A ConPaaS service may consist of three main entities: the manager, the agent and the frontend. The (primary) manager resides in the first VM that is started by the frontend when the service is created and its role is to manage the service by providing supporting agents, maintaining a stable configuration at any time and by permanently monitoring the service’s performance. An agent resides on each of the other VMs that are started by the manager. The agent is the one that does all the work. Note that a service may contain one manager and multiple agents, or multiple managers that also act as agents.
To implement a new ConPaaS service, you must provide a new manager service, a new agent service and a new frontend service (we assume that each ConPaaS service can be mapped on the three entities architecture). To ease the process of adding a new ConPaaS service, we propose a framework which implements common functionality of the ConPaaS services. So far, the framework provides abstraction for the IaaS layer (adding support for a new cloud provider should not require modifications in any ConPaaS service implementation) and it also provides abstraction for the HTTP communication (we assume that HTTP is the preferred protocol for the communication between the three entities).
You can see below the directory structure of the ConPaaS software. The core folder under src contains the ConPaaS framework. Any service should make use of this code. It contains the manager http server, which instantiates the python manager class that implements the required service; the agent http server that instantiates the python agent class (if the service requires agents); the IaaS abstractions and other useful code.
A new service should be added in a new python module under the ConPaaS/src/services folder.
In the next paragraphs we describe how to add the new ConPaaS service.
The first step in adding a new ConPaaS service is to choose a name for it. This name will be used to construct, in a standardized manner, the file names of the scripts required by this service (see below). Therefore, the names should not contain spaces, nor unaccepted characters.
To function properly, ConPaaS uses a series of configuration files and scripts. Some of them must be modified by the administrator, i.e. the ones concerning the cloud infrastructure, and the others are used, ideally unchanged, by the manager and/or the agent. A newly added service would ideally function with the default scripts. If, however, the default scripts are not satisfactory (for example the new service would need to start something on the VM, like a memcache server) then the developers must supply a new script/config file, that would be used instead of the default one. This new script’s name must be preceded by the service’s chosen name (as described above) and will be selected by the frontend at run time to generate the contextualization file for the manager VM. (If the frontend doesn’t find such a script/config file for a given service, then it will use the default script). Note that some scripts provided for a service do not replace the default ones, instead they will be concatenated to them (see below the agent and manager configuration scripts).
Below we give an explanation of the scripts and configuration files used by a ConPaaS service (there are other configuration files used by the frontend but these are not relevant to the ConPaaS service). Basically there are two scripts that a service uses to boot itself up - the manager contextualization script, which is executed after the manager VM booted, and the agent contextualization script, which is executed after the agent VM booted. These scripts are composed of several parts, some of which are customizable to the needs of the new service.
In the ConPaaS home folder (CONPAAS_HOME) there is the config folder that contains configuration files in the INI format and the scripts folder that contains executable bash scripts. Some of these files are specific to the cloud, other to the manager and the rest to the agent. These files will be concatenated in a single contextualization script, as described below.
Files specific to the Cloud:
(1) CONPAAS_HOME/config/cloud/cloud_name.cfg, where cloud_name refers to the clouds supported by the system (for now OpenNebula and EC2). So there is one such file for each cloud the system supports. These files are filled in by the administrator. They contain information such as the username and password to access the cloud, the OS image to be used with the VMs, etc. These files are used by the frontend and the manager, as both need to ask the cloud to start VMs.
(2) CONPAAS_HOME/scripts/cloud/cloud_name, where cloud_name refers to the clouds supported by the system (for now OpenNebula and EC2). So, as above, there is one such file for each cloud the system supports. These scripts will be included in the contextualization files. For example, for OpenNebula, this file sets up the network.
Files specific to the Manager:
(3) CONPAAS_HOME/scripts/manager/manager-setup, which prepares the environment by copying the ConPaaS source code on the VM, unpacking it, and setting up the PYTHONPATH environment variable.
(4) CONPAAS_HOME/config/manager/service_name-manager.cfg, which contains configuration variables specific to the service manager (in INI format). If the new service needs any other variables (like a path to a file in the source code), it should provide an annex to the default manager config file. This annex must be named service_name-manager.cfg and will be concatenated to default-manager.cfg
(5) CONPAAS_HOME/scripts/manager/service_name-manager-start, which starts the server manager and any other programs the service manager might use.
(6) CONPAAS_HOME/sbin/manager/service_name-cpsmanager (will be started by the service_name-manager-start script), which starts the manager server, which in turn will start the requested manager service.
