _concoord._tcp.counterdomain  # returns set of Replicas with ports

$ dig -t txt counterdomain                 # returns set of all nodes

$ dig -t ns counterdomain                  # returns set of nameservers

If you want to run the nameserver without proper delegation setup, you can query the nameserver bound to nsipaddr specifically as follows:

$ dig -t a counterdomain @nsipaddr         # returns set of Replicas

Before starting a slave nameserver node manually, you should have a master nameserver set up and running. The master nameserver should be set up to answer the queries for its slave nameservers. We provide OpenReplica Nameserver as a ready to deploy master nameserver and a Nameserver Coordination Object in our example objects set to keep track of slave nameserver information. Using this coordination object, the master nameserver can keep track of its slave nameserver delegations and the slave nameserver can update the master every time the view of its system changes.

Once your master nameserver is set up, you can start the slave nameserver as follows:

$ concoord nameserver -n counterdomain -o counter.Counter -b ipaddr:port -t 2 -m masterdomain

When the slave nameserver starts running, you can send queries for counterdomain and see the most current set of nodes as follows:

$ dig -t a counterdomain                   # returns set of Replicas

$ dig -t srv _concoord._tcp.counterdomain  # returns set of Replicas with ports

$ dig -t txt counterdomain                 # returns set of all nodes

$ dig -t ns counterdomain                  # returns set of nameservers

Before starting a nameserver connected to Amazon Route 53, you should have a Route 53 account set up and ready to receive requests. After your Route 53 account is ready, the nameserver can update the master every time the view of its system changes automatically.

To use Amazon Route 53 you can pass your credentials into the methods that create connections or edit them in the configuration file.

• AWS_ACCESS_KEY_ID - Your AWS Access Key ID
• AWS_SECRET_ACCESS_KEY - Your AWS Secret Access Key

Once you make sure that your Route53 account is set up and your credentials are updated, you can start the nameserver as follows:

$ concoord nameserver -n counterdomain -o counter.Counter -b ipaddr:port -t 3 -o configfilepath

Once you have a ConCoord instance running with your object, it is easy to access your object.

Now we will use the proxy object we generated at (4.1). You can import and use this proxy object in your code. Depending on how you set your nameserver node up, you can access your object with the ipaddr:port pair or the domainname.

>>> from counterproxy import Counter >>> c = Counter(domainname) >>> c.increment() >>> c.increment() >>> c.getvalue() 2

At any point to reinitialize an object after it is deployed on replicas, you should call __concoordinit__ function:

>>> from counterproxy import Counter >>> c = Counter(domainname) >>> c.increment() >>> c.__concoordinit__() >>> c.increment() >>> c.getvalue() 1

You can create bundles to use at the server and client sides using the Makefile provided under concoord/

Remember to add the objects you have created in these bundles.

To create a bundle that can run replica, acceptor and nameserver nodes:

$ make server

To create a bundle that can run a client and connect to an existing ConCoord instance:

$ make client

We have two kinds of loggers for ConCoord:: * Console Logger * Network Logger

Both of these loggers are included under utils.py. To start the NetworkLogger, use the logdaemon.py on the host you want to keep the Logger.

ConCoord provides a distributed and fault-tolerant threading library. The library includes:

• Lock
• RLock
• Semaphore
• BoundedSemaphore
• Barrier
• Condition

The implementations of distributed synchronization objects follow the implementations in the Python threading library. We will walk through an example below using the Semaphore object under concoord/object/semaphore.py

In the blocking object implementation, the method invocations that use an object from the threading library requires an extra argument _concoord_command. This argument is used by the calling Replica node to relate any blocking/unblocking method invocation to a specific client. This way, even if the client disconnects and reconnects, the ConCoord instance will remain in a safe state:

from concoord.threadingobject.dsemaphore import DSemaphore

class Semaphore:
  def __init__(self, count=1):
    self.semaphore = DSemaphore(count)

  def __repr__(self):
    return repr(self.semaphore)

  def acquire(self, _concoord_command):
    try:
  return self.semaphore.acquire(_concoord_command)
    except Exception as e:
      raise e

  def release(self, _concoord_command):
    try:
      return self.semaphore.release(_concoord_command)
    except Exception as e:
      raise e

  def __str__(self):
    return str(self.semaphore)

To create the proxy for this blocking object we will use the following command:

$ concoord object -o semaphore.Semaphore -p 1

This command creates the proxy that supports blocking operations. Now you can use blocking objects just like basic ConCoord objects. First, we start the replica, acceptor and nameserver nodes the same way we did before as follows:

To test the functionality, you can use multiple clients or print out the Semaphore object as follows:

Visit http://openreplica.org to see concoord in action and to get more information on concoord.

If you believe you have found a bug or have a problem you need assistance with, you can get in touch with us by emailing concoord@systems.cs.cornell.edu