Scripts (1), (2), (3), (4) and (5) will be used by the frontend to generate the contextualization script for the manager VM. After this scripts executes, a configuration file containing the concatenation of (1) and (4) will be put in ROOT_DIR/config.cfg and then (6) is started with the config.cfg file as a parameter that will be forwarded to the new service.
Examples:
Files specific to the Agent
They are similar to the files described above for the manager, but this time the contextualization file is generated by the manager.
Below you can find the directory structure of the scripts and configuration files described above.
In this section we describe how to implement a new ConPaaS service by providing an example which can be used as a starting point. The new service is called helloworld and will just generate helloworld strings. Thus, the manager will provide a method, called get_helloworld which will ask all the agents to return a ’helloworld’ string (or another string chosen by the manager).
We will start by implementing the agent. We will create a class, called HelloWorldAgent, which implements the required method - get_helloworld, and put it in conpaasservices/helloworld/agent/agent.py (Note: make the directory structure as needed and providing empty __init__.py to make the directory be recognized as a module path). As you can see in Listing [lst:helloworldagent], this class uses some functionality provided in the conpaas.core package. The conpaas.core.expose module provides a python decorator (@expose) that can be used to expose the http methods that the agent server dispatches. By using this decorator, a dictionary containing methods for http requests GET, POST or UPLOAD is filled in behind the scenes. This dictionary is used by the built-in server in the conpaas.core package to dispatch the HTTP requests. The module conpaas.core.http contains some useful methods, like HttpJsonResponse and HttpErrorResponse that are used to respond to the HTTP request dispatched to the corresponding method. In this class we also implemented a method called startup, which only changes the state of the agent. This method could be used, for example, to make some initializations in the agent. We will describe later the use of the other method, check_agent_process.
Let’s assume that the manager wants each agent to generate a different string. The agent should be informed about the string that it has to generate. To do this, we could either implement a method inside the agent, that will receive the required string, or specify this string in the configuration file with which the agent is started. We opted for the second method just to illustrate how a service could make use of the config files and also, maybe some service agents/managers need some information before having been started.
Therefore, we will provide the helloworld-agent.cfg file (see Listing [lst:helloworldcfg]) that will be concatenated to the default-manager.cfg file. It contains a variable ($STRING) which will be replaced by the manager.
Now let’s implement an http client for this new agent server. See Listing [lst:helloworldagentclient]. This client will be used by the manager as a wrapper to easily send requests to the agent. We used some useful methods from conpaas.core.http, to send json objects to the agent server.
Next, we will implement the manager in the same manner: we will write the HelloWorldManager class and place it in the file conpaas/services/helloworld/manager/manager.py. To make use of the IaaS abstractions, we need to instantiate a Controller which controls all the requests to the clouds on which ConPaaS is running. Note the lines:
1: self.controller = Controller( config_parser)
2: self.controller.generate_context('helloworld')
The first line instantiates a Controller. The controller maintains a list of cloud objects generated from the config_parser file. There are several functions provided by the controller which are documented in the doxygen documentation of file controller.py. The most important ones, which are also used in the Hello World service implementation, are: generate_context (which generates a template of the contextualization file); update_context (which takes the contextualization template and replaces the variables with the supplied values); create_nodes (which asks for additional nodes from the specified cloud or the default one) and delete_nodes (which deletes the specified nodes).
Note that the create_nodes function accepts as a parameter a function (in our case check_agent_process) that tests if the agent process started correctly in the agent VM. If an exception is generated during the calls to this function for a given period of time, then the manager assumes that the agent process didn’t start correctly and tries to start the agent process on a different agent VM.
We can also implement a client for the manager server (see Listing [lst:helloworldmanagerclient]). This will allow us to use the command line interface to send requests to the manager, if the frontend integration is not available.
The last step is to register the new service to the conpaas core. One entry must be added to file conpaas/core/services.py, as it is indicated in Listing [lst:helloworldservices]. Because the java and php services use the same code for the agent, there is only one entry in the agent services, called web which is used by both webservices.
So far there is no easy way to add a new frontend service. Each service may require distinct graphical elements. In this section we explain how the Hello World frontend service has been created.
As you have noticed in the Hello World manager implementation, we used some standard states, e.g. INIT, ADAPTING, etc. By calling the get_service_info function, the frontend knows in which state the manager is. Why do we need these standardized stated? As an example, if the manager is in the ADAPTING state, the frontend would know to draw a loading icon on the interface and keep polling the manager.
Several lines of code must be added to the two files above for the new service to be recognized. If you look inside these files, you’ll see that knowing where to add the lines and what lines to add is self-explanatory.